Rockfall Elimination Track B Managing the risk and the value of safety spending

Transcription

1 Safety in Mines Research Advisory Committee Final Project Report Year 3 Rockfall Elimination Track B Managing the risk and the value of safety spending Volume I W.C. Joughin, E. Nezomba, L. Rwodzi, A. Jager Research agency : SRK Consulting Project number: SIM Track B Date: July 2011

2 Executive summary Safety spending in the mining industry is often viewed as a necessary but expensive overhead cost. This mindset places safety and productivity in direct opposition to each other, while improved safety practices could actually lead to better productivity. Alternatively, from a pure safety perspective, the perception also exists that more is better in terms of safety spending, while the additional spending may not necessarily lead to a significant reduction in risk. In order to evaluate the real benefits of safety spending in mitigating rockfalls and their consequences a risk evaluation model has been developed as part of the ongoing SIM Track B research programme. Risk Evaluation Model This model enables the quantification of rockfall injuries and economic losses due to rockfalls for a support system in a given geotechnical environment. Different support systems can be compared and the effectiveness of monitoring and barring can be assessed in terms of the expected frequency of injuries and the total cost of support systems, including the expected economic losses resulting from failures of the systems. Variability is taken into consideration. The existing software, JBlock, was enhanced to perform the rockfall simulation. This tool can now be used to test and design support to cater for the full range of rockfall sizes that can be anticipated in a geotechnical environment. The simulated rockfalls can now be normalised by the area mined. The consequences of these rockfalls are evaluated using the new software RiskEval. The frequency of injuries is simulated by considering the temporal and spatial exposure of personnel. The expected losses associated with dilution, re-supporting, loss of production and loss of sweepings can be evaluated for each rockfall and accumulated. Importantly, rockfalls located in different zones within the stope panel have different consequences and are evaluated accordingly. Validation of the Risk Evaluation Model The model was tested by collecting joint characteristics and mapping rockfalls on two different reefs (UG2 and Merensky) in a structurally complex mine. Data collection is described in detail, which will be useful for anyone attempting this type of analysis. The following conclusions can be drawn from these analyses: 2

3 It is important to apply a clamping stress in JBlock to improve the agreement between rockfalls recorded in JBlock and the actual underground mapped rockfalls. A low clamping stress of 1 kpa to 2 kpa provides the best correlation for the very small rockfalls mapped underground. However, the results highlight the need to develop a method in JBlock to enable an increasing clamping stress with height above the stope to improve the calibration for the full range of rockfalls. A much longer period of comprehensive data collection would be required to provide a more complete distribution of rockfalls underground and for different geotechnical conditions and mining methods. Many of the smaller rockfalls had release surfaces comprising fractures, either blast or stress induced. This phenomenon requires further investigation. Additional rockfall data were obtained from a very large rockfall database, compiled by mine personnel. While this data was useful, it was certainly not complete and therefore the number of rockfalls per area mined could not be reliably compared. This again highlights the need for more comprehensive data collection. The mine s rockfall injury data was considered to be reasonably complete and was used to obtain a more reliable calibration of the model. A clamping stress of 10 kpa was found to provide the best correlation and was used for the case studies. Sensitivity analyses were conducted to show the effect of various input parameters: - clamping stress; - profit margins - quality of support installation - flexibility ( spare stopes) - effectiveness of barring Case Studies A series of case studies were carried out that demonstrate the benefits of using the risk evaluation model. Several different support systems were evaluated and compared in each of three different geotechnical environments. The model clearly shows the expected frequency of injuries and economic losses for each support system, which enables a fair comparison of each support system. 3

4 This work also provided an indication of the amount of work required to perform such analyses. The initial part of the analyses involves the collection of joint data through underground mapping, collecting financial data and investigating the methods of rehabilitating rockfalls used by the mine. This will involve a considerable amount of work as costing for specific task is not always readily available. However, once this has been done, the individual support system analyses are very quick and easy. Recommendations for further research This project has demonstrated that the risk evaluation model and the software developed is immediately useful, but further work is required to develop and improve the tools. The following research and development is recommended: Comprehensive rockfall databases needs to be compiled to determine the percentage of fall out with blasting, efficacy of barring, proportion of large rockfalls and their consequences. This will enable a more rigorous calibration of the risk model. The rockfall model needs to be improved to take varying clamping stress into consideration. Models need to be developed for other rockfall modes of failure: - Rockbursts and seismically induced rockfalls, - Pillar failure, - Unravelling, - Stress damage - Time dependant deterioration The model needs to be adapted to cover a wider range of mining methods. The reduction in risk due to monitoring needs to be researched further. This forms an important link with the results of the Track A project and collaboration between the two projects is required. The software requires ongoing development to ensure that it meets the needs of the rock mechanics practitioners who will use the software. 4

5 Table of contents 1 INTRODUCTION 12 2 REVIEW OF WORK DONE DURING YEAR 1 AND YEAR RISK EVALUATION MODEL Risk Evaluation process Quantifying the potential for rockfalls Keyblock generation Rockfall simulation Quantifying rockfall injuries Rockfall occurs during blast or is effectively barred Time exposure of personnel (personnel present) Monitoring and evacuation Spatial coincidence Calculating the expected number of injuries Severity of injuries and fatalities Evaluating accepted levels of injuries and fatalities Quantifying economic losses and loss of reserves Remediation strategies and consequences Cost of rockfall damage Costs of injuries and fatalities Economic evaluation Once off implementation costs and reserve loss Summary 54 4 VALIDATION OF THE RISK EVALUATION MODEL Joint Characteristics and Methods of Joint Data collection Joint Properties Joint Mapping Techniques Joint Mapping Errors

6 4.2 Field Mapping Sites Site Selection Mapping Programme Site Description Support Systems JBlock Input Data Geotechnical Data for JBlock Analysis Data Collection Data Analysis JBlock Geotechnical Input Data Underground Rockfall Mapping Data Collection Results Rockfall and Injury Data provided by Mine A Rockfall Data Injury Data Calibration and validation of JBlock simulated rockfall data Calibration based on underground rockfall mapping Calibration based on Mine A rockfall database and rockfall injury data Sensitivity Analysis RISK EVALUATION CASE STUDIES Case Study1 (UG2 Reef at Mine A) Scenario Descriptions Financial and Injury Data JBlock Results Risk Evaluation and Comparison of Results Case Study 2 (Merensky Reef at Mine A) Scenario Descriptions Financial and Injury Data JBlock Results Risk Evaluation and comparison results

7 5.3 Case Study 3 (Stope cable anchors at Mine B) Site Description Geotechnical Data Scenario Descriptions Financial and Injury Data JBlock Results Risk Evaluation and comparison of results Discussion CONCLUSIONS RECOMMENDATIONS FOR FURTHER RESEARCH ACKNOWLEDGEMENTS REFERENCES 132 7

8 List of Figures Figure 3-1: Uncertainty in slope design (after Tapia et. al 2007) 17 Figure 3-2: Risk Evaluation Process for rockfalls, modified from Stacey et. al 2006 and Terbrugge et. al Figure 3-3: Keyblock generation 21 Figure 3-4: Set of keyblocks generated by JBlock (frequency distribution of block volume) 22 Figure 3-5: Rockfall simulation 24 Figure 3-6: Results of rockfall simulation (Cumulative frequency distribution of rockfall width, length area and volume) 25 Figure 3-7: Results of rockfall simulation (Cumulative frequency distribution of the areas of rockfalls within each zone) 26 Figure 3-8: Event tree for injuries and fatalities. 29 Figure 3-9: Spatial coincidence 32 Figure 3-10: F-N Graph for bench marking safety impact (Contreras et al, 2006) 39 Figure 3-11: ALARP Principle (from Major hazard facilities regulations guidance notes) 40 Figure 3-12: International benchmarking using the F-N Graph (after Gonzalez and Karzulovic, 2004) 41 Figure 3-13: Platinum Industry Fatality Rate Towards the MHSC s Milestones. 42 Figure 3-14: Platinum Industry DIFR 43 Figure 3-15: Rockfalls in stope zones 45 Figure 3-16: Cleaning and re-supporting 46 Figure 3-17: Pillar re-establishing 47 Figure 3-18: Full panel re-establishing 47 Figure 3-19: Cost of small and medium rockfalls. 50 Figure 3-20: Cost of large rockfalls in Zone 1 51 Figure 3-21: Cost of large rockfalls in Zone 2 51 Figure 3-22: Cost of large rockfalls in Zone 3 52 Figure 3-23: Economic evaluation 54 Figure 4-1: Definition of dip (β) and dip direction (α) 57 Figure 4-2: Joint spacing error correction 59 Figure 4-3: Patton s experiment on the shear strength of saw-tooth joints (Patton 1966). 61 Figure 4-4: Area mapping (After Zhang and Einstein (1998).) 64 Figure 4-5: Photogrammetric mapping (From Sturzenegger and Stead, 2009) 65 8

9 Figure 4-6: Straight scanline mapping (Zhang, 2004). 67 Figure 4-7: Mapped stopes in the UG2 Reef 71 Figure 4-8: Mapped stopes in the Merensky Reef 72 Figure 4-9: UG2 support standard 73 Figure 4-10: Merensky support standard 74 Figure 4-11: Breithaupt compass 76 Figure 4-12: Friction angle distribution curve for Merensky J1 set 78 Figure 4-13: Dips pole plot of mapped joints in UG2 stopes. 79 Figure 4-14: Dips pole plot of mapped joints in Merensky stopes 80 Figure 4-15: Pole plot of mapped rockfall release surface in UG2 stope 84 Figure 4-16: Pole plot of mapped rockfall release surface in Merensky stope 84 Figure 4-17: Pole plot of mapped rockfall fracture release surfaces in UG2 stope 86 Figure 4-18: Pole plot of mapped rockfall fracture release surfaces in Merensky stope 86 Figure 4-19: Frequency distribution of rockfall volume 87 Figure 4-20: Frequency distribution of rockfall area 88 Figure 4-21: Frequency distribution of rockfall height 88 Figure 4-22: Rockfall location contour plot for UG2 Reef 89 Figure 4-23: Rockfall location contour plot for Merensky Reef 89 Figure 4-24: Cumulative rockfall frequency vs distance from face 91 Figure 4-25: Percentage of rockfalls per size category vs distance categories behind the face (mine rockfall data only) 92 Figure 4-26: UG2 Reef small rockfall keyblock distribution 95 Figure 4-27: Merensky Reef small rockfall distribution 95 Figure 4-28: UG2 Reef mapped rockfall volume and JBlock distributions for clamping stresses from 0 kpa to 20 kpa 97 Figure 4-29: UG2 Reef mapped rockfall area and JBlock distributions for clamping stresses from 0 kpa to 20 kpa. 97 Figure 4-30: Merensky Reef mapped rockfall volume and JBlock distributions for clamping stresses from 0 kpa to 20 kpa. 98 Figure 4-31: Merensky Reef mapped rockfall area and JBlock distributions for clamping stresses from 0 kpa to 20 kpa. 98 Figure 4-32: Keyblock size distribution for UG2 Reef at mine A 103 Figure 4-33: Keyblock size distribution for Merensky Reef at mine A 103 Figure 5-1: Dips pole plot for Mine B Merensky joint sets 119 9

10 Figure 5-2: 1.5 m Cable Anchor support system 121 Figure 5-3: 3 m Cable Anchor support system 122 Figure 5-4: Keyblock size distribution for Merensky Reef at mine B 123 List of tables Table 2-1. Summary of the direct and indirect consequences of rockfalls 13 Table 3-1: Exposure reduction due to rockfalls which occur during the blast or are barred out. 30 Table 3-2: Time exposure analysis (hours per day) 30 Table 3-3: Time exposure analysis (proportion of time) 31 Table 3-4: Expected injuries for different categories of workers. 35 Table 3-5: Severity of rockfall injuries and fatalities 36 Table 3-6: Disabling Injury and Fatal injury Rates 36 Table 3-7: Comparison between fatal accident rates 43 Table 3-8: Direct cost of injuries 52 Table 4-1: Joint Property and Source 57 Table 4-2: Example of a scanline logging sheet 67 Table 4-3: Mapping area covered 70 Table 4-4: JBlock joint input properties for UG2 Reef 80 Table 4-5: JBlock joint input properties for Merensky Reef 81 Table 4-6: Rockfall mapping data 83 Table 4-7: Joints making blocks and significant joint sets 85 Table 4-8: Rockfall related injury and fatality ratios 93 Table 4-9: Measures of accuracy for Area 101 Table 4-10: Measures of accuracy for Volume 101 Table 4-11: Rockfalls and Injuries for UG2 Reef at mine A with varying clamping stress 104 Table 4-12: Rockfalls and Injuries for Merensky Reef at mine A with varying clamping stress 104 Table 4-13: Clamping Stress Sensitivity 105 Table 4-14: Profit Margin Sensitivity 106 Table 4-15: Effect of quality of support installation on rockfall frequency 107 Table 4-16: Effect of quality of support installation on costs 107 Table 4-17: Costs associated with varying mining flexibility 108 Table 4-18: Costs and injuries associated with the effectiveness of barring

11 Table 5-1: Summary of rockfalls for case study Table 5-2: Summary of Risk Eval results for case study Table 5-3: Summary of rockfalls for case study Table 5-4: Summary of Risk Eval results for case study Table 5-5: Table of scanlines and number of joints mapped 117 Table 5-6: JBlock joint input parameters for mine B 119 Table 5-7: Case study 3 support scenario and strengths 120 Table 5-8: JBlock results for case study Table 5-9: RiskEval case study 3 results

12 1 Introduction Safety spending in the South African mining industry is often viewed as a necessary but expensive overhead cost. This mindset places safety and productivity in direct opposition to each other. In lean times, safety could be compromised for the perceived advantage of increased productivity. For certain circumstances, the perception also exists that more is better in terms of safety spending, while this is not necessarily the case. Very often the benefits of safety spending are expressed only in terms of reduced injury and fatality rates. The direct financial benefits of reducing rockfall rates are seldom quantified, and the indirect and downstream benefits are rarely acknowledged, let alone quantified. In order to evaluate the real benefits of safety spending in mitigating rockfalls and their consequences a risk model has been developed as Track B of the SIMRAC SIM Rockfall Elimination research programme. The aim of this research is to address this problem by demonstrating the financial benefits of safety spending, and to provide tools for quantifying this benefit under various conditions. The research has been conducted over a period of three years. However, at the end of year 2, a change in scope was requested in order to provide some calibration of the model through underground mapping of joints and rockfalls. A brief review of the work done in year 1 and 2 and the subsequent changes is presented in chapter 2. The risk evaluation model is described in chapter 3. This model incorporates a special upgrade of the software JBlock and the newly developed program RiskEval. The underground mapping and calibration of JBlock and RiskEval is presented in chapter 4. Case studies which demonstrate the use of the software are presented in chapter 5. Conclusions and recommendations for further research are provided at the end of this report. 12

13 2 Review of work done during Year 1 and Year 2 The work carried out during year 1 and 2 has been summarised in the two annual interim reports. This work has subsequently been written up more comprehensively in an MSc dissertation (Rwodzi, 2010). The following work was described: A review of risk in the industry Consequences of rockfalls Development of a preliminary model to evaluate the consequences Application of the model through case studies At the end of year 2, the project team realised that it proved difficult to calibrate the results, since geotechnical mapping data and rockfall data were scarce. A change in scope was therefore presented to the SIMRAC committee and accepted. Some of the earlier information has been used and developed further, while other information has been discarded since it is now redundant. Important aspects of the review of risk have been incorporated in the final risk evaluation model (section 3). A summary of direct and indirect consequences of rockfalls was determined during a workshop, which included experienced mining and rock engineering practitioners, at the start of the project and is presented in Table 2-1. The factors to consider when evaluating the consequences are described. The model has been developed to evaluate the consequences of stope excavations in great detail. This was done because the greatest exposure of personnel is in stopes and this is where most injuries occur. Also damage occurs more frequently in stopes. In the case of injuries and fatalities there are significant costs incurred in addition to the moral and societal risk. These costs were evaluated in some detail using a comprehensive thesis on the costs of rockfall injuries (Marx, 1996) and medical costs obtained from a mine hospital. A summary of the costs that were used is presented in section Table 2-1. Summary of the direct and indirect consequences of rockfalls 13

14 Direct Consequences Indirect consequences Loss of reserves (Net present value of lost reserves less insurance claim) - major damage or loss of access to stopes. Loss of production (revenue from lost production less variable costs) if there is no alternate source of production. Damage to excavations (access excavations and stopes) Replacement of access excavations (cost divided by life in years) large rockfalls Rehabilitation - proportional to size of rockfall and importance of excavation Dilution (in stopes) Re-deployment of machinery and personnel to maintain production. Clean up operations. (depends on size of rockfall) Insurance premiums (Increase due to claims) Stakeholder resistance (reputation, share price and cost of capital)-difficult to quantify Moral and societal risk Temporary mine closure (Mine Health and Safety Act, Section 54) (revenue from lost production less variable costs) up to 5 days of partial or full mine closure. Retraining of existing personnel, audits and loss of reserves Medical and rescue operation costs Wages and compensation Injuries and fatalities Investigation and inquiries cost of professional time Re-training cost of re-training new employees SIMRAC levies (depends on severity of injury) Legal costs (determined from precedent practice) Insurance premiums (Increase due to accident record) Industrial action difficult to quantify Stakeholder resistance (reputation, share price and cost of capital)-difficult to quantify Damage to equipment and machinery (mobile and fixed) Loss of production (revenue from lost production less variable costs) if there is no alternate source of production. - only production affected by equipment loss Cost of re-deployment of machinery and personnel to maintain production Replacement costs (based on depreciated value of damaged machinery) - large rockfalls Cost of repairs depends on extent of damage (size of rockfall) 14

15 In the final model, the consequences now also depend on the location of the rockfall within the stope. This is an important change and enables a more realistic evaluation of the consequences. In addition to this, different categories of workers have different exposure levels in different locations within the stope. This necessitated a change in the model for calculation of injuries. It was no longer possible to calculate the frequency of incidents of multiple injuries, since the grouping of personnel is no longer simple. The expected frequency of injuries is now more reliably calculated. A model was developed in some detail to evaluate the cost of damage to machinery. However, the lack of data on equipment damage and associated costs made it difficult to pursue this further. The case studies performed in year 2 were based on the original model and joint data was assumed. While these analyses provided important insight at the time, they are not presented in this report, since the model has changed considerably. New case studies with mapped joint and rockfall data have been included in section 5. 15

