Performance Characteristics of Deterministic Leak Detection Systems Michael Kasch 3S Consult München GmbH Germany Abstract The aim of leak detection and leak location is the limitation of losses and possible environmental damage in case of a leak, especially with hazardous products. The recognition of a leak, the knowledge of its leak rate and its location should lead to preassigned action taken by the operating personnel. A basic requirement for that is a reliable leak detection system that provides a well-defined detection threshold, which mainly depends on the metering precision of the instrumentation and the performance characteristics of the process data acquisition. On this basis it provides a well-defined response within an appropriate period of time and unique behaviour on operational intervention. A properly calibrated LDL system is a valuable assistance to the operating personnel. It is sensitive and at the same time shows a low false alarm rate. The pillars of deterministic leak detection (LD) and leak location (LL) for pipelines are: measuring technology, communication and SCADA technology, and the application of hydraulic and thermodynamic laws. Any leak detection system that uses process data no matter if based on deterministic methods or not faces the same limits with respect to the achievable performance characteristics. Sensitivity, response time and cumulated leakage volume are the most important performance characteristics to be discussed here. The focus of this paper is on liquid pipelines. An important advantage of deterministic methods for leak detection and leak location (LDL) for pipelines compared to statistic and other methods is the possibility to predict the performance characteristics of the deterministic LDL system already before its implementation. This will be demonstrated by three examples of predicted results compared to data obtained from field experiments. Vice versa the deterministic approach enables us to specify the requirements of the instrumentation, communication system and SCADA components to obtain the requested sensitivity and accuracy of the LDL system. This is last but not least an important aspect for choosing an economically feasible solution for each pipeline.
1 Requirements & Performance Characteristics Following the demands of the German technical rules for cross country pipelines (Technische Regel für Rohrfernleitungen TRFL) the equipment for leak detection (LD) and leak location (LL) should be implemented as a continually operating monitoring system. Depending on the transported medium and its hazard potential the requirements are graded. The performance characteristics for a pipeline are individually specified in coordination with the authorities. In general, all operation modes from downtime over steady-state to transient flow situations should be fully covered. In all these situations reliable detection and location of a leak should be possible. The performance characteristics of a LDL system are: detection thresholds (sensitivity) response time cumulated leakage volume (cumulated loss until leak alarm) location accuracy low false alarm rate (reliability) fail safety (recognition of missing or erroneous input data) It s an engineering task to select and combine the optimal set of instruments, SCADA components, and different LDL methods for each particular pipeline to comply with the demands of operational safety and environmental protection. 2 Deterministic Methods 2.1 Direct Detection Methods The idea of all methods which are based on direct monitoring of those process data, which are possibly related to a leak, is relatively simple. When a leak arises, flow and pressure will respond to that in a characteristic manner. Direct methods work fine for steady-state pipeline operation as well as during downtime. However, due to its dimension, line content, and other conditions every pipeline responds with an individual hydraulic behaviour. On the basis of a detailed and well-calibrated pipeline model we can simulate the hydraulic behaviour for all operation modes. These simulation results can be used for planning and testing the LDL system. 2.1.1 Flow/Pressure Monitoring When a leak arises in a pipeline, the pressure will drop in a characteristic manner. Figure 1 Continual Monitoring of process data reading V 1.3 2