16 3 Risk Evaluation Model Uncertainty and risk are central features of rock engineering. Rock strengths have very high variability and the rock mass is made up of multiple discontinuities with varying orientations and shear strengths. These characteristics can also change and deteriorate with time. Mining is not perfect. The dimensions of excavations and pillars vary considerably due to both local geotechnical conditions and lack of compliance or organized mining. This can significantly affect the stability of excavations and pillars. The strength of support units is inherently variable, particularly if timber is used. Support standards are not always followed and the support system may not perform as intended. This leads to a number of questions. Which is the best safety strategy or support system to address the risk? How do you decide when a design is acceptable with so much uncertainty? How is uncertainty incorporated in a design? Christian (2003) suggests that engineers deal with uncertainty in four different ways: 1. Ignoring it. While on its face, such a head-in-the-sand approach would seem insupportable, it is surprisingly widespread. 2. Being conservative (eg worst case scenario). This is an obvious and frequently sound approach. Rather than get involved in the details of how often undesirable things might happen and what their consequences might be, the engineer makes the structure or system so robust that it will resist anything. While this works in many cases, it is usually expensive, it may drag the project out to unacceptable completion times, and in some cases it may simply not be possible. Eventually one must ask how conservative is conservative enough. 3. Using the observational method. The observational method has established itself as the preferred way for geotechnical engineers to deal with uncertainty in situations for which simple conservatism is unsatisfactory. It involves (1) considering possible modes of unsatisfactory performance or other undesirable developments; (2) developing plans for dealing with each such development; (3) making field measurements during construction and operation to establish whether the developments are occurring; and (4) reacting to the observed behaviour by changing the design or construction process. 4. Quantifying uncertainty. This is the purpose of reliability approaches. Quantifying the uncertainty is consistent with the philosophy of the observational method; it might be 16

17 considered a logical extension of the observational method that accommodates modern developments in probabilistic methods. Figure 3-1 demonstrates the use of probabilistic methods. A distribution of factors of safety (FOS) can be determined using a Monte Carlo simulation (or other similar methods) taking into account the variability in the geotechnical input parameters. The area under the curve to the left of a factor of safety of 1.0 indicates the probability of failure. This approach has also been applied to pillar stability (Joughin et. al 2000 and Esterhuizen, 1993). Figure 3-1 also shows that the simple deterministic methods are not always satisfactory. A more conservative design does not necessarily have a lower inherent risk, than a less conservative design, if the uncertainty is greater. The factor of safety, based on the mean input parameters is higher for case B, but the probability of failure is greater. Figure 3-1: Uncertainty in slope design (after Tapia et. al 2007) One difficulty in interpreting results is that most people, including engineers, have difficulty establishing an allowable probability of failure or dealing with low values of probability. A risk evaluation process has been developed to address this. 17

18 3.1 Risk Evaluation process A risk evaluation process has been developed for addressing geotechnical risk in surface mining (Terbrugge et al, 2006) and a slightly modified version for underground mining was presented by Stacey. et. al. (2006). In this process, the probability of slope/stope failure is determined using a fault tree, then the risks are determined using an event tree and these risks are then evaluated against pre-defined accepted levels of risk. The basic concept has been maintained in the risk evaluation process presented in Figure 3-2, but it has been adapted to evaluate the risk associated with multiple rockfalls of varying size rather than one major rockfall. This is done because both large and small rockfalls have consequences. Large rockfalls have more severe consequences, but the small rockfalls occur more frequently and their cumulative effect can be as significant. Statistical mode of failure analysis to determine the potential for rockfalls Event tree to determine the risks Evaluation of risk levels Keyblock Injury to personnel Expected injuries Evaluate against accepted injury risk level Stope Collapses Pillar failures Rockburst Set of rockfalls of varying size per area mined Damage to equipment Loss of production Excavation damage Expected economic loss Loss of reserves Evaluate loss of revenue against cost of improved risk control Evaluate effect on NPV Seismically induced rockfall Human Resources Industrial Action Accepted level of risk based on above? Other modes of failure Public Relations Stakeholder resistance Accepted level of risk based on above? Figure 3-2: Risk Evaluation Process for rockfalls, modified from Stacey et. al 2006 and Terbrugge et. al

19 Statistical mode of failure analysis to determine the potential for rockfalls The first step is to develop a model which will provide a large set of rockfalls of different sizes, which represents a given area mined or hangingwall exposed (column 1, Figure 3-2). It is important to consider different rockfall sizes, because the consequences depend on the size of the rockfall. Normalising the set of rockfalls to an area mined enables reasonable comparisons between different support systems and to actual data. In order to develop the model, potential modes of failure that could result in rockfalls need to be identified. In order to carry out a comprehensive analysis, one should address all relevant modes of failure. The list of failure modes in Figure 3-2 is by no means complete, but these are general failure modes. For comparison purposes, it is possible to analyse a single mode of failure and compare the performance of two mitigating strategies (eg support or mining layout). An analytical model, which can represent the failure mode and determine the volume of failure for variable geotechnical input parameters should be used. Then a monte carlo simulation should be carried out to provide a frequency distribution of the rockfalls. If multiple modes of failure are analysed, then the sets of rockfalls can be combined, providing they represent the same area mined or hangingwall exposed. In this project, only the keyblock mode of failure analysis was developed, because there was software available, JBlock (Esterhuizen, 2003), which could perform this type of analysis. This software has been developed further to provide the necessary input for the remainder of the risk evaluation process. Most rockfalls in Platinum and Chrome mines are due to jointed bounded keyblocks and therefore this project will be valuable to the industry. The analyses are done on a comparative basis. Different support systems, mining layouts or other safety strategies can be compared. Event tree to determine the risks The next step is to develop an event tree, which is used to determine the consequences and determine the risk (column 2, Figure 3-2). The event tree in Figure 3-2 is very simple and just demonstrates possible outcomes. The outcomes with solid outlines can be quantified quite reliably using the simple models outlined in sections 3.3 (injuries) and 3.4 (economic loss and loss of reserves). The consequences of industrial action and stakeholder resistance are not easily defined, but these are probably related to the frequency of injuries and fatalities. 19

20 RiskEval was developed to read the output from JBlock and calculate the frequency of injuries and costs of rockfalls. Accepted risk levels The final step is to evaluate the level of risk against accepted risk levels. It is important to note that the risk of injuries and fatalities is evaluated separately from economic considerations. Defining accepted risk levels for injuries and fatalities is controversial and is discussed in more detail in section The evaluation of economic risk and loss of reserves is discussed in sections and Quantifying the potential for rockfalls The program, JBlock, is used to quantify the potential for rockfalls. The program generates a set of keyblocks of varying size, which is normalised by the area mined. A rockfall simulation is then carried out using the proposed support Keyblock generation The first step in the process is to generate the set of keyblocks or potential rockfalls. The following information is required for this process: Stope dip and dip direction Dip and dip direction of each joint set (mean and standard deviation) Spacing and trace lengths of joints within each joint set (mean, minimum and maximum) Friction angles for each joint set (mean and standard deviation) (Barton 2002) Parting planes parallel to the stope hangingwall (mean and standard deviation of height above stope hangingwall) The joint information needs to be collected by mapping joints in the underground stopes. The data should represent a ground control district. Fault sets can also be included. Parallel parting planes are prominent bedding planes or weak horizons above the reef (eg chrome stringers and Green Bar). These can be identified in geological boreholes. The orientated hangingwall surface of an excavation is simulated with the traces of joints (Figure 3-3). The joints are projected in 3D and JBlock then determines whether blocks are formed. Primary blocks are discrete blocks which are bounded by joints and do not include joints within the block. Secondary blocks (combined blocks) are bounded by joints and also 20

21 include joints within the block and may include primary blocks. JBlock records all relevant information pertaining to each block: shape, geometry, area, volume, length (parallel to the stope face), height and shear strength parameters (cohesion and friction) for each surface bounding the block. JBlock also keeps track of the hangingwall surface area of unformed blocks. The total area is the sum of the hangingwall surface area of all blocks and represents the hangingwall surface area exposed during the simulation and can be equated to the area mined. Primary block Secondary block 3D Blocks formed Unformed block Figure 3-3: Keyblock generation Total area Represents area mined JBlock also allows additional geological features to form blocks. These include: Ramp structures (size and number per square metre). These features frequently interact with other joints. Dome structures (size and number per square metre). These can represent pillow lavas or through cross bedding and do not need to interact with other joints. The block data is stored in a file, which can then be used for the rockfall simulation (section 3.2.2). An example of a set of keyblocks is included in Figure

22 Figure 3-4: Set of keyblocks generated by JBlock (frequency distribution of block volume) When generating a set of keyblocks it is necessary to specify the minimum block size and the minimum number of keyblocks or minimum hangingwall area exposed. Figure 3-4 shows many times more small blocks are generated than large blocks. Small rocks can cause injuries and therefore must be included in the data set (section 3.3). The largest rockfalls result in the major production losses (3.4.2) and it is therefore essential to ensure that sufficient large keyblocks are included in the data set to provide a realistic estimation of the damage costs (section 3.4). Very large block sets are therefore required to provide a realistic model. The more jointed the ground control district the more blocks are required. Some experimentation is required to determine an ideal block set, but the following parameters are suggested for the preliminary analyses: Minimum block size: 0.01 m 3 Minimum number of keyblocks: 100,000 Minimum hangingwall surface area: 50,000 m Rockfall simulation The stope excavation and stope support are then simulated as shown in Figure 3-5. The following types of support can be simulated: 22

23 Point support (timber elongates, props) Line support (headboards) Tendon support (end anchored or grouted) Area support (packs) Membrane support (shotcrete, mesh, thin sprayed liner and nets) The mean and standard deviation of the support spacing can be specified. The strength of the support unit also has a mean and standard deviation, which is used to represent the inherent variability of the support (eg timber elongates) and/or installation practice (not perpendicular to the hangingwall). JBlock also allows different support to be installed for different shifts. Normally, during drilling before the blast, temporary support and an additional line of support are installed, while during cleaning after the blast, the support is further from the face. The stope can then be divided into zones of interest. This is important since personnel exposure (section 3.3.2) and the remediation strategies (section 3.4) will differ in each of these zones. For the purposes of this study, the zones are defined as the face, sweepings, gully and back area. JBlock places the entire set of blocks in the stope area, one by one, and tests the stability of each block. Some of the blocks can then be tested with the additional support installed for the drilling shift and others will be tested without this support installed. The user must specify the proportion of blocks to be tested on the different shifts. In reality more rockfalls occur close to the face and therefore JBlock places more blocks close to the face. This is controlled by an input cumulative frequency distribution of rockfall distance from the face, which is best determined from actual rockfall data. Blocks are placed randomly in the direction parallel to the face. The program first checks the location of the block relative to support and checks that tendons are protruding beyond the block and anchored. It then tests for support failure, failure in between support or rotational failure. The bond length and bond strength of grouted tendons is also analysed. A horizontal clamping stress can be applied during the analysis. The resulting failed blocks make up the set of rockfalls. 23

24 100% 90% 80% 70% Frequency 60% 50% 40% Cumulative frequency distribution of rockfall distance from the face) Frequency 30% 20% 10% 0% Distance From Face Zone 3 Block must fail support Stope face Zone 1 Zone 2 Zone4 Block failure in between support Block rotation Figure 3-5: Rockfall simulation The simulation is carried out with many thousands of blocks and a set of rockfalls is created for further analysis. During the simulation, JBlock keeps track of the total hangingwall area exposed (area mined) for the set of blocks and this can be used to normalise the data. Figure 3-6 is an example of the output. Figure 3-7 shows a cumulative frequency distribution of the areas of the portions of rockfalls within each zone. 24

25 Figure 3-6: Results of rockfall simulation (Cumulative frequency distribution of rockfall width, length area and volume) 25

26 Cumulative Frequency Legend Face Area Sweepings Area Gully Area Back Area Area (m 2 ) Figure 3-7: Results of rockfall simulation (Cumulative frequency distribution of the areas of rockfalls within each zone) The rockfall set that is generated can be normalised by the area mined per panel, per annum as follows: where: A m = blasts per month x months (12) x advance per blast x panel length (expected area mined per panel, per annum, based on the production target or call). A h = hangingwall area exposed (simulation area that the block set represents) This is used to normalise the number of injuries (section 3.3.5) and the costs of damage (section 3.4.2). It is possible to test each block in the set several times, while keeping track of the hangingwall area exposed, to ensure that there are sufficient large rockfalls for the analysis, since many of the large blocks will be supported. This is often necessary to estimate the damage costs realistically (section 3.4). However the block set must have an appropriate distribution of rockfall size and include sufficient large blocks (section 3.2.1). 26

27 JBlock provides a report in a csv (comma separated values) file, which details the following information for every block tested: Width-P (m): block length or width parallel to the stope face Height (m): height of block. Volume (m 3 ): volume of block FaceArea (m 2 ): area of the block face in the hangingwall Parting? (Y/N): top surface of the block is formed by a parting plane Sim no: if blocks are tested more than once, then the simulation number is reported Fall? (Y/N): did the block fail? Mode (1-7): mode of failure/stability: failed support, failed between support, failed by rotation, stable by friction, stable by support, stable over solid, stable no sliding No. Supts: number of support units of all types installed in the block No. Bolts: number of bolts installed in the block No. tooshort: number of bolts installed in the block that were too short No. longenough: number of bolts installed in the block that were long enough No. Bondfailure: number of bolts installed in the block that failed due to bond failure No. Steelfailure: number of bolts installed in the block that failed due to steel failure FaceDist (m): distance from the face to the nearest point of the block Blast (A/B): was the block tested before or after installation of temporary support Fric 1-6: friction angle for each block surface. JBlock also reports the following summary information: Total hangingwall area exposed (represents the equivalent area for the number of blocks tested) Number of blocks tested Total failed area Total failed volume Total number of bolts that were long enough Total number of bolts that were too short Total number of bolts that failed in tension Total number of bolts that failed due to bond failure Area of each zone (face, sweepings, gully and back area) 27

28 This information is read by the program RiskEval and used for the estimation of rockfall injuries (section 3.3) and cost of damage (section 3.4). The information was also used for forensic testing and debugging of the JBlock output. 3.3 Quantifying rockfall injuries The quantification of rockfall injuries is carried out using the program RiskEval. The model makes use of the event tree methodology to analyse the probability of a rockfall resulting in an injury. Event trees start with an initiating event, such as a rockfall (Figure 3-8). This is followed by the formulation of a set of subsequent possible events (circumstances, conditions or mitigating measures), which could follow as a result of the initiating event. Paths are developed and the probability of occurrence of each event is calculated or estimated. The probabilities of the final outcome at the end of each path are the product of the probability of occurrence of the initiating event and the probabilities of each event along the path. Multiplying the probabilities reduces the probability of the final outcome. Figure 3-8 illustrates the event tree developed for injuries and fatalities. The event tree analysis is carried out separately for each rockfall generated by the statistical keyblock analysis (section 3.2). The red path indicates the route towards injuries, where the outcome is unfavourable at each event. Alternate paths do not result in injuries. The probabilities can vary depending on the worker category exposure and the rockfall size, location and mode of failure (those shown in Figure 3-8 are for a single rockfall example). Explanations of events are provided in sections to The calculation of the number of injuries for all workers and all rockfalls is explained in section The severity of injuries is covered in section and the evaluation of accepted risk is discussed in section

29 Rockfallsof varying sizes (JBlock output) Rockfall occurs during blast? No (35%) Effectively Barred? No (10%) Yes (65%) Yes (90%) No (95%) Personnel Present? Monitoring and evacuation? Yes (5%) Yes (0%) No (100%) No (99.8%) Spatial Coincidence? Yes (0.2%) No injury Injuries: Severity: Dressing LTI Reportable Fatality Figure 3-8: Event tree for injuries and fatalities Rockfall occurs during blast or is effectively barred A large proportion of the rockfalls occur during or shortly after the blast, which will not cause injuries, since personnel will not be present. During early examinations, unstable blocks are identified and barred down, rendering them harmless. The success of barring has been measured as the proportion of rocks that are barred versus the total number of rocks that fall and are barred. Different categories of workers may be more or less exposed to unbarred rocks. Barring is usually poor on nightshift and therefore these workers are exposed to unbarred rocks. Team leaders could also have a higher exposure level since they are actually carrying out the barring. Data was captured during underground monitoring (section ) to determine the proportion of rocks that fall out with the blast or are barred out, but unfortunately the data is quite limited and likely to vary for different areas. While this data provides useful insight, some judgement is still required to determine appropriate probabilities. Table 3-1 provides example values for exposure reduction due to blasting and barring. This information must be specified in RiskEval. Loose rocks are barred out before support is installed or in between installed support. In practice, it is rare that supported rocks are barred out. The model therefore applies this factor only to unsupported rocks. 29

30 Table 3-1: Exposure reduction due to rockfalls which occur during the blast or are barred out. Category Shift Rockfall occurs during blast Rockfall is barred out Stope Driller Day Gully Driller Day Stope Timber Day Winch Driver Day Team leader Day Winch Driver Night Team leader Night Time exposure of personnel (personnel present) Rockfalls can occur at any time of day, but personnel are only exposed for a limited time during each shift. The exposure time will be different for each category of worker and will differ in each stope zone (see section 3.2.2). An exposure analysis should be conducted, such as that shown in Table 3-2 and captured in RiskEval. The following nomenclature applies to the table: H jk = the hours per shift in zone j for worker category k N k = the number of persons per worker category k Table 3-2: Time exposure analysis (hours per day) Category Shift N k H 1k H 2k H 3k H 4k Stope Driller Day Gully Driller Day Stope Timber Day Winch Driver Day Team leader Day Winch Driver Night Team leader Night

31 The proportion of time exposed per worker category, per zone can be calculated as follows: Where: S = working shifts per month Table 3-3 shows the resulting time exposure (E jk) for each type of worker in each zone. Table 3-3: Time exposure analysis (proportion of time) Category Shift Nk E1k E2k E3k E4k Stope Driller Day Gully Driller Day Stope Timber Day Winch Driver Day Team leader Day Winch Driver Night Team leader Night This analysis is based on the assumption that the compliment of workers will always be filled and when someone goes on leave they are replaced. When determining the exposure of individual person, the proportion of time exposed should be reduced to take into consideration the time spent on leave or absent as follows: Individual exposure in a given zone is therefore E jk L. The value of L will typically be around 90% to 95% and this is therefore a relatively minor adjustment Monitoring and evacuation Monitoring systems are being developed, which could provide a warning prior to a rockfall. Injuries could be avoided if a warning is given and personnel are successfully evacuated prior to the rockfall. Such systems are not widely used, but the results of experimental monitoring systems are encouraging. As more data is obtained from these experiments, the success rates of identifying rockfalls could be determined. The ability to evacuate personnel will depend on 31

32 the amount of time given and the practical constraints during evacuation. The probability of success is the product of the probabilities of successful monitoring and successful evacuation. Effective systems could dramatically reduce the risk of rockfalls. Timber elongates usually provide a warning prior to large rockfalls. Invariably the failure is progressive and not immediate. The failure of the timber elongates is noisy and alerts personnel that a large failure is about to take place Spatial coincidence If personnel are exposed, they also need to occupy the same space as the rockfall if they are to be injured. Rockfalls can occur entirely in one zone or they can overlap zones. Personnel spend different proportions of their time in the different zones (section 3.3.2). Each rockfall in the set could potentially injure somebody. Figure 3-9 illustrates this concept. Zone 3 Zone 1 Zone 2 Zone4 Figure 3-9: Spatial coincidence The probability of spatial coincidence (C ij ) for a single rockfall within a zone can be simply estimated as follows: C ij = R ij/z j where R ij is the area of the portion of rockfall i within zone j, and Z j is the area of the zone j. 32

33 A correction needs to be made when the rockfalls are very small, because a very small rockfall can strike a person anywhere on their body: C ij = P/Z j where P is the area that a person occupies (about 0.3 m 2 ), and R i < P It follows that small rockfalls are less likely to injure people than large rockfalls. However there are many times more small rockfalls than there are large rockfalls (see Figure 3-6in section 3.2.2). The total spatial coincidence, in a given zone, for all rockfalls can be determined as follows: This represents the number of times that any person could be struck by a rockfall in a given zone if they were exposed all the time. It is quite likely that C j will be greater than 1.0 when a large number of rockfalls are simulated. However, injuries can only occur while personnel are exposed, so the expected number of injuries per person will be a small fraction of C j. The method of calculation is discussed in the following section (3.3.5) Calculating the expected number of injuries Expected injuries per person in a given category for all rockfalls The expected number of injuries per person in a given category for all rockfalls, taking into consideration both temporal exposure and spatial coincidence can be calculated as follows: where: Bl k =probability of the rockfall falling during the blast per worker category k (section 3.3.1) Ba k = probability of the rockfall being barred per worker category k (section 3.3.1) M = the reduction in exposure due to monitoring and evacuation (section E jk = proportion of time spent in zone j per worker category k (section 3.3.2) C j = The total spatial coincidence in zone j for all rockfalls (section 3.3.4) 33

34 Risk of injury for an individual in a given work category The risk of injury for an individual person in a given work category can be assessed by determining the expected annual frequency of injuries as follows: Expected individual injuries per annum = I k L B where: L is the adjustment for leave days (section 3.3.2) B is the annual normalisation parameter (section 3.2.2). This can be used to evaluate the individual risk against benchmarks or international standards (section 3.3.7). Total expected injuries The total number of injuries for all categories of workers and all rockfalls is calculated as follows: where: N k is the number of workers per category k (section 3.3.2) It is useful to normalise the expected injuries by the hangingwall area exposed or area mined (A h in section 3.2.2) as follows: Expected injuries per square metre mined = I/A h This value can be used to calibrate the expected injuries with the actual rockfall injuries being experienced by the mine. The expected injuries can also be normalised per panel per annum as follows: Expected injuries per panel per annum = I B Injury Frequency Rate The Injury Frequency Rate (IFR) can be calculated using the expected injuries per annum. This rate can be determined per 1000 workers per annum or per million man hours as follows: 34

35 where: H is the total hours per shift per person (including travelling time) S is the number of shifts per month (section and 3.3.2) The IFR can be converted to a disabling injury frequency rate (DIFR) or fatal injury frequency rate (FIFR) by taking the injury severity (section 3.3.6) into consideration. Modelled DIFR and FIFR can be compared directly with the actual mine data and industry benchmarks (section 3.3.7). It should be noted that the actual mine data includes the hours spent by other workers and should therefore be lower than the modelled values. Example injury calculation Table 3-4 is an example of the calculated expected injuries. Table 3-4: Expected injuries for different categories of workers. Injuries Category Crew size I Individual injury risk per annum Injuries per annum per panel Injuries per 100,000 m 2 Stope Driller Gully Driller Stope Timber Winch Driver Team leader Winch Driver (N/S) Team leader (N/S) Total/Average IFR (per 1000 workers per annum) 29 IFR (per million man hours) Severity of injuries and fatalities The severity of injuries cannot easily be determined using this model. The size of the rockfall obviously has a significant effect, but this also depends on where the rockfall strikes the person. It is therefore suggested that data from the operation is used to determine the proportion of 35

36 injuries falling the various severity categories as shown in Table 3-5. Different terminology is used on different operations, but the severity of injuries is invariably recorded in four categories: 1. Non Lost Time Injuries (NLTI) or Dressing Cases 2. Lost time Injuries (LTI) or Disabling Injuries (DI) 3. Reportable or Severe Injuries 4. Fatal Injuries Some judgement is required when deciding to include category 1 or not. It is necessary to first consider the reliability of the data. Then the minimum rockfall size used for the analysis should also be considered (see section 3.2.1). If the smallest rockfall is likely to cause at least a category 2 injury, then the proportions should be estimated excluding category 1 injuries as shown in Table 3-5. Table 3-5: Severity of rockfall injuries and fatalities Year Severity Category Total Proportion (all) 48% 29% 22% 1% Proportion (excluding 1) 55% 43% 2% The disabling injury and fatal injury frequency rates can be determined by multiplying the injury results (Table 3-4) with the severity category proportions (Table 3-5.). This is shown in Table 3-6. Table 3-6: Disabling Injury and Fatal injury Rates 36

37 Individual Worker Category fatality risk per annum Stope Driller 4.0 x Gully Driller 1.3x Stope Timber 3.2 x Winch Driver 2.9 x Team leader 3.7 x Winch Driver 4.2 x Team leader 5.5 x Average 3.6 x DIFR (per 1000 workers per annum) 15 DIFR (per million man hours) 7 FIFR (per 1000 workers per annum) 0.40 FIFR (per million man hours) Evaluating accepted levels of injuries and fatalities In any job or operation, there exists a risk. Some risks are intolerable, some are tolerable and some are acceptable. Intolerable risk is an imminent threat and results in the job or operation being stopped for safety and or economic reasons. However, where careful controls can be implemented to reduce the risk, such risk is known as tolerable risk. Tolerable risk is reduced to acceptable risk levels. As mentioned in section 3.1, defining the accepted risk level for injuries is controversial. Most mining companies target zero harm. While this is certainly a mission, it is not realistic to design for zero fatalities. Mine workers are currently exposed to risks, which do result in injuries and fatalities. There will always be a residual risk that remains as long as mining continues, but this can be reduced through effective risk management strategies. It essential to reduce the number of injuries and fatalities, but realistic targets need to be defined. Different industries have different perceptions on acceptable risk. In the nuclear power generation and dam engineering industries where large numbers of the public could be exposed, accepted frequencies of fatalities have been defined, which are not zero, but are extremely low (Bacher and Christian, 2003, Christian, 2004 and Terbrugge et. al, 2006). The question then posed is What is acceptable risk? Rowe (1979) discusses a risk as being acceptable when those affected are generally no longer (or not) apprehensive about it. In 37

38 defining acceptable risk, Baecher and Christian (2003) quoted the four conclusions drawn by Starr (1969): i. The public is willing to accept voluntary risks roughly 1000 times greater than involuntary risks; ii. iii. iv. Statistical risk of death from disease can be taken as a psychological yardstick for establishing the level of acceptability of other risks; Acceptability of risk is proportional to the third power of benefits; Public awareness of the benefits of an activity influences the societal acceptance of risk. International risk acceptance criteria The F-N curves are a common way of expressing societal risk in various industries. The method is a graphical representation of the relationship between the annual probability of an event causing N or more fatalities, and the number of fatalities, Figure The graph shows zones of negligible, As Low as Reasonably Practicable (ALARP) and Intolerable risk for different industries. The commonly applied benchmark is the annual frequency of a fatality by natural causes of a teenager aged between 10 and 14, which represents the lowest health risk population group. This annual frequency is 10-4 and it represents negligible risk. The negligibility line in Figure 3-10 represents a constant annual frequency of 10-4 for a single fatality. Steffen (2007) refers to this as the divide between voluntary and involuntary risk. This implies that the risk should be negligible, if the risk is involuntary and the ALARP region should be used for setting criteria for voluntary risk. 38

39 Figure 3-10: F-N Graph for bench marking safety impact (Contreras et al, 2006) The ALARP principle is best explained by The Edward versus National Coal Board case (HSE, 1999 Annex 3, paragraph 2) which established this principle as law in Britain in It says, it is stated that the case established that a computation must be made in which the quantum of risk is placed on one scale and the sacrifice, whether in money, time or trouble, involved in the measures necessary to avert the risk, is placed on the other; and that, if it be shown that there is a gross disproportion between them, the risk being insignificant in relation to the sacrifice, the person upon whom the duty is laid discharges the burden of proving that compliance was not reasonably practicable. A diagrammatic representation of this principle is illustrated in Figure However, within the ALARP region a philosophy of constant improvement is implied. 39

40 Figure 3-11: ALARP Principle (from Major hazard facilities regulations guidance notes) An international benchmarking exercise across different industries, including surface and underground mining carried out by Gonzalez and Karzulovic (2004) using the F-N graph is presented in Figure Incidents, which are non-fatal are also represented on the graph. The graph shows the risk associated with underground mines (internationally, including South African mining industry data). It is apparent that the frequency of underground mining fatal accidents is higher than most industries. However, multiple fatalities with large numbers are more common in other industries, since miners are only exposed in relatively small groups. In this analysis, merchant shipping, mobil drill rigs and super tankers are higher risk than underground mining. 40

41 Figure 3-12: International benchmarking using the F-N Graph (after Gonzalez and Karzulovic, 2004) Individual risk can be benchmarked and the value of 10-4 can be treated as negligible risk. 41

42 South African Mining Industry Milestones In 2003, the Tripartite Mine Health and Safety Summit held for the South African Mining Industry agreed on targets and milestones to be attained by 2013 measured in fatalities per million man hours worked: Industry Target: Zero rate of fatalities and injuries Milestones: In the Gold Sector: By 2013 achieve safety performance levels equivalent to current international bench marks for underground metalliferous mines, at the least; In the Platinum, Coal and Other Sectors: By 2013 achieve constant and continuous improvement equivalent to current international benchmarks, at the least. This required a 20% annual reduction of the injury and fatality rates across all commodities. The Platinum milestones are shown together with the actual Platinum industry fatality rates in Figure Platinum Industry Fatality Rate Towards the MHSC's 2013 Milestones Implats Anglo Lonmin Norplats Aquarius Eastplats Xstrata Pt Royal Bafokeng Pt Industry Milestones Figure 3-13: Platinum Industry Fatality Rate Towards the MHSC s Milestones. Disabling injury rates for 12 months are included in Figure 3-14 for comparison. 42

43 Jul-08 Aug-08 Sep-08 Oct-08 Nov-08 Dec-08 Jan-09 Comparison of Platinum Industry - D.I.F.R. (12 Month Rolling) Feb-09 Mar-09 Apr-09 May-09 Jun-09 Jul-09 Aug-09 Sep-09 Oct-09 Nov-09 Dec-09 Jan-10 Feb-10 Mar-10 Apr-10 May-10 Jun-10 Jul-10 Aug-10 Sep-10 Oct-10 Nov-10 Dec-10 Jan-11 Feb-11 Mar-11 Impala Amplats Lonmin Norplats Aquarius EastPlats Royal Bafokeng Xtrata_Pt Pt Industry Figure 3-14: Platinum Industry DIFR Suggested method of evaluating injury risk A comparison between the different ways of calculating accident rates is shown in Table 3-7. The platinum industry 2013 milestone is highlighted in green. It can be seen that this is below the international negligibility criteria of 10-4, which is commendable. The actual fatality rates (Figure 3-13) are currently higher than this, but in the lower part of the ALARP region in Figure One of the challenges of benchmarking the model results with the international and the South African Mining Industry Milestones is that workers are exposed to many risks, not just rockfalls. Rock related fatalities are about 40% of the total rockfall injuries. The target rockfall FIFR for the 2013 milestone should therefore be significantly lower than 0.02 to ensure that the milestone is achieved. The risk evaluation model is used to compare different support systems or safety strategies. An incremental reduction in the number of injuries is a positive improvement. Small changes can lead to a positive improvement. Table 3-7: Comparison between fatal accident rates 43

44 FIFR (per million man hours) FIFR (per 1000 per annum) Individual fatality risk per annum x x x x x x x x x Quantifying economic losses and loss of reserves The cost of rockfall damage is also evaluated using the program RiskEval. In order to quantify the economic losses and loss of reserves associated with rockfalls, one needs to first determine the consequences of rockfall damage, then determine the method of remediation and then calculate the cost of damage. Finally it is necessary to evaluate and compare the economic losses and reserve losses for different strategies Remediation strategies and consequences When a rockfall occurs there are two typical strategies for remediation: Cleaning and re-supporting, and Panel re-establishing The decision as to which strategy to use, depends on the zone in which the rockfall occurs (Figure 3-15). In the face area (zone 1) and sweepings area (zone 2), the decision is invariably based on the size of the portion of the rockfall within the zone. Critical rockfall dimensions must be specified for RiskEval to determine the choice of strategy. These dimensions and strategies should be based on actual practice on the given operation. The gully (zone 3) is always required for access and removal of broken rock. It is therefore invariably necessary to clean and re-support the portion of the rockfall in the gully area. The back area (zone 4) is typically clean, swept and barricaded off. As a result there is no need to remediate and there are no consequences or costs associated with rockfalls in this zone. It is possible that a collapse in the back area affects the stability of the gully and access to the stope 44

45 face, which could result in substantial losses. This eventually is currently not catered for in the software, but should be in future. Zone 3 Zone 1 Zone 2 Zone4 Figure 3-15: Rockfalls in stope zones Cleaning and re-supporting If the portion of area of the rockfall that occurs in zone 1 or 2 is relatively small, it will invariably be cleaned out (Figure 3-16). In zone 1, the stope face cannot be worked until the rockfall is cleaned out and made safe. The gully (zone 3) will always be cleaned irrespective of the size of the rockfall (Figure 3-16). If there is support failure, the support will need to be replaced to ensure safety. Rockfalls between support generally do not require additional support. The losses are determined as follows: Dilution: When the rockfall is cleaned out, it will be transported from the stope to the plant and will be processed along with the ore. The volume of the rockfall is recorded as dilution. Re-support: This is only applied if support failure occurs. The area of the portion of the rockfall within the zone multiplied by a specified factor (>1) is re-supported. 45

46 Production loss: The time it takes to clean is a function of the volume of the rockfall. This may result in a lost shift if the rockfall is large or occurs towards the end of the shift. Lost production time is cumulated for each rockfall. If there are enough small rockfalls, the cumulated time could represent a full shift. The expected or typical production that would have been achieved during this time is the lost production. Plan View 30m Z4 Z2 Z1 Conditions Rockfall Area: <15m 2 Rockfall Height: Any Any height 2m Z3 Stope Section Vie Figure 3-16: Cleaning and re-supporting Re-establishing panels If the portion of area of the rockfall in zone 1 is large (the height is equivalent to or greater than the stoping width and it occupies a significant area), it is not considered safe to clean and resupport. The panel is re-established by mining part of the face and leaving a pillar, as shown in Figure It will be necessary to re-raise for the length of the pillar to re-establish the face. The pillar will be longer than the rockfall by a specified safety margin. The width of the pillar, reraise width and face advance must be specified, which should be determined from the mine standard. If the portion of the area of the rockfall in zone 1 is large enough that it is necessary to stop the entire panel, it will be necessary to re-raise to establish the panel (Figure 3-18). The pillar will be equal to the length of the panel and the width must be specified. Re-raise dimensions and face advance must be specified based on standard practice. It is important to calibrate the modelled frequency of large rockfalls with the actual frequency of large rockfalls on the mine. A reliable set of rockfall data is required for this. 46

47 If the portion of the rockfall in zone 2 is too large to clean and support, the rockfall will not prevent any portion of the face from being blasted. Plan View 10m 4m Conditions 30m 2 > Rockfall Area (Zone 1) >15m 2 Rockfall Height: >1m Z4 Z2 Pillar 30m Area of Lost Sweepings Z1 >1m Stope Z3 2m Section View Figure 3-17: Pillar re-establishing 10m 4m 30m Area of Lost Sweepings Pillar Conditions Rockfall Area (Zone 1) >30m 2 Rockfall Height: >1m >1m Stope 2m Section View Figure 3-18: Full panel re-establishing The losses are determined as follows: Sweepings loss: Broken ore beneath the rockfall will not be recovered. The quantity of sweepings usually scales with the size of the rockfall. The height of the muckpile and the bulk density of the broken material must be specified. Production loss: The software calculates the time it will take to re-establish the panel and the area mined during re-establishment. The production loss is the difference between expected production during this time (based on the called or typical production, 47

48 which must be specified in RiskEval) and the production achieved during reestablishment. Reserve loss: The volume of the pillar is recorded to represent the reserve loss. If the panel crew can be redeployed to another available stope panel, there is no production loss. RiskEval requires the input of a parameter called flexibility, which represents the probability of another stope being available for mining. The production loss is multiplied by 1 - flexibility Cost of rockfall damage The losses are evaluated from a business perspective. The reduction in profits as a direct result of the rockfall damage is considered. Fixed costs and labour are not considered in these calculations, since they remain constant whether the rockfalls occur or not. The cost of certain consumables will be affected by the remediation of rockfall damage. Dilution cost The dilution cost is the cost of transport and processing. The transport costs are usually difficult to extract from the mine financial information, since the expenses are captured by department, rather than by function. However, the combined transport and processing costs are estimated to be 20% of the total operational costs based on SRK project experience. The dilution cost per ton must be specified in RiskEval and this is multiplied by the dilution volume and the specified density of hangingwall rock. Re-support cost For re-supporting, the cost of standard support is used, which is multiplied by the re-support area. Loss of profit due to lost production Loss of profit, due to lost production, is the revenue from production that would have been achieved, less the cost of consumables that should have been used during that mining eg. explosives, support), transport and processing. For loss of production, RiskEval requires the profit margin per ton of ore to be specified. The revenue per ton of ore is readily obtained from mine financial statements or can be determined from the ore grade and long term commodity prices. The cost of consumables is also difficult to extract from mine financial information, but 48