At the same time the flow will rise in the counter-flow direction and will drop in the flow direction. These effects can be detected by continual recording of all pressure and flow readings. An algorithm is used to filter out any significant variation in a recorded value from its noisy measurement signal. Such variations, of course, can also be induced by the usual operational interventions. Therefore, all events detected by direct methods should be checked for plausibility. If pressure drops while no operational changes have been executed, this event is likely to be caused by a leak. The time span t mentioned in Figure 1 has to be chosen according to the hydraulic response of the pipeline (section) and the scenario to be monitored. For steady-state operation (i.e. constant flow and pressure) the time slot t is normally just seconds. During downtime it can be stretched out up to minutes or even hours (cf. 2.1.3). Another application for pressure drop detection is time tagging pressure wave fronts when they pass the metering stations (cf. 2.1.4). These time tags are used to locate a leak. 2.1.2 Flow or Volume Balance Balancing the line pack of a pipeline section by simply comparing what flows out and what flows in is another direct leak detection method. Its accuracy, however, is frequently not acceptable for large pipelines. During transient flow situations this simple balance is blind. The larger the pipeline volume and the elasticity of the medium the longer last the transient periods. Figure 2 Flow or volume balance using metering values directly While operating a pipeline in steady-state mode this balance yields a constant difference. Any change in this difference without executing operational interventions might indicate a leak. 2.1.3 Downtime Pressure Monitoring Downtime Pressure Monitoring is a topic of interest, because it s a sensitive LD method and LD during downtime is separately mentioned in the demands of the TRFL. A requirement to achieve high sensitivity with this LD method is a hydraulically sealed off pipeline section. For this case there is a tight relation between line pack and line pressure. The drop in pressure and leak flow caused by a leak turns out to be: and () p t Qi = p 1 t pi 2 i χ (1) Qi () Q t = Qi χ t pi 1 (2) V 1.3 3
Factor χ is specific for each sealed off pipeline section. While pressure drops as a quadratic function the decrease of leak rate is linear. Both functions are shown in the following figure. It depends on χ, the initial leak rate Q i, and the initial pressure p i, when the quadratic contribution in equation 1 becomes significant. Figure 3 Drop of pressure and leak flow If the algorithm that is used to detect pressure drops (cf. 2.1.1) provides a reliable detection threshold of p, the response time of the Downtime Pressure Monitoring is given by: 1 pi t resp = 1 1 p (3) χ Qi pi Figure 4 Response time as a function of initial leak rate The figure above shows the typical relation between response time and leak rate. Small leaks take more time to be detected than large ones. The initial pressure is not important for Downtime Pressure Monitoring. The total leakage volume that accumulates until the leak is detected is calculated by integrating the leak rate within the response time: V 1.3 4
V leak = t resp 0 () Q t dt = Q i t resp Qi 1 χ tresp (4) 2 pi Comparison with field data recorded during leak experiments using orifices of different cross section area lead to a relation between leak width and leak rate. Figure 5 Pressure drop for different leak rates comparison to field data 0 Time [min] 0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 Pressure drop [mbar] (initial pressure 5 barg) -50-100 -150-200 -250-300 Cooling down 1.8 mm² 3.1 mm² 7.0 mm² theory 3.2 litre/min theory 6.3 litre/min theory 12.5 litre/min -350 For tiny leak rates the results of the Downtime Pressure Monitoring must be checked against a possible pressure drop that might be caused by decreasing pipeline temperature (broken curve in Figure 5). 2.1.4 Pressure Wave Detection If a leak arises abruptly, two pressure wave fronts are generated which propagate along the pipeline in opposite direction with a defined wave speed c. In oil pipelines the wave speed is around c = 1000 m/s. The exact value can directly be measured or calculated from the oil density and the pipeline material, its diameter, and wall thickness. Pressure waves which might be related to a leak show negative amplitude. If such a wave is detected and time tagged, the leak position between two adjacent metering stations at distance L can be calculated (cf. Figure 7): x leak 1 = ( L ± c ( t2 t1 )) (5) 2 There are two aspects related to the performance characteristics of this LL method: sensitivity and location accuracy. The detection threshold for pressure drop p is the limiting factor for sensitivity. For largeer distances the attenuation of pressure waves has to be considered. Depending on the distance of the leak to the metering station the minimum required pressure drop p * is larger than the detection threshold itself. V 1.3 5