49 can be estimated to be 30% of the total operational costs. The volume of lost production is multiplied by the ore density and the production profit margin per ton of ore. Loss of profit due to sweepings Loss of profit, due to lost sweepings, is the revenue from sweepings that should have been achieved, less the associated transport and processing cost. For loss of sweepings, RiskEval requires the profit margin per ton of ore to be specified. This should be slightly higher than the profit loss (production). The revenue per ton of ore should be assessed in the same way and specified in RiskEval. The cost of transport and processing can be estimated to be 20% of the total operational costs. The volume of lost sweepings is multiplied by the ore bulk density and the profit loss (sweepings) per ton of ore. Each rockfall is simulated and the total cost is sum of the cost of all the individual rockfalls. The costs of rockfalls of various sizes were calculated to demonstrate how the model is applied. The following input parameters were used: Dilution: R134/t of rockfall cleaned, transported and processed Profit loss (production): R711/t of lost production Profit loss (sweepings): R778/t of lost sweepings Re-support cost: R88.50/m 2 Typical costs of rockfalls of varying size up to 15 m 2 is shown in Figure

50 R R R R Lost Production Re-Support Dilution R R Area (m 2 ) Figure 3-19: Cost of small and medium rockfalls. Typical costs of rockfalls of varying size, but greater than 15 m 2, for the different zones, are shown in Figure 3-20 to Figure

51 R R R R R R R Sweepings loss Lost Production R R R Area (m 2 ) Figure 3-20: Cost of large rockfalls in Zone 1 R R R R R R Sweepings loss R R R R Area (m 2 ) Figure 3-21: Cost of large rockfalls in Zone 2 51

52 R R R R R R R Lost Production Re-Support Dilution R R R Area (m 2 ) Figure 3-22: Cost of large rockfalls in Zone Costs of injuries and fatalities The direct cost of injuries and fatalities were estimated using SIMRAC levies, medical costs from a mine hospital and cost data from Marx (1996) that were escalated according to inflation. These are presented in Table 3-8. Table 3-8: Direct cost of injuries Lost time Cost injury Major injury Fatality Lost time SIMRAC Levies R 25 R 494 R 21,480 Medical costs R 10,018 R 10,018 R 1,465 Wages and Compensation R 7,527 R 7,527 R 26,583 Total R 17,577 R 18,177 R 55,528 52

53 These direct costs are not very high when compared with the cost of the rockfall damage associated with large rockfalls (see Figure 3-20) and it is unlikely that they will play a major role in the investigation. However, substantial indirect costs are incurred when the mine is closed. The mine may be closed during rescue operations and the Department of Minerals Resources may instruct the mine to close (Section 54 of the MHSA). Mines have been closed for 5 to 10 days by the DMR after fatalities. Depending on the production revenue and the amount of production lost, these costs can vary, but are always significant. For example, if 50 panels are stopped for 5 days, the loss would be R11,720,000, using the profit loss (production) in section The injury costs are estimated from the total number of injuries (section 3.3.5), the severity of the injury (section 3.3.6) and the cost of the injuries, including lost production. The total injury costs should be added to the total cost of rockfall damage to provide the total expected cost of rockfalls Economic evaluation When carrying out an economic evaluation, one needs to consider both the cost of the risk control (eg. support system) and the expected cost of rockfalls. Figure 3-23 shows that the expected cost of rockfalls increases with increasing hazard. Conversely, the cost of the risk control must be increased to limit the rockfall hazard. The total cost is the sum of the cost of the risk control and the expected cost of rockfalls. The optimum cost lies somewhere in the middle. It is suggested that two scenarios with different risk controls are compared. 160 High Cost Scenario 1 Scenario 2 Cost of of risk Risk control Control Expected costcost Total Cost cost Low 0 Low High 1 3 Rockfall 5 Hazard

54 Figure 3-23: Economic evaluation It is essential to ensure that the same production over the same period is used for both scenarios. It is suggested that annual costs are used and the exercise can be scaled to represent the whole mine. It should be emphasised, that while it is not essential that the model is extremely accurate, since the evaluation is made on a comparative basis, it is important that the frequency of large rockfalls (section 3.4.1) and the financial input parameters are reasonable. If the frequency of large rockfalls or the profit margins are overestimated, the results will be biased towards the more effective and more expensive risk control. The model can be calibrated by analysing an existing risk control or support system and comparing the modelled frequency of large rockfalls with the actual frequency of large rockfalls. A comprehensive rockfall database (at least for large rockfalls) is essential for a reasonable calibration. If the rockfall database is not comprehensive, then the rockfall injury frequency (section 3.3.5) should be used for calibration, since the rockfall injury data is invariably more accurate Once off implementation costs and reserve loss If a new risk control is implemented, there is often a large implementation cost. For example, the introduction of backfill requires the installation backfill plant. Reserve losses effectively shorten the life of the mine and this loss should be accounted for at the end of the life of the mine. It is suggested that a discounted cashflow is used to determine the net present cost of the two scenarios. Implementation costs can be treated as a capital cost at the start of the cashflow and the reserves are accounted at the end of the cashflow. 3.5 Summary This risk model can be used to evaluate uncertainty and determine the risk of rockfalls. It can be useful for comparing two support systems, to determine which will be more beneficial to the mine. The number of rockfall injuries and fatalities for each support system can be estimated. It is suggested that the 2013 safety benchmarks should be used as a target. The cost of rockfalls can be estimated using the model. The total cost for each support system can be compared. 54

55 Both the economic risk and safety risk need to be evaluated before deciding which is the better support system. The existing software JBlock was enhanced to perform the rockfall simulation and a new program (RiskEval) has been developed to perform these analyses. The initial part of the analyses involves the collection of joint data through underground mapping, collecting financial data and investigating the mines methods of rehabilitating rockfalls. This will involve a considerable amount of work. Once this has been done, the analyses are very quick and easy. 55

56 4 Validation of the Risk Evaluation model A representative rockfall distribution for the Ground Control District (GCD) is a key input in the evaluation of rockfall related risk using the model described in section 3. It is necessary to evaluate and calibrate the results to make sure that the rockfalls generated are in reasonable agreement to actual rockfalls occurring. The evaluation and calibration exercise of a keyblock stability software has never been done before. Rock joints underground delineate the rock mass into rock blocks that may eventually result in rockfalls. Underground joint and rockfall mapping was carried out on Merensky Reef and UG2 Reef panels in a Platinum mine in order to evaluate and calibrate JBlock rockfall frequency distributions. The joint characteristics and method of joint data collection is described in section 4.1. The underground mapping sites (section 4.2), geotechnical information (section 4.3), rockfall mapping data (section 4.4), rockfall and injury data collated by mine personnel (section 4.5) and the calibration of the software (section 4.6) are described in this chapter. 4.1 Joint Characteristics and Methods of Joint Data collection Since joints have a significant influence on the engineering properties of rock masses, it is vital to have reasonably accurate descriptions of the properties of such joints. Joint surveys to statistically describe a rock mass are an integral component of site characterization studies in rock engineering as stated by several authors (Baecher, 1983; Call and Nicholas, 1978; Dershowitz and Einstein 1988; Barton, 1995). The motivation for this exercise is the need to take the distributive character of joint properties into account and obtain a representative distribution for each property to be used in a stability analysis. The important properties in describing joints are orientation, spacing, length and shear strength properties (Priest and Samaniego, 1983, Priest, 1985; Goodman and Shi, 1984). determined for each joint set Joint Properties These properties are normally The present systems of joint data collection can be from either orientated drill cores and/or exposed rock surfaces i.e. outcrops and or tunnel walls. Table 4-1 below lists joint properties and the corresponding source of information for that particular property. This data can be one 56

57 dimensional or at best two dimensional. In this research, joint data was obtained by the careful mapping of underground stope hangingwall exposures. Table 4-1: Joint Property and Source Joint Property Orientation Spacing Length Shear Strength Properties Source - Measurement of: Oriented borehole cores, outcrops and excavation walls Oriented borehole cores, excavation walls, outcrops Excavation walls and outcrops Excavation walls and borehole cores (observation of joint surfaces and/or laboratory testing) Joint Orientation Orientation or attitude of a joint in space, is described by the line of maximum declination on the discontinuity surface measured from the horizontal and dip direction of this line is measured clockwise from the true north (Figure 4-1). It has been noted (Priest and Samaniego, 1983) that in many cases joints generally have a planar geometry and are oriented in sub-parallel groups or sets. It is usual to quote joint orientation data in the form of dip (two digits)/dip direction (three digits) for example 46⁰/009⁰ and 70⁰/320⁰. Joint dip and dip direction can be measured simultaneously using a Breinthaupt magnetic compass. N=000⁰ α β Figure 4-1: Definition of dip (β) and dip direction (α) Joint orientation data can be represented on stereographic projections (Shi and Goodman, 1981, 1985; Priest, 1985; 1993) for the purpose of visualisation and clustering the joints into sets. Joint sets are identified from joint orientation information and stereographic projection software such as DIPS, which is used for this function. Those joints that do not fall within the defined sets are known as random joints. In the further analysis of the joint data, all joints that are included in a defined set are given equal weighting for any characteristic property being 57

58 investigated, thus the definition of the set has to be as comprehensive as possible, providing the mean, minimum and maximum or standard deviation. The shapes of blocks are determined by the mutual orientation of the joint sets. Further to this, the relative orientation of the proposed excavation to the joint sets will control the possibility of unstable conditions. The importance of joint orientation increases when the other conditions for instability are present, such as low joint shear strength, and a sufficient number of joint sets to delineate a failure mode. Orientation has been observed to follow a normal distribution (Stacey and Bartlett, 1990; Gumede, 2006; Cameron-Clark and Kirsten, 1986, Grady, 1983). However, other researchers (Priest, 1984; 1993; Brown, 2003, Thompson, 2002) have found the Fisher distribution (Fisher, 1953) to represent the orientation distribution better. JBlock currently uses a normal distribution currently, but should be changed in future to use a Fishcer distribution. Correction for sampling bias to joint orientations has been done (Kulatilake et al., 1990) and shown to be dependent on joint shape and size. However, the last two factors are difficult to measure in 2 D making the correction exercise difficult. Joint Spacing Joint spacing is the perpendicular distance between two adjacent joints of the same set and is often expressed as a mean spacing for a particular joint set. However, in addition maximum and minimum spacing estimates are required when working with JBlock. The spacing of joints determines the sizes of the blocks making up the rock mass. This parameter can be measured on rock exposures on surface or underground and it can also be estimated from orientated drill cores. It should be noted though that spacing cannot be measured directly from cores when they intersect the joint at an angle nearly parallel to the joint. It is recommended that a sampling length of not less than 3m be used and that the sampling length should be greater than ten times the estimated spacing for reliable results (ISRM, 1978). Joint spacing has been found to follow either an exponential distribution or a lognormal distributions (Priest and Samaniego, 1983; Steffen et al, 1975; Cameron-Clark and Kirsten, 1986, Grady, 1983; Gumede, 2006; Baecher and Einstein, 1977; McCullagh and Lang, 1984; Pahl, 1981; Warburton, 1980). JBlock uses the former distribution in creating rock blocks. An error in spacing calculation is often brought about by the sampling line cutting across the joint set at an angle less than 90 resulting in the overestimation of joint spacing (Terzaghi, 1965). This error is illustrated in Figure 4-2 and can be corrected as follows: 58

59 Where: l is the distance between successive joints of the same set along the scanline d is the true spacing distance θ is the angle between the scanline and the mean orientation of the joint set. In order to obtain relative angle (θ) between the joint and the scanline, the respective orientations of the joint set and scanline should be known. Use of a number of scanlines with different with different orientations can be used to improve spacing information. The complete calculation of the spacing error is a three dimensional problem, this is a complex exercise and will not be described here. If the plunge of the scanline is less than 10, the three dimensional problem can be simplified by a two dimensional solution. Scanline(Tape) d l Joint θ Figure 4-2: Joint spacing error correction Length Joint traces are lines formed by the intersections of joints with excavations in rock. The individual lengths of these traces indicate the extent a joint persists in the rock mass and is denoted by the straight line distance between end points of the joint. Joint length measurements can be taken from natural outcrops, underground and open cut excavation walls. 59

60 Persistence is an equivalent to joint length when describing joints in three dimensions. However, frequently rock exposures are small compared to the area or length of continuous joints and so joint length can only be estimated. Joint trace lengths have been found to follow lognormal or negative exponential (Baecher and Einstein, 1977; Bridges, 1976; Zhang, 2004). Joint lengths measurements are often biased and difficult to determine accurately. The main reasons for this bias are that joints may extend beyond the boundaries of the exposure or joints less than a chosen minimum length e.g 0.5 m are ignored. These errors are discussed in section Joint shear strength In addition to joint geometry when analyzing rock block failures, joint shear strength is also a very important property. This property is a measure of the resistance to sliding of block surfaces on opposing sides. The nature of the joint walls and infill material between the walls on a joint can aid or restrict the shearing behavior. A number of models have been developed for determining the shear strength of joint surfaces. These models are among others, the Coulomb model (1776), Landanyi Archambault model (1969), Bilinear shear strength model (Roberds and Einstein, 1978), Barton-Bandis model (Barton, 1973; 1976; Barton et. al, 1985; Bandis et.al, 1983) and Barton model (2002). The most widely used models are described below: Coulomb model The simplest shear strength model of discontinuities is the Coulomb failure criterion. Coulomb (1766) suggested that the shear strength of a surface is made up of two components i.e, a constant cohesion and a normal stress-dependent frictional component This model can be represented by the following equation: (3) Where joint shear strength joint cohesion effective normal stress internal friction angle of the joint 60

61 The weakness of this model is that it predicts that the relationship between joint shear strength and normal stress is linear. This assumption leads to an overestimation of the shear strength at high normal stress. Bilinear shear strength model Usually the shear-normal stress relation of the joints is non-linear. Patton (1966) addressed the weakness of the Coulomb model by formulating a bilinear model as shown in Figure 4-3. The shear strength of a joint at low normal stresses is given by: Where Peak shear strength Effective normal stress Basic friction angle for a apparently smooth surface Effective roughness angle of a saw tooth face The main weakness of this model is that it is only valid at low normal stresses. At higher normal stresses, the strength of the intact material will be exceeded and the teeth will tend to break off and there is a transition from sliding along the joint to fracturing through the intact material. This then results in a shear strength behaviour, which is more closely related to the intact material strength than to the frictional characteristics of the surfaces as represented by the change in gradient in the second line in Figure 4-3. As a result Jaeger (1971) proposed another shear strength model to provide a curved transition between the straight lines of the Patton model. Figure 4-3: Patton s experiment on the shear strength of saw-tooth joints (Patton 1966). 61

62 Barton-Bandis model This model is an improvement of the Patton s approach by Barton (1973, 1976). In this case changes in shear strength with increasing normal stress are gradual rather than abrupt. Barton studied the behaviour of natural rock joints and proposed that Patton s equation could be rewritten. In the Barton-Bandis model, simple measurements of joints were developed to have a practical way of evaluating the shear strength of joints (Barton, 1976; Barton and Choubey, 1977; Barton and Bandis, 1990). The joint shear strength for the Barton-Bandis model can be represented by the following equation: Where joint shear strength effective normal stress basic friction angle Joint Roughness Coefficient in the range 0(smooth) to 20(rough) Joint Wall Compressive Strength The weakness of this model is that its focus is on expressing joint behaviour at peak shear strength. This therefore means that it cannot be relied upon when describing the pre-yield and post-yield behaviour of rock joints. Barton model Barton (2002) provided a method of determining friction angles as follows: where: is the joint friction angle is Joint roughness condition ( = smooth for sepentinized joints and rough undulating for calcite filled joints and joints with no fill) is the joint alteration number ( = unaltered for joints with no infill, nonsoftening sandy particles for calcite fill and clay mineral fillings of varying thicknesses for serpentinised joints) 62

63 This model has an underlying assumption that the joints are cohesionless. This is normally the case with sheared joints. This model provides an easy, cheap and intelligent field estimate of of joint shear strength properties. The other important joint properties are, thickness of filled joints, type of infill and water conditions. These properties contribute towards the joint shear strength. It should be noted here that joint set properties contributing toward the shear strength cannot be described by distributions and therefore mean values will be used Joint Mapping Techniques In mapping exposures, four main techniques are used; these are spot mapping, scanline mapping, area or window mapping and photogrammetric mapping. Spot mapping This is where the observer selectively samples only those discontinuities that are considered to be important. This method has a weakness in that it ignores the influence of the other joints towards excavation stability. Random joints have been shown in literature to have a significant influence towards rock block formation (Brown, 2003). Area or Window mapping This technique involves selecting an exposed area of the rock face and mapping only all joints within the selected area as in Figure 4-4 (Brady and Brown, 2004; CSIRO, 2009, Pahl 1981). The portions of the joint traces that are within the window are measured, while the portions of traces intersecting such a window but lying outside are ignored. This method has an advantage over spot and scanline mapping because it reduces sampling biases for orientation and size. However, it suffers practical difficulty in application to underground operations and sensoring bias of joint lengths proceeding beyond the area window. 63

64 Figure 4-4: Area mapping (After Zhang and Einstein (1998).) Photogrammetric mapping More recently photogrammetric mapping making use of high resolution cameras in mapping joints as shown in Figure 4-5is being used in joint mapping. There are three steps involved in mapping joints using this method as listed below (Birch, 2006; Gaich et al, 2006): Using high resolution, stereoscopically arranged, digital cameras to collect images of rock face Producing an oriented point cloud, and a digital 3D surface called a Digital Terrain Model (DTM) which is made of spatial data points of the face Analysis of the DTM to characterise the rock mass (e.g. take joint dip direction, dip, spacing measurements etc.). This can be seen on the right side of Figure

65 Figure 4-5: Photogrammetric mapping (From Sturzenegger and Stead, 2009) The only drawback of photogrammetric mapping over traditional mapping is that it is relatively expensive. However, photogrammetric face mapping techniques far outweighs traditional methods of field data capturing (Tonon and Kettensette, 2006; Poropat, 2006). The advantages of photogrammetric face mapping techniques over traditional methods are listed below: The ability to analyze large portions of rock masses, including inaccessible areas by capturing images from up to 3km. Limited size and access to rock exposures for mapping had been cited by Villaescusa (1993) as a hindrance to describing joint set characteristics. The ability to collect large data sets yields a more realistic picture of the fracture orientations; new fracture sets are oftentimes discovered that would be missed by manual investigations. New insights into the rock mass structure are gained by selecting fractures or fracture sets on a stereonet and seeing their location on the 3D model. The ability of zooming in and out of the face allows one to identify features, such as major shears or fractures, which are otherwise not apparent when working close to the rock face. These methods are typically five times faster than manual data collection, without taking into account the much larger number of data points. 65

66 Permanent documentation of the rock face condition for reporting and contractual or legal issues. Permanent documentation of excavation stages with the added bonus of precisely determining excavation volumes. When each stage is documented, valuable information can be extracted, such as fracture persistence and fracture clustering that allows for a reliable three-dimensional characterization of the rock mass. It is relatively easy to use i.e. there is no need for a photogrammetry expert; as was the case with traditional photogrammetry, picking corresponding points in a pair of photographs required a well-trained photogrammetry expert. This expert is now substituted by the software adjustment capability. Digital photogrammetry is relatively range invariant. Indeed, if the object distance changes, one may change the focal length with the aim of keeping the image on the camera sensor the same length Scanline mapping This method involves setting up a line on the surface of a rock mass and mapping all joints intersecting a given straight line as in Figure 4-6 ( Priest and Hudson, 1976; 1981; Baecher 1983; Steffen et al, 1975, Haines and Marker, 1992; Zhang, 2004; Goodman, 1995).Brady and Brown (2004) suggested the use of a measuring tape pinned to the rock face with masonry nails and chalk lines drawn on the face as a scanline and only the properties of those joints that cross the tape are recorded. The tape needs to be taut at all times along the length in-order to have reliable joint spacings. To reduce orientation bias, a three dimensional view of the joint data is needed. This is achieved by having two orthogonal scanlines (one on dip and the other on strike) in the stope hangingwall and a borehole drilled into the hangingwall. This mapping technique has been recommended when using probabilistic methods for key block analysis (Kemeny and Post, 2003). An orientated drill core is an indirect scan-line that can be used to measure orientations and joint spacings only in unexposed rock masses. Scanline mapping of outcrops and surface exposures offers the extra opportunity of measuring joint length which cannot be done on oriented drill cores. Circular scanlines are also used for joint mapping and offer an advantage of eliminating directional bias. Scanline mapping is by far the most common mapping technique in use because of its relative ease to use. One drawback of this method is that it provides unreliable information when the mean orientation of a joint set is less than 20 from the sampling line (Baecher and Einstein, 1977). 66

67 Figure 4-6: Straight scanline mapping (Zhang, 2004). Table 4-2 is a snapshot of the log sheet used in scanline mapping. Recorded are the location of the joint along the scanline, this is used to calculate joint spacing. Also recorded, are the joint orientations, length and how the joint ends with the J indicating ending at another joint, R ending in the rock and O indicating that the joint progresses beyond the exposed rock surface. Table 4-2: Example of a scanline logging sheet Orientation Infill Joint wall Joint set Joint wall Rock or Location roughness Joint Joint Dip Thicknerationess Micro Macro Sepa- Hard- Type Dip Type length ends structure Direction Joint 0.27 Pyro mm J/R Joint 0.46 Pyro mm J/J Joint 0.82 Pyro mm J/J Joint 1.02 Pyro mm Calc 1mm O/O Joint 1.10 Pyro mm J/O Joint 1.36 Pyro mm Calc 1mm J/O Joint 1.44 Pyro mm Sep 2mm J/O From field mapping experience, Priest (1993) suggested that joint observations need to be made using scanline mapping techniques. The lower number and higher numbers are ideal for a rock mass containing three and six joint sets respectively. On the other hand, Robertson and Mac (1970) suggested that 100 observations be made per joint set in order to reduce the effect of potential errors. Call et al (1976) has suggested that the number of observations required be dependent on the fracture intensity and the number of sets. The scanline mapping technique is the one that was chosen and utilised in this particular exercise because it is comprehensive, cheap and will provide the data required for a probabilistic keyblock stability analysis. 67

68 4.1.3 Joint Mapping Errors The quality of the joint input information will determine the quality of the output from JBlock. It is important to highlight the potential sources of errors in carrying out joint surveys using the scanline method. The potential sources of error in joint mapping apart from human bias and instrumentation error are joint size or length, orientation, censoring and truncation (Baecher and Lanney, 1978; Baecher, 1983; Einstein et al., 1979; Priest and Hudson, 1981; Kulatilake and Wu, 1984; Mauldon and Mauldon, 1997). These errors are discussed in detail below; Joint Size Joint size or length bias is the most important of these errors as described by various authors. This bias leads to non-equal sampling probabilities among traces of varying lengths in two ways: (a) a larger discontinuity is more likely to appear in an outcrop than a smaller one; and (b) a longer trace is more likely to appear in a sampling area than a shorter one. Censoring Joint censoring bias takes place when traces observed on a joint exposure run off into the rock walls and cannot be observed in their fullness. This gives a false trace length of the joint in that it reduces the effective joint trace length. Truncation This bias takes place when trace lengths below a cut off length (0.5 m for this study) are ignored in data collection. This has an effect of increasing the sample mean of the joint trace lengths. Joint spacing values are overestimated in this case as some of the joints in the same set are omitted. However, the short length joints are less likely to form unstable blocks. Orientation Orientation bias is where joints sub-parallel to the rock face have less chance to intersect the rock face than joints perpendicular to the rock face (Terzaghi, 1965; Stacey and Bartlett, 1990). This usually arises from outcrops that may form along joint surfaces resulting in a strong possibility that an entire set of joints being systematically underestimated in the survey results. This error can be reduced significantly by using three mutually orthogonal scanlines in the same area. An orientation bias correction has been effected for all the mapped joint data in the current study. 68

69 4.2 Field Mapping Sites In this section, a detailed description of the field mapping sites and mapping programme is made. The success of the field evaluation exercise hinges on conducting the mapping programme at a suitable site that is typical of the Ground Control District (GCD) Site Selection In-order to be confident about the usefulness of the joint data to be acquired for the evaluation process, a thorough process of selecting suitable mapping sites was undertaken. There are two main mineral deposits that are being exploited in South Africa, that is the gold bearing Witwatersrand deposit and the Bushveld Complex. The Bushveld Complex is a huge layered igneous intrusion, in which platinum group metals, comprising almost 70 % of the world s known resources, are concentrated into two exploitable horizons. The Bushveld Complex was selected as candidate mapping environment because it is often characterised by blocky hangingwall conditions as alluded to in the photographs by Rangasamy (2010) and accident statistics from the DMR. Many rock mechanics related problems in this area are driven by the blocky nature of the hangingwall. However, the ground condition in this environment varies from area to area with some places having competent and non-blocky hangingwalls. A search was made for a mine with a significantly blocky hangingwall and where rockfalls are relatively common. A mine with a well maintained rockfall database was selected and an initial visit to the mine was made to discuss the conditions expected at the shaft selected. Mine plans were scrutinised to ascertain stope layouts, face advance rates, rock mass ratings, F.O.G statistics for the GCD and logistical detail. A subsequent underground visit was made to the selected stopes on the Merensky and UG2 Reefs to confirm the suitability for investigation Mapping Programme During the process of visiting candidate mapping stopes as part of the site selection process, the research teams were trained on the activities involved in the mapping exercise. The mapping programme involved two teams of two people each mapping joints and recording all the rockfalls that took place in both the UG2 and Merensky Reefs in the selected stopes. The mapping involved the teams going underground with the mining morning shift crew. The mine team leader and his crew would make safe in the working area in the presence of the mapping team and loose rocks that were barred down were noted and recorded. Once the working area had been made safe, the 69

70 mapping team again scanned through the hangingwall to search for rockfalls that had already occurred or extension of previously recorded rockfalls. Joint mapping would then follow after all the rockfalls had been mapped. These procedures were repeated daily for both stopes on each reef as the mining faces advanced. The on-site mapping programme lasted 5 weeks. Midway through the mapping programme, mapping teams took a one week break to evaluate and review the data that had been gathered and make suitable adjustments to improve on the quality of information being collected Site Description The two sites in which the mapping was done for each reef will be described in detail in this subsection. The summary information on mapping area covered for both the Merensky and UG2 Reefs during the mapping exercise is presented in Table 4-3. For each stope reference positions were marked on the excavation walls from the start position of the mapping programme and the advance measured until the end of the mapping exercise. This information is important for the purpose of normalising the mapped rockfalls to the area mined. Table 4-3: Mapping area covered UG2 Stope Face Face Area Advance(m) Length(m) (m 2 ) Merensky 1 South Merensky 2 South UG2 4 North UG2 3 North The two panels selected for the mapping exercise in this reef were on the same raise line and are shown on plan in Figure 4-7. They are at a depth of approximately 674 m below surface. The dip and dip direction of these stopes is 10⁰ and 045⁰ respectively. For the UG2 Reef mapping was mostly done on the 15C53 4N stope. The 15C53 3N stope was stopped two weeks into the mapping programme because of the bad hangingwall conditions brought about by the pothole shown on plan. The mining face for the 15C 53 4N stope was 28 m long and was a distance of 52.4m from the main raise line close to the end of the mapping programme. There was no other mining activity in the immediate vicinity of these stopes. 70

71 15C53 4N 15C53 3N Figure 4-7: Mapped stopes in the UG2 Reef Merensky Reef The two stopes selected for the mapping exercise in the Merensky Reef are shown on plan in Figure 4-8 and are at a depth of approximately 1093 m below surface. The dip and dip direction of these stopes is 10⁰ and 045⁰ respectively. Most of the mapping for the Merensky Reef was done on the S stope. When the mapping programme began, the S stope was still being ledged. Mining production on this stope commenced two weeks into the mapping programme. The mining face for the S stope was approximately 29 m long and was a distance of 75 m from the main raise line at the end of the mapping. There was no other mining activity in the immediate vicinity of these stopes. 71

72 Figure 4-8: Mapped stopes in the Merensky Reef Support Systems The support layout and properties used for supporting the UG2 and Merensky hangingwalls is presented in Figure 4-9 and Figure 4-10 respectively. The UG2 stope support system comprises of 200mm diameter mine poles spaced 2 m on strike and 1.5 m on dip with a maximum distance of 4 m from the face after the blast. Breaker lines of 3 cluster sticks are installed 10 m apart on strike. The Merensky stope support system is made up of a combination of 1.2 m long hydrabolts and mine poles. The hydrabolts are on a spacing of 1 m on strike and 1.5 m on dip and are installed 0.5 from the face after every blast. The mine poles are spaced 2 m on strike and 2.4 m on dip with a maximum distance from the face after the blast of 6 m. A line of three 1.2 m long hydrabolts are installed in the gully for both the UG2 and Merensky Reefs at a spacing of 1 m on strike by 0.5 m on dip. The Hydrabolts are installed at angles of greater than 55 in the gully and greater than 70 in the stope. 72

73 Figure 4-9: UG2 support standard 73

74 4.2.5 JBlock Input Data Figure 4-10: Merensky support standard The statistical keyblock stability software, JBlock is used in simulating rockfalls in underground stopes. The stope geometry is drawn and saved as a JBlock file format called an excavation file. The orientation of the stope hangingwall, advance per blast, the advancing mining face and mining direction for the excavation are specified at this stage as well. Support elements are added in the excavation to create a support file. The spacing of the support elements is as presented in section and properties of these supports are as follows: Hydrabolts: Mine Poles: installed as grouted tendons in JBlock with a peak strength 100kN and grout bond strength of 320kN/m (Hyrabolts contact) peak strength of 270kN 74

75 Hazard zones have been applied in JBlock as discussed for the UG2 and Merensky Reef analysis, these are listed below: Face area, this is the zone from the face line to 4 m behind. This is the area in which most of the work i.e, drilling, blasting, supporting and scraper cleaning, in the stope is done and most of the workforce is confined to this area Sweepings area, this is the zone from the face area boundary going back to 10 m. In this region, cleaning of ore left behind during scraper cleaning is done and appropriate man hours are allocated to this area. Gully area, this is the zone used for access of the stope and movement of ore from the stope. Back area, this is the zone behind the sweepings area, where access is restricted. The hazard zones together with images of the excavation and support defined in JBlock are presented in Appendices B.3 and C.3 for the UG2 and Merensky Reefs respectively. The hazard zones have been labelled for the Merensky Reef excavation and support image. 4.3 Geotechnical Data for JBlock Analysis Joint information is required in order to generate keyblocks for use in a stability analysis. This section describes the method used and the data gathered during joint mapping Data Collection The joint mapping exercise took place in both the UG2 and Merensky Reefs at a platinum mine in the Bushveld Complex. The selection and description of sites in which the mapping was carried out is described in section 4.2. The scanline joint mapping technique was employed. Its relative ease of use, cost and comprehensiveness in collecting joint information favoured its selection over the other methods. The distributive nature of the joint properties which is critical for a statistical analysis is also comprehensively covered by this method. In carrying out the scanline mapping underground, a measuring tape was laid in the stope hangingwall by tying it at short intervals to the support elements in-order to make it taut. To reduce joint orientation bias, two tapes were laid in the hangingwall orthogonal to each other such that joints sub-parallel to one tape would be intersected by the other tape. One tape was laid parallel to the face from the bottom of the panel (approximately 29 m long) to the top. The other tape was laid parallel to the gully and at least 30 m long. All joints greater than 0.5 m in 75

76 length that intersected the scanline were considered to be of significance and were mapped for orientation, location, length and shear strength properties (joint roughness and alteration numbers). This lower cut off joint length has also been used by Dight and Baczynski (2009). The orientation of the joint was measured on an exposed surface of the joint in the hangingwall using the Breithaupt compass (Figure 4-11). This instrument has an advantage of simultaneously recording both the dip and dip direction of the joint. Care was taken in taking the orientations as far away as possible from the magnetic influence imposed by metallic objects in particular roof bolts. An investigation of the magnetic influence of the orebodies on the compass was done and it was found out that this is negligible. The location of the joint is the distance along the tape to the point where the joint crosses the tape. The length of the joint was measured on the strike extent of the joint where it intersects the hangingwall. The joint shear strength properties were recorded using the method described by Barton (2002). The joint walls and infill material were examined by visual analysis, touching the surfaces and gouging out the infill to determine it properties. Blast fractures were excluded from the joint mapping exercise. The orientation of each scanline was also recorded as this is important for spacing error correction as illustrated in section This exercise was repeated after a number of blasts both on dip and strike as the mining face advanced. Thorough washing down of the hangingwall before the mapping exercise assisted in joint identification and collection of joint properties. All the information collected was recorded on log sheets that are presented in the appendices. Figure 4-11: Breithaupt compass 76

77 4.3.2 Data Analysis After the joint mapping exercise, the initial step in analysing the joint data is to group the orientations into sets of sub-parallel joints. For this purpose, the recorded joint orientations are entered into DIPS and processed into sets. DIPS is a Rocscience program designed for the interactive analysis of orientation data. Since a magnetic compass was used in collecting the orientation data, a magnetic declination of 17 degrees west at this mine was used to effect a correction on the recorded orientation of the joint data to obtain the true orientation. The standard deviation for joint dip direction was obtained by using the Fisher distribution in DIPS whilst the standard deviation for dip of joints has been approximated using the normal distribution. With the joint sets having been defined, further processing of joint properties is done per each joint set. Since joint lengths were challenging to measure accurately underground due to limited exposures, average joint lengths have been calculated from the recorded approximations. The most prominent joint sets (with the highest number of poles) and longest joints from mapping were given the longest lengths in the joint parameters. This is consistent with observations in the Bushveld Complex open pits by Bye and Bell (2001) which suggest that some of the prominent joints are continuous laterally and vertically over hundreds of metres. The approach adopted here is supported by extensive joint mapping experience from other sites (Call et al, 1976), which pointed out that minimum joint length is more accurately defined than maximum length because the maximum length is usually beyond the excavation walls. However, the logging exercise in this study, indicated how the joint ends or whether it progressed beyond an exposure (Appendices B.1.1 and C1.1). The joints that ended in the rock are designated R, those that ended at a joint J whilst the joints that extended beyond the exposure O. From the logs, it can be observed that the majority of the joints continued beyond the exposed area. Joint spacing has been determined as the distance between successive joints of the same set. The joint locations are used to calculate spacing. In order to obtain the true spacing, an orientation correction is applied using the relative angle between the joint set and the scanline as described in section4.1. Joint friction angles have been calculated for each joint set based on the Barton (2002) model as described in section In general there are three clusters of joint friction angle data. The first cluster comprises the joints with serpentinite infill of varying thicknesses, the second comprises calcite infilled joints and lastly joints without any infill. Within these clusters, the thickness of the infill material also differentiates these friction angles. This information was 77

78 utilised in calculating the friction angles. A normal distribution curve was fitted to the friction angle data in order to obtain the mean and standard distribution for each joint set. An example of the curve fitting is presented in Figure 4-12 for the Merensky J1 joint set. The friction angle graphs for each joint set are presented in Appendices B1.2 and C Cummulative Distribution Normal Dist Actual Data Friction Angle (⁰) Figure 4-12: Friction angle distribution curve for Merensky J1 set UG2 For the UG2 Reef, a total of 273 joint occurrences were mapped over a total scanline distance of approximately 205 m. The joint orientations from mapping have been plotted as poles in DIPS for the purpose of further processing of the joint data. Clustering of the poles of the joints resulted in the delineation of two main joint sets, J1 and J2 and two random sets as shown in Figure The poles for each joint set are plotted in a distinct colour. Joints in the J1 and J2 sets are the most frequent and as such the joint spacings for these sets joints are small as compared to the other less frequent random joints which are sparsely spaced. Summarised joint properties from this analysis are listed in Table

79 N 4m 4w R2 Reef Plane Set 1 [102] 2 [165] 3 [10] 4 [15] [no data] [7] W E J2 2m 2w R1 3m 3w 1m 1w J1 Equal Angle Lower Hemisphere 299 Poles 299 Entries S Figure 4-13: Dips pole plot of mapped joints in UG2 stopes. Merensky A total of 304 joint occurrences were mapped over a total scanline distance of approximately 187 m on the Merensky Reef hangingwall. The joint orientations from mapping have been plotted as poles in DIPS for the purpose of further processing of the joint data. Clusters of joint orientations occurred that delineate three main joint sets (J1, J2 and J3) and one random set as shown by the DIPS pole plot in Figure The poles for each joint set are plotted in a distinct colour. Joints in the J1, J2 and J3 are the most frequent and as such the joint spacings for these sets are small as compared to the other less frequent random joint which is sparsely spaced. Summarised joint properties from this analysis have been presented in Table 4-5. It is apparent from this that the degree of jointing in the Merensky Reef stope hangingwall is higher than that of the UG2 Reef stope hangingwall. 79