The minimum pressure wave amplitude that is necessary to be detectable is related to the minimum (initial) leak rate by the Joukowski relation: Q min > A p ρ c * (6) Simulation of pressure surges on the basis of a detailed and calibrated pipeline model yields the attenuation characteristics. The next figure shows the aspect of sensitivity which is best, if the leak is in the middle between two metering stations and worst if the leak is near a station. Figure 6 Sensitivity of leak location by tracing pressure waves Because the wave speed in oil pipelines is about 1000 m/s, the accuracy of time tagging is very important to obtain location accuracy less than one kilometre. Precise time tagging requires a common time basis across all stations. The tagging precision can be estimated as δt = δt sync + 2 δt cyc + δt alg. If the synchronisation and cycle time are both within 100 ms and the algorithm that recognises pressure drops is able to tag the onset of a pressure drop with an uncertainty of two cycles, the predictable location accuracy is 500m. Real tests on a 47 km section of an 18 inch crude oil pipeline with approximately the above mentioned parameters found deviations from the true leak position between 30 and 320m. The leak position was near the middle of the section, the leak rates were 37-45 m³/h, and the pressure drop detection threshold was 0.25 bar. Both, the accuracy and homogeneity of the wave speed δc along the line and the accuracy of time tagging δt contribute to the location accuracy δx leak, which is given by the next equation. c δc δx = t t + t leak 2 1 2 δ (7) 2 c V 1.3 6
Figure 7 Accuracy of leak location by tracing pressure waves Also the location accuracy turns out to be best in the middle between two stations. 2.2 Model Based Detection Methods A leak detection and location system, which is based on or assisted by a wellcalibrated hydraulic model, is sensitive and provides reliable results even for nonsteady-state operation. Parallel to the operation of the pipeline its hydraulic behaviour is calculated in real time. Deviations are continually monitored and analysed, if they could be caused by a leak. 2.2.1 Dynamic Mass Balance A standard model based leak detection method is the Dynamic Mass Balance. The principle of the Dynamic Mass Balance (DMB) is similar to the simple Volume Balance discussed in 2.1.2. The deciding improvement is that the real time line pack is included in the Dynamic Mass Balance. Figure 8 Principle of the Dynamic Mass Balance The periodically calculated result of the DMB, which is usually displayed, is the (negative) leak rate. V Vout Vin VLP = + (8) t t t t DMB V 1.3 7
The first two terms of the above equation are direct process values of the volume metering stations (if necessary, beforehand integrated from flow to volume and further converted to standard volume). The third term is provided by the Real Time Hydraulic Simulation (RTHS) as sketched in the next figure. Figure 9 Input data for the Dynamic Mass Balance The analysis of the performance characteristics of the Dynamic Mass Balance is definitely more complicated than that for the direct methods. A general calculation of errors applied to the last term of equation 8 yields: V t VLP = p p V δ + t T T δ t LP LP δ (9) The impact of the accuracy of the RTHS itself on the accuracy of the line pack results can only be determined by comparison to the real system. While the -terms on the right hand side of the above equation depend on the elastic properties of the pipeline including its line pack, the δ-terms depend on the accuracy of the instrumentation and performance characteristics of the process data acquisition as well as on the particular operation mode. For non-steady-state operation these contributions have to be analysed with the help of detailed hydraulic simulations covering all expected operation modes. For steady-state operation the line pack contributions to the DMB are less important but not negligible. It has to be estimated how much all the pressure and temperature readings can vary during the time slot t (cf. equation 8/9). An example for the steady-state case precisely this means: all pressure and temperature readings show just little noise for otherwise constant values. Further, the pressure values are only transmitted and hence evaluated by the DMB, if the measured changes exceed a defined transmission threshold of e.g. 0.2bar. To calculate the pressure term of equation 9 some assumptions have to be made. The probability of the event that a pressure reading will send a new value (i.e. changed more than 0.2bar) during steady-state operation is here assumed to be 0.05. This assumption depends on the fraction of repeatability and transmission threshold and the noise amplitude of the reading in the particular operational mode. If there are pressure transmitters on each side of the section, a statistical argument leads to an average expectation value for the pressure variation of 0.021bar. Then, for a crude oil pipeline section of 40km length, 18inch diameter, and 7mm wall thickness the pressure contribution of equation 9 yields an expectable value of V 1.3 8