80 N Set Reef Plane 1 [137] 2 [96] 3 [60] 4 [8] W 4m 4w R J3 3m 3wE [no data] [1] J1 1m 1w J2 2m 2w Equal Angle Lower Hemisphere 302 Poles 302 Entries S Figure 4-14: Dips pole plot of mapped joints in Merensky stopes JBlock Geotechnical Input Data Geotechnical input data used in JBlock is presented in this sub section. The processed joint mapping data provides JBlock input parameters. Since JBlock is a statistical keyblock stability programme, the distributions of the input parameters are also required. Three of the four joint sets defined for the UG2 Reef are steep dipping with a dip ranging between 71 and 90. The other joint set comprises the low angled joints. The low angled joints assist in defining keyblocks that will eventually fallout from the hangingwall. The height of a parting plane representing chrome stringers or Triplets in the UG2 hangingwall was obtained from the geology department at the mine. The parting plane is often involved in creating loose keyblocks in the hangingwall. All the joint properties used in JBlock for the UG2 Reef are listed in Table 4-4. Table 4-4: JBlock joint input properties for UG2 Reef 80

81 Set Dip Orientation (⁰) Dip Direction Mean Stdev Mean Stdev Friction Angle (⁰) Spacing (m) Length (m) Mean Stdev Min Mean Max Min Mean Max J J R R Parting plane at 2.5m ± 0.5m tensile strength of 15kN ± 2kN The majority of the joints for the Merensky Reef are steep dipping with three of the four having a steep dip ranging between 82 and 88. The other joint set is made of the low angled joints. The low angled joints are important in assisting the creation of keyblock release surfaces. All the joint properties used in JBlock for the Merensky Reef are listed intable 4-5. Table 4-5: JBlock joint input properties for Merensky Reef Orientation (⁰) Friction Angle (⁰) Spacing (m) Length (m) Set Dip Dip Direction Mean Stdev Mean Stdev Mean Stdev Min Mean Max Min Mean Max J J J R Underground Rockfall Mapping The thorough selection process for ideal underground mapping panels described in section 4.2 was focused on collecting a useful set of rockfalls. The underground mapping of rockfalls and joint mapping were done in the same stopes. The main objective of this activity was to record a set of actual rockfalls and their respective properties and use this information to compare with JBlock simulated rockfalls based on joint mapping data from the same area Data Collection The rockfall mapping process was a daily activity from Monday to Friday. On the other hand, the mining roster took place from Monday to Saturday. Underground rockfall mapping on each Monday would endeavour to cover the falls that occurred on Saturday. 81

82 On all mapping days, each rockfall mapping team went underground to their respective stopes with the morning shift mining crews during re-entry. The re-entry procedures to make the working panels safe were done by the mining crews in the presence of the rockfall mapping teams. During this process, rockfalls would be identified. A rockfall is generally identified by the mould it leaves in the hangingwall when it falls out. In some instances rockfalls can be identified as they occur, i.e. when barring out loose blocks to make a working area safe. All the rockfalls were recorded as to whether they occurred with the blast or were as a result of barring. For identification purposes, paint was used to mark the boundaries of the mapped rockfall mould. When inspecting the same stope after a subsequent blast, a check was made for new rockfalls and any further increase in the size of a previously mapped rockfall mould by inspecting the paint marks on its boundary. Thorough watering down of the hangingwall was done to make the paint visible for ease in the identification of rockfalls. The mapping team remained in the stope for the entire shift to check for any rockfalls that took place during the shift. In collecting rockfall data, a measuring tape was tied to the stope hangingwall and parallel to the stope face in-order to record the position of the rockfall from the gully. The properties describing each rockfall were recorded and are listed below; When and how the rockfall occurred i.e. with the blast or by barring, Distance of the rockfall from the gully, Distance of rockfall from the face, Whether it had failed any support units or not, Rockfall shape, this will assist in a more accurate volume approximation, The average dimensions of the rockfall. The following convention was adopted in recording this data: - Rockfall length was taken to coincide with the dimension almost parallel or parallel to face - Rockfall widths was taken in a direction perpendicular to face - Rockfall height was taken as an average of maximum and minimum height for each rockfall, Take note of the release surfaces that caused the rockfall to slide out, i.e. whether it is a joint or blast fracture. Take note of any off-line drilling in the hangingwall, 82

83 Record the properties of all the rockfall release surfaces, i.e. orientation, joint roughness condition and joint alteration, Record any subsequent increase in the size of the rockfall, All this information was recorded on log sheets shown in Appendices B.2 and C.2. identification and mapping of rockfalls on the stope hangingwall was a repetitive process. The Results The data acquired from the rockfall mapping program has been processed in order to get a better understanding of the rockfalls. The statistics of rockfalls recorded underground and their respective time of failure are listed in Table 4-6. The rockfalls have been categorised as either failure during the blast, by barring or the cause of the rockfall being unkown. The time when the rockfalls that occurred during the period when the team took the one week break could not be ascertained with confidence. The effectiveness of barring was then determined as a percentage of the total rockfalls mapped. Most of the recorded rockfalls occurred during the blast. Table 4-6: Rockfall mapping data Reef Blasting Barring Unknown Total % Falls due to blast Merensky UG Total An analysis of the joints forming rockfall release surfaces has been done. The first part involved processing orientations of the joints that formed the release surfaces of rockfalls using the software program Dips. In grouping the rockfall release surface joint poles into sets, a similar approach to that used for the joints in section was adopted. The orientation pole concentrations of the joints that were part of the release surfaces of the rockfalls for UG2 and Merensky Reefs are shown below in Figure 4-15 and Figure 4-16 respectively. It is worth noting that the pole concentrations identified from joint mapping in section (Figure 4-13 and Figure 4-14) are the same as those found in rockfall mapping. 83

84 SIMRAC: SIM Review and final Report N Set 4w 4m R2 1 [44] Reef Plane 2 [53] 3 [12] 4 [3] [no data] [10] W E J2 2w 2m R13m 3w J1 Equal Angle Lower Hemisphere 122 Poles 122 Entries 1w 1m S Figure 4-15: Pole plot of mapped rockfall release surface in UG2 stope N Set 1 [95] Reef Plane 2 [126] 3 [93] 4 [17] [no data] [7 R W 4w 4m 3w 3m J1 J3 E J2 Equal Angle Lower Hemisphere 404 Poles 404 Entries 2w 2m 1w 1m S Figure 4-16: Pole plot of mapped rockfall release surface in Merensky stope The predominant joints defining the rockfall release surfaces are listed in Table 4-7 below. Two steep joints J1 and J3 combined with the shallow dipping random joint, R are the predominant joints delineating the rockfalls in the Merensky Reef stopes. A similar trend can also be observed for the UG2 rockfalls, the rockfall release surfaces here are the steep J1 and J2 joints 84

85 in combination with the shallow random joint R delineate the rockfalls. In both reefs the shallow joints are critical for delineating and providing a release surface for keyblocks. Table 4-7: Joints making blocks and significant joint sets Number of Rockfalls Joints per Reef Block Significant sets J1 J2 J3 R F Max Min Merensky J1,J3 &R UG J1,J2 &R The total number of joints that resulted in the formation of the failed keyblocks has been listed in Table 4-7. It is interesting to note that there are some rockfalls that are bounded by two joints. This type of wedge forms when two sub-vertical joints opposing each other intersected on an undulating hangingwall. In the Merensky and UG2 Reefs, a maximum of four and three joints respectively formed the surfaces of a loose keyblock. These observations can be related to observations made by Hatzor (1992) when analysing rockfalls in a tunnel where 97% of the 69 rockfalls were bound by three joints only. Tharp (1985) and Goodman (1995) also agree that an average of three joint sets in general define a keyblock. These relative numbers are reflective of the degree of jointing of the respective reef hangingwalls, with the Merensky Reef hangingwall being more jointed than the UG2 Reef. Some rockfalls however, had fractures forming some of the release surfaces of the rockfall wedges (Figure 4-17 and Figure 4-18). It could not be established with confidence whether these fractures were formed as a result of blast or stress effects on the rock. However, fractures formed keyblock release surfaces particularly for the Merensky Reef where there was a significant number of low angle blast fractures recorded. 85

86 Figure 4-17: Pole plot of mapped rockfall fracture release surfaces in UG2 stope Figure 4-18: Pole plot of mapped rockfall fracture release surfaces in Merensky stope The shapes of the rockfall moulds have been used in determining the base area and volume of the rockfalls. The typical shapes of the recorded rockfalls are wedges, pyramids and tetrahedrons. These shapes could be truncated in some instances. Consideration of the rockfall shape is important for an accurate approximation of the keyblock base area and volume. 86

87 The rockfall volume, area and height from underground mapping is shown in Figure 4-19, Figure 4-20 and Figure 4-21 respectively. Each reef is represented by two curves, one for all the rockfalls and the other for rockfalls that had no fractures. This is important for a comparative analysis with JBlock, since JBlock reports rockfalls created by joints only. The rockfalls have been normalised per one thousand square metres mined. It is noteworthy that about a third of the rockfalls on both reefs had at least one fracture as a release surface Rockfalls/1000m Merensky-All Merensky-No fractures UG2-All UG2-No Fractures Volume (m 3 ) Figure 4-19: Frequency distribution of rockfall volume 87

88 Rockfalls/1000m Merensky-All Merensky-No Fratures UG2-All UG2-No Fractures Area (m 2 ) 120 Figure 4-20: Frequency distribution of rockfall area Frequency Distribution of Rock Fall Height 100 Rockfalls/1000m Merensky-All Merensky-No fractures UG2-All UG2-No Fractures Height (m) Figure 4-21: Frequency distribution of rockfall height It is worth noting that most of the rockfalls identified during the mapping exercise for both reefs are relatively small (<1.4 m 2 ). No subsequent rockfalls were mapped in the back areas during this period. It is suspected though, that the majority of the large rockfalls would occur in the back areas. The Merensky Reef registered more rockfalls per thousand square metres mined than the UG2 Reef. This corresponds with the observed relatively higher degree of jointing for the Merensky Reef than the UG2 Reef. 88

89 The locations of recorded rockfalls are shown in the density contour plots in Figure 4-22 and Figure 4-23 for the UG2 and Merensky Reefs respectively. The rockfalls occurred at an average distance of 1.30 m and 1.69 m behind the face for the UG2 and Merensky Reefs respectively. The results here signify that, the UG2 rockfalls occurred closer to the face than the Merensky rockfalls. For comparative purpose the 4 m boundary for the recorded rockfalls corresponds to the face area defined in JBlock. Face position Distance (m) behind face Gully position Legend Barring Blasting Distance (m) from gully Figure 4-22: Rockfall location contour plot for UG2 Reef Face position Distance (m) behind face Gully position Legend Barring Blasting Distance (m) from gully Figure 4-23: Rockfall location contour plot for Merensky Reef The time dedicated for recording rockfalls underground and the small dataset of rockfall has been deemed to be inadequate to do a conclusive rockfall correlation exercise. A much larger 89

90 data set including all Ground Control Districts from several mines around the Bushveld Complex is required to better understand rockfalls. 4.5 Rockfall and Injury Data provided by Mine A A larger set of rockfall data and a record of rockfall injuries were provided by Mine A and this was used to assist with the calibration Rockfall Data A large database of rockfalls has been accumulated at Mine A. Strata control observers at the mine record rockfalls that occur in their respective working areas on a monthly cycle. In addition to this rockfalls are recorded, by exception, when an injury or a major rockfall incident occurs. In these records there are over recorded rockfalls since the year 1969, but more diligently since However, most of the records were measured from brows and only the height was recorded. There are approximately complete records of rockfalls. In addition to the dimensions, each record includes information regarding the location, reef and whether it is stoping or development. There is other information, but this is not always consistently recorded. In analysing these rockfalls, only the rockfalls that fell in the stopes have been considered since stope rockfalls are the focus of this research. The number of large (>15 m 2 ) rockfalls was determined from the database to be per 1000 m 2. The area of these rockfalls is approximately 0.42% of the area mined. These values were used for comparison with the simulation results (section 4.6.2). It should be noted that while the data has been diligently collected and the database is very large, it clearly does not contain all the rockfalls that have occurred. It is expected that the larger rockfalls will be more consistently recorded and a clear bias toward large rockfalls was evident in the database. However, it was not possible to establish a level of confidence in the numbers of rockfalls of any size recorded in the database. Figure 4-24 presents the profile of cumulative rockfall frequency versus the distance behind the working face at which the rockfall occurred. The blue curve represents the historical mine rockfalls whilst the red curve represents the small rockfalls recorded during the course of this research. The graph shows that 32% of the historical rockfalls fell on the face and up to 70% occurred within the face working area. While the rockfall data may not be complete, it is believed that this sample will generally be representative of the distance from the face for all rockfalls. However, it should be noted that because the historical rockfalls are recorded on a monthly cycle, there is a possibility that the data was captured a few days after the occurrence 90

91 of the rockfall and the face may have advanced. Therefore it is anticipated that the actual rockfall release profile will have more rockfalls closer to the face. Figure 4-25 shows the percentage of rockfalls per size category versus distance category behind the face. This shows that the trend is consistent for rockfalls of different sizes. 100% 90% 80% 70% Frequency 60% 50% 40% 30% Legend Mine Rockfalls Mapping 20% 10% 0% Distance from face (m) Figure 4-24: Cumulative rockfall frequency vs distance from face 91

92 60% 50% 40% 30% 20% Area (m 2 ) 0 to 1 1 to 2 2 to 4 4 to 20 >20 10% 0% 0 to 2 2 to 4 4 to 8 8 to 15 >15 Distance from face (m) Figure 4-25: Percentage of rockfalls per size category vs distance categories behind the face (mine rockfall data only) Injury Data An analysis of rockfall related injuries at a number of shafts over an 8 year period is presented in Table 4-1. The proportions of the different injury categories have been calculated whilst including and excluding the non-lost time injuries. For the purposes of calculating the cost of injuries, the non- lost time injuries have been excluded as these have no significant financial impact. 92

93 Table 4-8: Rockfall related injury and fatality ratios Year NLTI LT Reportable Fatality Stope production per annum (m 2 ) Total Injuries / m Total Ratio 48% 29% 22% 1% Ratio 55% 43% 2% Average 8.7 Considering only the lost time injuries, reportable injuries and fatalities it was established that injuries and fatalities occurred at an average of 8.7 times/ m 2 mined. This average is important when correlating injuries calculated with JBlock simulated rockfalls. 4.6 Calibration and validation of JBlock simulated rockfall data This chapter describes an exercise to correlate simulated rockfall data to underground rockfall data. The correlation exercise has been split into two, firstly to correlate the rockfall size distribution and then secondly to check that the number of large rockfalls and rockfall injuries, generated through this process, are representative of the actual numbers recorded on the operation. Due to the relatively short duration of underground rockfall mapping, the mapped rockfalls are all relatively small (section 4.4). This was the only data available for reliably correlating the rockfall size distribution. The mine s rockfall database and rockfall injury data (section 4.5) was used for the second part of the analysis. A sensitivity analysis was also carried out to investigate the influence of various input parameters on the results. 93

94 4.6.1 Calibration based on underground rockfall mapping The objective of this investigation was to determine whether JBlock reliably simulates the actual rockfall size distributions that occur underground. A JBlock analysis was done to simulate a set of rockfalls for comparison with mapped rockfall data. This research resulted in a significant upgrade of JBlock. The output from JBlock is in a spreadsheet format with all the information on failed and stable keyblocks together with their respective properties. Further processing of the rockfall information can be done on these outputs to enhance the understanding of results. The joint set data from underground mapping presented in section was used to statistically simulate keyblocks in JBlock for both the UG2 and Merensky Reefs. The excavation geometry used in JBlock is a representation of the stopes in which mapping was done and is shown in section Using the excavation hangingwall and the joint properties, JBlock creates keyblocks, as described in section 3.2.1, while keeping track of the simulated area, which represents the hangingwall are exposed or are mined. JBlock simulation In creating blocks for both the UG2 and the Merensky Reefs a minimum volume of 0.001m 3 has been specified so as to generate blocks of similar size to those mapped underground. JBlock created keyblocks for the UG2 Reef, which required a simulation area of m 2 (0.82 keyblocks / m 2, 82% of area comprises keyblocks). The block size frequency distribution for UG2 is shown in Figure For the Merensky Reef, keyblocks were generated, which required a simulation area of m 2 (1.6 keyblocks / m 2, 90% of area comprises keyblocks). The block size frequency distribution for Merensky is shown in Figure This simulation area was considered adequate for this analysis, since there were no very large rockfalls measured underground and the distribution of rockfall size is more than adequate for this calibration. There are also many more small rockfalls (<0.08 m 2 ) generated with the Merensky data than the UG2 data. The Merensky distribution is lognormal while the UG2 distribution is normal. This is all due to the closer joint spacing in Merensky hangingwall. 94

95 Figure 4-26: UG2 Reef small rockfall keyblock distribution Figure 4-27: Merensky Reef small rockfall distribution 95

96 The keyblocks generated have been tested in the supported excavation to evaluate the effectiveness of the support system in holding the keyblocks. The support properties and spacings used came from the support standards at the mine and are presented in section In reality there is a horizontal clamping stress in the hangingwall which also acts to stabilize keyblocks in the hangingwall. The clamping stresses were varied in the analyses to calibrate the JBlock simulated rockfalls to the underground mapping rockfalls. Initially, no clamping stress was applied when testing the keyblocks for stability. Then increasing clamping stresses were applied iteratively to the JBlock keyblocks until a trend comparable to the mapped rockfalls was produced. It has been illustrated in section that most of the rockfalls recorded underground were within 4 m from the working face. Therefore, for comparison purposes the JBlock results presented here have been screened to look at rockfalls that occurred in the face area only, i.e an area within 4 m from the working face. For comparative purposes, the rockfalls mapped underground have been plotted on the same graphs as the JBlock rockfall results. The graphs for UG2 Reef rockfall volume and area are presented in Figure 4-28 and Figure 4-29 respectively. Similarly, graphs for Merensky Reef rockfall volume and area are presented in Figure 4-30 and Figure 4-31 respectively. In each of these graphs the underground mapping results are represented by the solid green and red curves. The red curve represents all mapped rockfalls inclusive of fracture bound rockfalls whilst the green curve represents the mapped rockfalls that were bounded by joints only. This distinction is important when comparing with JBlock since, the JBlock rockfalls are bound only by joints. 96

97 Figure 4-28: UG2 Reef mapped rockfall volume and JBlock distributions for clamping stresses from 0 kpa to 20 kpa Figure 4-29: UG2 Reef mapped rockfall area and JBlock distributions for clamping stresses from 0 kpa to 20 kpa. 97

98 Figure 4-30: Merensky Reef mapped rockfall volume and JBlock distributions for clamping stresses from 0 kpa to 20 kpa. Figure 4-31: Merensky Reef mapped rockfall area and JBlock distributions for clamping stresses from 0 kpa to 20 kpa. The curved for the mapped rockfall data for the UG2 crosses several clamping stress curves as the size increases. This trend indicates that JBlock needs to be modified to allow increasing clamping stresses for increasing rockfall size or in fact rockfall height. There was a better visual 98

99 correlation for Merensky at low clamping stresses for rockfalls between m 3 and 0.03 m 3, but for larger rockfalls a similar trend is observed. Simulation Accuracy When comparing how one data set fits or compares to another, a measurement of accuracy is one good way of comparing the agreement between the two. There are five main types of accuracy measures to be discussed here of which only one will be chosen to compare the data sets. In many instances, the word accuracy refers to goodness of fit, which in turn refers to how well a simulation model is able to reproduce the data that are already known. A single observed (actual) value is represented by. The individual simulated (JBlock) values are denoted by and the error by, where the error is the difference between the actual value and the simulated value for observation t: The first error measure is the mean error (ME). The ME is the sum of all the individual errors averaged. The mean absolute deviation (MAD) is based on the mean error (ME) but is defined by first making each error positive by taking its absolute value, and then averaging the result. The MAD is the preferred choice of the aforementioned errors as the ME is likely to be small since positive and negative errors tend to offset one another. The idea is the same for the root mean squared error (RMSE). The errors are made positive by squaring each one, then the sum of the squared errors are averaged. The standard statistical measures can be defined as: Each of the aforementioned statistics deals with measures of accuracy whose size depends on the scale of the data and therefore do not facilitate comparison across data sets. In order to make comparisons like these, one needs to work with relative percentage error measures. The relative percentage error (PE) is defined as: 99

100 The relative percentage error can be used to compute the percentage error for any number of observations. The percentage error can then be averaged to give the mean percentage error (MPE). As with the ME, the MPE is likely to be small since positive and negative PEs tend to offset one another. Hence the MAPE is defined using absolute values of PE. The statistical measures for the MPE and MAPE can be defined as: The above error measures are all calculated in a very similar way, where denotes the exact (observed) values and denotes the calculated (simulated) values. The MAPE is the preferred method of preference as the size of the error does not depend on the scale of the data as well as the fact that it allows for positive and negative errors not offsetting another. The relationship between the actual rockfalls ( ) and the simulated rockfalls ( ) has been evaluated for volume and area properties. The closer the simulated and actual mapping data are to another, the smaller the error and the better the rockfall simulation exercise. In order to compare the data sets, the following need to be noted: The number of points representing the actual mapped data is two orders of magnitude less than that simulated by JBlock, The error measure should be applied only to the range of values which corresponds to the minimum and maximum values of the actual mapped data (the mapped data provides these parameters as JBlock has a much wider range). In order to compare the data sets, one need to transform the simulated data sets to have the same number of entries and the same range as the actual mapped data sets, while still accurately representing the original simulated data set: This is done by taking each x value (area or volume) from the actual mapped data set and determining an equivalent interpolated y value for the JBlock set. The x-values for both the actual mapped data and the simplified JBlock set are the same. 100

101 Using the simplified JBlock set and the actual mapped data, the different measures of accuracy were calculated for the area and volume properties. The results are presented in Table 4-9 and Table 4-10 for area and volume respectively. These results should be compared with Figure 4-28 to Figure Table 4-9: Measures of accuracy for Area UG2 (Area) Merensky (Area) Clamping Stress (kpa) MAD RMSE MAPE MAD RMSE MAPE % % % % % % % % % % % % % % % % Table 4-10: Measures of accuracy for Volume Excluding Blast Fractures UG2 (Volume) Merensky (Volume) Clamping Stress (kpa) MAD RMSE MAPE MAD RMSE MAPE % % % % % % % % % % % % % % % % Consider the MAPEs of both reefs for rockfall area and rockfall volume in Table 4-9 and 4-10, respectively. JBlock simulations with no clamping stress is the worst representation of the actual data for both the Merensky and UG2 reefs in terms of both rockfall area and rockfall volume. It is interesting to note that the JBlock simulations with a clamping stress of 1kPa for UG2 and Merensky rockfall area (276.47% and %) performs fairly poor compared to similar simulations produced for UG2 and Merensky rockfall volume (53.03% and 32.18%). Discussion The mapped rockfall data set includes very small rockfalls, but is limited to maximum rockfall size of 0.46 m 3. It is clear that by applying some low value of clamping stress to this range of block sizes greatly improves the correlation between mapped and simulated results. However, it is quite apparent that the JBlock program needs to be modified further to improve the 101

102 correlation. The actual clamping stress in the hangingwall increases with increasing height above the stope and the results show that a low clamping stress is appropriate for small rockfalls, while a higher clamping stress is appropriate for larger rockfalls. A longer and more extensive rockfall data collection exercise will be required to collate the full range of rockfall sizes Calibration based on Mine A rockfall database and rockfall injury data A validation of the risk evaluation model, which included the RiskEval program, was undertaken in-order to measure its performance against observations at the mines. The objective of this exercise is to determine whether the simulated results correlate with the actual number of rockfall injuries and large rockfalls recorded at the mines. Rock blocks for this case study have been generated in JBlock v3028 using the UG2 and Merensky Reef geometry and joint properties described in Sections and respectively. In creating the rock blocks, JBlock was set to create a minimum block volume of 0.01m 3 in order to create a balanced set of blocks with both the large and small block sizes included (see section 3.2.1). The distribution of keyblock sizes generated in JBlock for the UG2 and Merensky Reefs are shown in Figure 4-32 and Figure It should be noted that these distributions are different from Figure 4-28 and Figure in section 4.6.1, since the minimum keyblock size was set 0.01 m 3 and more keyblocks were generated. It is still apparent that more small keyblocks are generated using the Merensky joint data than when using the UG2 joint data. For the UG2, keyblocks were created, which required a simulation area of m 2 (0.32 keyblocks / m 2, 80% of the simulated area comprises keyblocks). For the Merensky, blocks were generated, which required a simulation area of m 2 (0.91 keyblocks / m 2, 87% of the simulated area comprises keyblocks). 102

103 Figure 4-32: Keyblock size distribution for UG2 Reef at mine A Figure 4-33: Keyblock size distribution for Merensky Reef at mine A 103

104 Rockfalls were simulated using these block sets and the UG2 and Merensky support standards (section and 4.2.5). The expected frequency of injuries as a result of these rockfalls was simulated using RiskEval (section 3.3). As in section 4.6.1, the clamping stress was varied from 0 kpa to 20 kpa. The results are presented in Table 4-11 and Table Note that dilution is reported as the average thickness of dilution (dilution volume / simulated area). Table 4-11: Rockfalls and Injuries for UG2 Reef at mine A with varying clamping stress Clamping Stress Large Rockfalls /1000m 2 % Area All Rockfalls % Area Large Rockfall Dilution (m) Injuries / m 2 0kPa kPa kPa kPa kPa kPa kPa Table 4-12: Rockfalls and Injuries for Merensky Reef at mine A with varying clamping Clamping Stress Large Rockfalls /1000m 2 % Area All Rockfalls stress % Area Large Rockfall Dilution (m) Injuries / m 2 0kPa kPa kPa kPa kPa kPa kPa The number and percentage area of large rockfalls determined from the mine A rockfall database was rockfalls per 1000 m 2 (section 4.5.1), which is in the same order as the 10 kpa clamping stress analysis for the UG2. The Merensky 10 kpa analysis shows a much higher frequency of large rockfalls. However, the injuries reported in this table are in the same 104

105 range as the historical injuries reported in The rockfall injury data is considered more reliable than the large rockfall data. For the purposes of this project, it was therefore decided to use a clamping stress of 10 kpa based on the Merensky Reef injury results for the case studies in Chapter 5. However, this again highlights the requirement to use an increasing clamping stress with increasing height above the stope to improve the correlation for the full range of rockfalls Sensitivity Analysis A sensitivity analysis was conducted to investigate the influence of certain critical parameters on the results of the simulation. The sensitivities considered here are changes in clamping stress, profit margin, support quality, flexibility in mining and barring efficiency. Both JBlock and RiskEval were used for this analysis (section 3.2 and 3.3). All the sensitivities have been done on the Merensky Reef at Mine A using the standard stope support system. Detailed information on the parameters is listed in Appendix C. Clamping Stress Data from Table 4-12 in section was used for this analysis. The effect of clamping stress on the expected losses and expected frequency of injuries is shown in Table This shows that it is important to use a realistic clamping stress to ensure that the results do not bias the comparisons of different support systems. Table 4-13: Clamping Stress Sensitivity Clamping stress Support Cost Dilution Sweepings Cost/m 2 Expected losses due to rockfalls Resupport Production Injury Total Total Cost Injuries per m 2 0kPa R59.00 R76.28 R8.87 R6.06 R82.06 R70.04 R R kPa R59.00 R69.12 R8.08 R5.40 R72.79 R64.15 R R kPa R59.00 R66.50 R8.25 R5.14 R78.63 R61.41 R R kPa R59.00 R59.25 R8.02 R4.62 R74.05 R56.31 R R kPa R59.00 R47.58 R7.30 R3.76 R73.02 R46.16 R R kPa (Base R59.00 R24.33 R5.11 R1.95 R49.96 R25.49 R R Case) 20kPa R59.00 R5.22 R2.66 R0.55 R31.31 R9.38 R49.12 R

106 Profit Margins The effect of changing the profit margin on the results is shown here. A profit margin is the profit realised after processing a tonne of ore. The base case used was the support standard for the Merensky Reef at Mine A. The profit margin used in the base case is based on the profit margins listed in Appendix C.2 determined for this the mine. The profit margin has been varied to 20% higher and lower than the determined value. The results of this sensitivity are listed in Table Table 4-14: Profit Margin Sensitivity Profit Margin Support Cost Dilution Sweepings Cost/m 2 Expected losses due to rockfalls Resupport Production Injury Total Total Cost Injuries per m 2 Base Case R59.00 R24.33 R5.11 R1.95 R49.96 R25.49 R R % Higher R59.00 R24.33 R6.13 R1.95 R59.96 R30.42 R R % Lower R59.00 R24.33 R4.09 R1.95 R39.97 R20.56 R90.90 R It can be deduced that a higher profit margin results in higher losses and vice versa for a lower profit. It shows that on operations where the profit margin is high, it makes sense to use more effective support from an economic perspective. It also illustrates the importance of an accurate calculation of the profit margin per tonne. Quality of support installation The consequences of deviation from standard support spacing has been analysed. The base case is the mine standard with no deviation or error in the support spacing. Then the support spacing is varied on both dip and strike, while keeping the mean spacing to standard. The mean support spacing is then increased by the same amount as the error. Finally, an unsupported stope is also investigated to show the reduction in rockfalls offered by support. The effect on the frequency of rockfalls and associated costs experienced with each of these variations in the support system installation are listed in Table 4-15 and Table 4-16 respectively. Note that dilution is reported as the average thickness of dilution (dilution volume / simulated area). 106

107 Support Spacing Mean Spacing Table 4-15: Effect of quality of support installation on rockfall frequency Error (m) Large Rockfalls % Area All /1000m 2 Rockfalls % Area Large Rockfalls Dilution (m) Mine Standard (Base case) Mine Standard ± Mine Standard ± Mine Standard + 0.2m ± Mine Standard + 0.5m ± No support Support Spacing Mean Spacing Standard (Base Case) Table 4-16: Effect of quality of support installation on costs Error (m) Support Cost Dilution Sweepings Cost/m 2 Expected losses due to rockfalls Resupport Production Injury Total Total Cost Injuries per m R59.00 R24.33 R5.11 R1.95 R49.96 R25.49 R R Standard ±0.2 R59.00 R26.28 R6.35 R2.04 R57.70 R29.24 R R Standard ±0.5 R59.00 R26.46 R5.86 R2.04 R55.97 R28.13 R R Standard+0.2m ±0.2 R45.00 R30.42 R7.78 R1.88 R72.70 R34.24 R R Standard+0.5m ±0.5 R32.00 R39.17 R8.34 R1.74 R71.28 R40.26 R R No Support R0.00 R88.99 R15.72 R0.00 R R36.35* R R * *The model for determining injuries is not appropriate for unsupported stopes. It is interesting to note that increasing the variability, while maintaining the average support spacing results slightly more injuries and losses. As expected, increasing the support spacing does have a significant effect on the expected losses and injuries. If no support was installed many more rockfalls would occur and the expected losses would be far greater than the cost of support. However, the current model for determining injuries is not appropriate for unsupported stopes (section 3.3). The exposure reduction for falls which occur after the blast and rocks that are barred is applied to rockfalls in between support and not to the larger rockfalls that fail support. Therefore this reduction is applied to all the rockfalls in the unsupported case, including the larger rockfalls and results in a significant underestimation of the expected injuries. This does not affect the expected losses, with the exception of the cost of injuries, which would be much greater. 107

108 Flexibility The effect of increasing flexibility or increasing the available face length is investigated here. Mining flexibility is the availability of a spare stope to which labour can be redirected in the event that there is a large rockfall and the stope panel needs to be re-established. This is possible in the case where there are dedicated stope re-establishment crews. Table 4-17 lists the expected losses associated with varying mining flexibility. It can be concluded that there is a reduction of production related losses as flexibility increases. This analysis makes a very strong case for creating flexibility and employing these rehabilitation crews. However, the additional cost of creating this flexibility and maintaining these crews should be determined, for a comprehensive risk evaluation. This detail has not been done as part of this project. Table 4-17: Costs associated with varying mining flexibility Flexibility 0% (Base Case) Support Cost Dilution Sweepings Cost/m2 Expected losses due to rockfalls Resupport Production Injury Total Total Cost Injuries per m 2 R R R 5.11 R 1.95 R R R R % R R R 5.11 R 1.95 R R R R % R R R 5.11 R 1.95 R R R R % R R R 5.11 R 1.95 R 0.00 R R R Effectiveness of barring Barring of loose rocks in the stope hangingwall is an important part in making an underground working area safe. The effect of improving barring on the number of injuries and associated costs is investigated here. The base case of 90% effective barring is based on underground mapping results (Section 4.4.2). The injuries and costs for different levels of barring effectiveness are listed in Table It can be observed that as the effectiveness of barring reduces, so the expected injuries and costs increase. The more than 40% increase in injuries when barring is only 50% effective, highlights the importance of carrying out the making safe procedures properly. 108

109 Table 4-18: Costs and injuries associated with the effectiveness of barring Barring Effectiveness 90% Base Case Support Cost Cost/m 2 Expected losses due to rockfalls Dilution Sweepings Re-support Production Injury Total Total Cost Injuries per m 2 R59.00 R24.33 R5.11 R1.95 R49.96 R25.49 R R % R59.00 R24.33 R5.11 R1.95 R49.96 R31.19 R R % R59.00 R24.33 R5.11 R1.95 R49.96 R36.88 R R % R59.00 R24.33 R5.11 R1.95 R49.96 R58.72 R R

110 5 Risk Evaluation Case Studies Three Risk Evaluation case studies are described in this chapter that were selected to demonstrate the benefit of using the risk evaluation approach to quantify and compare the cost and safety impacts of the different stope hangingwall support scenarios. The case studies are based on mine A, UG2 and Merensky (see Chapter 4), and a Merensky ground control district at mine B, another platinum mine in the Bushveld Complex. 5.1 Case Study1 (UG2 Reef at Mine A) This case study is based on data collected on the UG2 Reef from Mine A. The site selection and data collection process for this case study has been described in sections 4.2 and 4.3 respectively Scenario Descriptions The base case for the scenarios considered here is the support system that is currently being used for the UG2 at this mine. The current support system at mine A comprises mine poles only in the stope and 1.2m Hydrabolts in the gully. A number of support scenarios which are mainly modifications to the current support system were considered to compare the risk associated with rockfalls in each of the scenarios. The support scenarios considered are listed in detail below; 1. Mine poles only, this is the base case for comparative purposes. The support system layout for this scenario, which is the mine standard, is presented in Figure 4-9. This support layout is simulated in JBlock and the associated support properties are presented in Appendix B m Hydrabolts and mine poles. This support system is an enhancement of the base case support by introducing 10T, 1.2 m Hydrabolts into the stope support. Any potential safety and profitability associated advantages as a result of this enhancement will be quantified m Hydrabolts and mine poles. In this support system, the effect of increasing the Hydrabolt length from the initial standard 1.2 m to 1.5 m is investigated and quantified m Hydrabolts with mine poles. The purpose of testing this support system is to establish the cost and safety benefit that can be derived from having an even longer Hydrabolt length than the two lengths previously considered. 110

111 m end anchored cable bolts with mine poles. This support system is an enhancement of the standard support by replacing the hydrabolts with 25T, 1.5 m cable bolts in the stope support. The cable bolt support is stronger than the Hydrabolt support. Any potential safety and profit associated advantages as a result of introducing the high strength support system will be investigated and quantified m cable bolts with mine poles. Potential safety and profit benefits associated with increasing the cable bolt length will be quantified in this support system. The support scenarios described and their properties are presented in detail in Appendix B.3. In choosing these support scenarios consideration for the feasibility of installing the various support system without a significant change in mining cycles and productivity in the stopes from the original support scenario has been made. The support elements chosen are those readily available on the market to the Bushveld Complex underground mines Financial and Injury Data The parameters used to quantify financial and injury risk associated with different support strategies in this case study have been quantified and presented in Appendix B for the UG2 Reef at mine A. Production and sweepings losses are calculated using the method described in section The production losses are based on long term price forecasts of platinum group metals (PGM) by five major banks presented by BMO Capital Markets (BMO, 2011). The PGM grade and metal ratios of the constituent elements are based on figures provided in annual results of the owners of mine A and are consistent with observations done by Cawthorn (1999) and Cawthorn et al (2002) in the surrounding areas of the mine. Total mining costs have been obtained from annual results of the owners of mine A. When calculating production losses, provision for an blasting efficiency of 70% during re-establishing has been made. The costs associated with injuries have been determined as described in section The mineral processing parameters have been adapted from SRK project experience for UG2 in the Bushveld Complex. The costs of different lengths of cable anchors and Hydrabolts were obtained through personal communication with the suppliers (M and J Mining Supplies, 2011 and New Concept Mining, 2011). Costs for mine poles were obtained from data at the mine. 111