V LP p p 0. 021 δ 0. 62 m³/h = 0. 16 m³/h t 1 12 for t = 5 minutes. To estimate the expectation value for N such pipeline sections is not straight forward, because, due to its volume, each segment responds individually. However, the probability that all segments contribute their worst case at the same time decreases with growing N. The total average pressure contribution of equation 9 for the pipeline we compare with turns out to be 0.47m³/h. Temperature effects are almost negligible here, because a proper temperature model cares for a weak response on changing temperature values. The accuracy of the volume metering itself is the second important contribution to the possible error of the DMB. For steady-state operation at a (constant) flow of Q meas we can estimate its contribution: V in / out δt δq meas δ + 2 Qmeas (10) t t Qmeas The first term in brackets is the accuracy of the volume metering synchronisation divided by the time interval t of the DMB. It is negligible in most cases. The second term is the relative accuracy of the volume metering itself which is different for different flow rates. An 18inch crude oil pipeline is typically operated up to more than 1500 m³/h, in the example here it was about 1000m³/h. Then, an accuracy of the volume metering of 0.1% yields an error of 2 1000 m³/h 0.001 = 2 m³/h. In total this leads to an expected average error for steady-state operation of 2 2m³/h + 0.47m³/h + temperature impact 3.5m³/h. This is in agreement with daily experience as shown in the next figure. Figure 10 Dynamic Mass Balance During transient operation modes the curves of the Dynamic Mass Balance show larger deviations. Within a period of 72 days of continual batch operation (crude oils) 8 deviations of more than 5m³/h (including 3 deviations of more than 10m³/h) have been reported. All these deviations could definitely be assigned to abrupt operational intervention. In average, this is less than one false alarm per week. 2.2.2 Real Time Hydraulic Simulation & Deviation Monitoring A basic requirement for reliable hydraulic simulation is a complete and well-calibrated pipeline model. The pipeline can be roughly divided into pipeline sections and V 1.3 9
stations. The main attributes of the sections are their dimensions and the profiles of altitude, wall thickness, etc. All stations have to be modelled in detail. The next step is to parameterise all elements of the stations. Finally, the whole hydraulic model has to be calibrated until the simulation results coincide with the measured process data. The software suite SIR-3S (shown in the next two figures) is a powerful tool for this task. Figure 11 Detailed model of a pumping station drawn with SIR-3S Figure 12 Pump characteristics parameterised with SIR-3S For transient operation modes it is frequently necessary to do the simulation on time steps of one second or even less. All control loops and their hydraulic feedback is fully included. With a well-calibrated hydraulic model it is possible to simulate all operational modes very close to reality and this is the solid basis for what we call Deviation Monitoring. Two instances of the same simulation are configured V 1.3 10
differently. One is based on real time process data like pressure, flow, etc., and the other is based on control variables, setpoints, and status information of the valves. Both instances are operated simultaneously and are synchronised. Their results are continually compared to each other. The principle of Deviation Monitoring is sketched in the following figure. Figure 13 Principle of Deviation Monitoring The two simulation results are displayed and continually refreshed as the so called Hydraulic Profile. Parallel to the Hydraulic Profile the result of the Deviation Monitoring is displayed as shown in the figure below. Figure 14 Display of the Hydraulic Profile and the Deviation Monitoring In this way any deviation is directly visible. If its magnitude exceeds a defined limit, the operator can easily see where the problem is located and can take action for remedy. V 1.3 11
3 Summary To fully cover all operational modes from downtime over steady-state operation to transient flow the presented methods for leak detection and leak location must be implemented as a cooperating system. Three examples have been chosen to demonstrate the influence of the accuracy of the measured input data and the performance characteristics of the data acquisition on the achievable accuracy of a Deterministic Leak Detection and Leak Location System. Comparison to field data shows that it s possible to predict the performance characteristics. For steady-state pipeline operation these error calculations can be done analytically. For the nonsteady-state case this is only possible, if a complete and well-calibrated pipeline model is used. V 1.3 12