112 The time and costs associated with drilling longer holes for longer support units has not been factored in the analysis due to unavailability of drilling costs from the mine. It is however, anticipated that the longer holes will increase costs associated with the support scenario. It is assumed that mining crews will retain the same complement for the various scenarios and thus labour costs will be constant. The proportional sub-division of accidents into different injury categories presented in Table 4-8 were derived from a study of rockfall related accidents over an 8 year period from a number of mine shafts operated by the owners of mine A. The worker categories and the time spent in each zone is based on the mining strategy at mine A and were established by communication with production personnel at this mine. All these scenarios have a uniform set of RiskEval parameters and strategies used for calculating the losses for the case study. A detailed list of the parameters and strategies for this case study are in Appendix B JBlock Results Rock blocks for this case study were generated (Section 4.6.2) using the UG2 Reef geometry and joint properties described in Sections and respectively. A clamping stress of 10 kpa was adopted in testing this set of keyblocks in all the described support scenarios in this case study. The JBlock rockfall results for case study 1 are presented in Table 5-1. Note that dilution is reported as the average thickness of dilution or increase in stoping width (dilution volume / simulated area). The large rockfalls are those that are large enough to require re-establishment of the panel. The frequency of large rockfalls in the UG2 is significantly lower than that determined for the Merensky (Section 4.6.2). Support System Table 5-1: Summary of rockfalls for case study 1 Large Rockfalls/1000m 2 % Area All Rockfall % Area Large Rockfall Dilution (m) Mine Poles Only m Hydrabolt + Mine Pole m Hydrabolt + Mine Pole m Hydrabolt + Mine Pole m Cable Anchor + Mine Pole m Cable Anchor + Mine Pole

113 5.1.4 Risk Evaluation and Comparison of Results Each single rockfall that occurs for each scenario is evaluated for the risk it is likely to cause. This risk for each support scenario is as described in Chapter 3 and is based on rockfall size and location and the zone in which it falls. The summary for risk evaluation results of this case study are presented in Table 5-2. Table 5-2: Summary of Risk Eval results for case study 1 Support System Support Cost Cost/m 2 Expected losses due to rockfalls Dilution Sweepings Re-support Production Injury Total Total Cost Injuries per m 2 Mine Poles R R 7.23 R 1.10 R 0.08 R 4.28 R 4.27 R R m Hydrabolt + Mine Pole R R 3.12 R 0.59 R 0.23 R 1.96 R 2.10 R 8.00 R m Hydrabolt + Mine Pole R R 2.42 R 0.66 R 0.18 R 1.92 R 1.70 R 6.89 R m Hydrabolt + Mine Pole R R 1.67 R 0.30 R 0.13 R 0.48 R 1.08 R 3.66 R m CableAnchor+ Mine Pole R R 0.90 R 0.02 R 0.08 R 0.34 R 0.61 R 1.97 R m Cable Anchor + Mine Pole R R 0.73 R 0.01 R 0.06 R 0.14 R 0.46 R 1.41 R The introduction of 1.2 m hydrabolts halves the number of injuries and this reduces substantially with each improvement in the support system. The same trend can be seen for all of the losses. The dilution is the largest contributor to the losses, followed by production losses. However the total loss for the Mine Pole support system is low relative to the cost of the other support systems and therefore the Mine Pole support system has the lowest total expected cost. This is due to the relatively low frequency of rockfalls, particularly large rockfalls. In this case the motivation for improving support would be based on safety only. 5.2 Case Study 2 (Merensky Reef at Mine A) This case study has been based on the mapping exercise done on the Merensky Reef at mine A. The site selection and data collection process for this case study has been described in sections 4.2 and 4.3 respectively Scenario Descriptions The base case for the scenarios considered here is the support system that is currently used at this mine. The current support system at mine A comprises a combination of mine poles 1.2 m Hydrabolts in the stope and 1.2 m Hydrabolts in the gully. A number of support scenarios 113

114 derived mainly from the current support system have been considered to compare the risk associated with rockfalls in each of the scenarios. The support scenarios considered are listed below; m Hydrabolts and mine poles. A combination of the 100kN, 1.2 m Hydrabolt and mine pole support system in the stope is currently being used for supporting the Merensky Reef hangingwall at mine A. From the rockfalls experienced with this support scenario, associated potential safety and profit benefits will be quantified m Hydrabolts and mine poles. In this support system, the effect of increasing the Hydrabolt length in the stope and gully from the standard 1.2 m is investigated and quantified m Hydrabolts with mine poles. The purpose of testing this support system is to establish the cost and safety benefit that can be derived from having a longer Hydrabolt length than the two lengths previously considered m end anchored cable bolts with mine poles. This support system is an enhancement of the base case support by introducing 250kN, 1.5 m cable bolts in the stope support. The cable bolt support is stronger than the hydrabolt support. Any potential safety and profitability associated advantages as a result of introducing the high strength support system is investigated and quantified m cable bolts with mine poles. Potential safety and cost benefits associated with increasing the cable bolt length is quantified for this support system m Hydrabolts, mine poles and safety net. This support scenario is a combination of the 100kN, 1.2 m Hydrabolt, mine pole and temporary safety net support (Skarbøvig et. al, 2011) system. The safety and cost benefits of introducing a safety net in the face area are evaluated in this support scenario. The support scenarios described and their properties are presented in detail in Appendix B.3. Consideration for the feasibility of installing the various support systems without a significant change in mining cycles and productivity in the stopes from the original support scenario has been made in choosing these support scenarios. The support elements chosen are those readily available in the market for the Bushveld Complex underground mines. 114

115 5.2.2 Financial and Injury Data The parameters used to quantify financial and injury risk associated with different support strategies in this case study have been quantified and presented in Appendix B.4 for the Merensky Reef at mine A. The parameters are derived as for the previous case. Selection of strategies used to deal with each rockfall has also been presented in Appendix B. It is assumed that mining crews will retain the same complement for the various scenarios and thus labour costs will be constant JBlock Results Rock blocks for this case study were generated (Section 4.6.2) using the Merensky Reef geometry and joint properties described in Sections and respectively. A clamping stress of 10 kpa was used in testing this set of keyblocks in all the described support scenarios for this case study. The JBlock rockfall results for case study 2 are presented in Table 5-3. Note that dilution is reported as the average thickness of dilution (dilution volume / simulated area). Table 5-3: Summary of rockfalls for case study 2 Support Scenario 1.2m Hydrabolt + Mine Pole 1.5m Hydrabolt + Mine Pole 1.8m Hydrabolt + Mine Pole 1.5m Cable Anchor + Mine Pole 2.0m Cable Anchor + Mine Pole 1.2m Hydrabolt + Mine Pole + safety net Large Rockfalls/1000m 2 % Area All Rockfalls % Area Large Rockfall Dilution (m) Note that there are many more rockfalls per area mined than that determined for the UG2 (Table 5-1). The original mine support used in this case study represents the base case and the remaining support systems are potential improvements. Increasing the Hydrabolt length reduces the overall rockfall percentage area and dilution significantly, but the frequency and percentage of large rockfalls remain essentially the same (given the stochastic simulation process). This indicates that the hydrabolt support system does not have a high enough support resistance to prevent the larger rockfalls. Their main purpose is to reduce the frequency of injuries. Introducing stronger and longer cable anchors results in a reduction in 115

116 both the overall and large rockfall frequency. The safety nets show a significant reduction in the percentage area of all rockfalls, but the improvement for large rockfalls is not significant, since the nets are not designed for this purpose. The reduction in dilution will not be realised, since the nets will be removed Risk Evaluation and comparison results The Risk Evaluation process involves an analysis for each single rockfall that takes place in each scenario for the risk it is likely to cause. This risk for each support scenario is based as described in Chapter 3 on rockfall size and location i.e., the zone in which it falls. The summary of risk evaluation results for this case study is presented in Table 5-4. The expected losses for the nets are not included in this table, since in reality the nets will be removed and the damage consequences will remain. Nets can reduce the expected frequency of injuries only. Table 5-4: Summary of Risk Eval results for case study 2 Support System 1.2m Hydrabolt+Mine Pole 1.5m Hydrabolt+Mine Pole 1.8m Hydrabolt+Mine Pole 1.5m Cable Anchor + Mine Pole 2.0m Cable Anchor+Mine Pole 1.2m Hydrabolt + Mine Pole + safety net Support Cost Dilution Sweepings Cost/m 2 Expected losses due to rockfalls Resupport Production Injury Total Total Cost Injuries per m 2 R59.00 R24.33 R5.11 R1.95 R49.96 R25.49 R R R65.00 R21.88 R5.18 R1.71 R55.03 R22.20 R R R71.00 R17.44 R4.90 R1.41 R52.55 R18.79 R95.09 R R94.00 R13.28 R3.08 R1.48 R31.59 R13.98 R63.41 R R98.00 R11.40 R2.23 R1.14 R24.55 R10.57 R49.89 R R The results show that each improvement in the support system results in a reduction in the expected frequency of rockfalls. Safety nets are also shown to reduce the risk of injuries, in this case by 15%. In this case study, the expected losses due to rockfalls are significant and offset the cost of the improved support, so the improved support could be justified on both a safety and economic basis. The losses due to production are the greatest due to the relatively high frequency of large rockfalls. However, it must be noted that the frequency of large rockfalls could not be calibrated with confidence (section 4.6.2) and these losses could be overestimated. The losses due to injuries and dilution are also significant. Sweepings losses and the cost of re-supporting are relatively insignificant. The support cost of nets is relatively insignificant, since they are re- 116

117 used. However, it is likely that the nets will need to be replaced when significant rockfalls occur and the evaluation of these losses is not currently modelled appropriately. 5.3 Case Study 3 (Stope cable anchors at Mine B) Mine B is a platinum mine in the Bushveld Complex. The Merensky Reef at this mine is extremely blocky, as such the case study has been focused entirely on the Merensky Reef. The mine changed its stope support system from its old 120 kn, 1.5 m cable anchors used in combination with mine poles to a new 250 kn, 3 m cable anchor support with no mine poles. The mine wanted to better understand the performance of their current 250 kn, 3 m cable anchor support system in comparison to the old support system and also investigate possible improvements Site Description Joint mapping was done in a total of five stopes in the Merensky Reef hangingwall at mine B. These stopes were at a depth of approximately 674 m below surface with a dip and dip direction of 10⁰ and 045⁰ respectively. The areas covered approximately 500 m on dip and over 3 km of strike length of the main working areas in the Merensky Reef. A list of the areas mapped and the number of joints intersected per each scanline is presented in Table 5-5. Mapping in these panels provided two dimensional data for the joints but in reality joints are three dimensional quantities. In order to get a three dimensional view or impression of joints in the hangingwall of the stope, a hangingwall development was mapped in a strike gully. At this site, a normal fault displaced the reef downwards whilst exposing the reef hangingwall at the same time. Table 5-5: Table of scanlines and number of joints mapped Location Scanline start position position No of joints Scanline Length (m) Joints/m 10/58 1W Peg D m from gully into stope on dip /58 1W D1212 in gully /40E 3W 5m from centre gully into strike gully /38 EA Peg R m on dip /38 EA Peg R8284 on strike /40E 1W Peg D m on dip+3.1m on strike /40 3W Peg D from gully into stope /40 3W Peg D1752 on strike in gully /46 4W Peg D m into stope /46 6W Peg D m from ASG Total In carrying out the joint mapping exercise, two orthogonal scanlines were used, one in-stope parallel to the face and the other in and parallel to the strike gullies. Both scanlines were 117

118 required for each mapping site so as to intersect those joints that may have been parallel or sub-parallel to one of the scanlines. Each and every joint that intersected the scanline was mapped to determine its orientation (dip and dip direction), location, length and the shear strength properties i.e. joint roughness and alteration of the joints. Over a total scanline length of m, 418 joints were mapped and are listed for each respective site in Table 5-5. The rock forming the immediate Merensky hangingwall is pyroxenite. A Breithaupt compass was used in measuring the orientations. Care was taken to measure the orientations as far away as possible from the magnetic influence imposed by metallic objects, in particular roof bolts. An investigation of the magnetic influence of the orebody on the compass was done and it was found that there is no influence on the orientation results Geotechnical Data In analysing the joint orientation data, use was made of an orientation data analysis programme DIPS. A total of 418 joint entries were made into the DIPS programme to obtain the average joint poles (Figure 5-1) for the different sets. Since the orientation measurements were done using a magnetic compass, an adjustement of 15 degrees west was effected to cater for magnetic declination in the area. The processed average dip and dip direction for the joints are listed in Table 5-6. There is a large scatter of poles with the majority of the joints being subvertical as illustrated by the pole plot in Figure 5-1. Though the joints could be clustered into six sets, the four most common joints observed were used in the analysis because not more than four joint sets were identified during mapping of individual areas. 118

119 N Number of Poles 1 pole 2 poles 3 poles 4 poles 5 poles W J3 J1 J2 J4 E Equal Area Lower Hemisphere 418 Poles 418 Entries S Figure 5-1: Dips pole plot for Mine B Merensky joint sets Table 5-6: JBlock joint input parameters for mine B Orientation (⁰) Friction Angle (⁰) Spacing (m) Length (m) Set Dip Dip Direction Mean Stdev Mean Stdev Mean Stdev Min Avg Max Min Avg Max J J J J Joint friction angles were determined using the Barton (2002) method described in section and the distribution of friction angles for each joint set is presented in Appendix D Scenario Descriptions Mine A had been using 1.5 m long 120 kn end anchored cable bolts with in-stope pencil sticks plus breaker lines of grout packs and temporary support (Figure 5-2). It was decided to change this support replacing it with a 3 m cable anchor timber-less system with grout packs at a similar spacing (Figure 5-3). This case study is an evaluation of the risk cost implications associated with this change of support strategy. 119

120 The main difference between these two support layouts is that the former has pencil sticks and the cable anchors are on a spacing of 0.9 m by 1.5 m whilst the latter does not have elongate support and the cable anchors have a spacing of 1 m by 1.5 m. The support layout and properties utilised in JBlock for this case study were adapted from the support layouts presented in Figure 5-2 and Figure 5-3. The stope mining area has been divided into four, i.e. face, sweeping, gully and back areas. The face area has been defined as the area within four metres of the mining face where most of the stoping operations take place. The sweeping area has been defined as the area up to 10 m behind the face area boundary. The rest of the area going backwards is defined as the back area. The gully area is the normal gully used for personnel and material access into the stope and for ore transportation. The strength of the 250 kn cable bolts has been downgraded to 200 kn to cater for poor bolt installations. The cable anchor lengths were varied from 1.5 m to 4 m in 0.5 m increments whilst the spacing for support units was kept the same. All the support scenarios considered for this case study are presented in Table 5-7. Table 5-7: Case study 3 support scenario and strengths Support Scenario Support unit strength 1.5m Cable Anchor Support 200kN 2.0m Cable Anchor Support 200kN 2.5m Cable Anchor Support 200kN 3.0m Cable Anchor Support 200kN 3.5m Cable Anchor Support 200kN 4.0m Cable Anchor Support 200kN 1.5m Cable Anchor + Mine Poles 120kN Cable Anchors and 250kN for Mine poles The stope layout for the support scenarios and the support properties used for this case study are detailed in Appendix D

121 1.5-m Cable Anchors, Grout Packs and 3 x 3 m Pillars 2m 1.5m 1.5 m Max 3m 0.5m 0.5m 2m 1.5m Fault 3m Legend 1.5m Mechanical Prop 1m RSS Grout-pack 0.5m 1.8m 3m 1.5m FCG Cable Anchor Am-strap Not to Scale Min 4m 0.5m Max 20m 1m Safety net 5.4m 1.5m Max 0.9m Max 0.9m 6m 1.5m 1.5m 1m 1m Pencil stick 1. Use Am-straps in areas where hanging wall requires it. 2. No drilling may take place before the temporary support and safety nets have been installed. 3. Cable anchor holes must be drilled to at least 1.6 m to get an effective anchor of 1.5 m. 4. Cable anchors must be installed 0.5 m from the face before the blast, 0.9 m apart on strike and 1.5 m apart on dip. 5. Cable anchor holes must be drilled as close to perpendicular as possible, not less than 70 0 with the hanging wall. 6. Cable anchors must be full column grouted (FCG). 7. Support must be within 0.5 m from both sides of prominent geological structures, concurrent with existing rows. 1.5m Figure 5-2: 1.5 m Cable Anchor support system 121

122 3-m Cable Anchors, Grout Packs and 3 x 3 m Pillars 2m 1.5m 1.5 m Max 3m 0.5m 0.5m 2m Fault 3m Legend 1.5m Mechanical Prop RSS Grout-pack Max 1.8m 3.2m 3.0m FCG Cable Anchor Am-strap Not to Scale 0.5m Min 4m 6m Max 10m Safety net 5m 1.5m New Line Max 1m Max 1m 1.5m 1.5m 1m 1m 1. Use Am-straps in areas where hanging wall requires it. 2. No drilling may take place before the temporary support and safety nets have been installed. 3. Cable anchor holes must be drilled to at least 3.1 m to get an effective anchor of 3 m. 4. Cable anchor holes must be drilled as close to perpendicular as possible, not less than 70 0 with the hanging wall. 5. Cable anchors must be full column grouted (FCG). 6. Support must be within 0.5 m from both sides of prominent geological structures, concurrent with existing rows. ASG Centre Line 1.5m Max 1.8m Financial and Injury Data Figure 5-3: 3 m Cable Anchor support system The parameters used to quantify financial and injury risk associated with different support strategies in this case study have been quantified and presented in Appendix D.4 for the Merensky Reef at mine B. The parameters are derived as for the previous cases. Selection of strategies used to deal with each rockfall has also been presented in Appendix D.4. The mining crew complement remained the same when changing from the old support system to the new support system and thus labour costs can be assumed to be constant. However, if the cable lengths are increased, the labour cost may also increase, but this has not been evaluated. 122

123 5.3.5 JBlock Results The Merensky Reef geometry and joint properties are as described in sections and respectively. In creating the blocks, JBlock was set to create blocks with a minimum volume of 0.01m 3, in-order to create a balanced set of blocks with large and small blocks (section 3.2.1). For this analysis, keyblocks were created, which required a simulation area of m 2 (0.72 keyblocks / m 2, 80% of the simulated area comprises keyblocks). The distribution of keyblock sizes generated in JBlock for the Merensky Reef is shown in Figure 5-4. Figure 5-4: Keyblock size distribution for Merensky Reef at mine B A clamping stress of 10 kpa has been adopted in testing all the described support scenarios in this case study. A summary of the JBlock results, in the face area, for case study 3 are presented in Table