Reliability Analysis of an Anti-lock Braking System using Stochastic Petri Nets

Transcription

1 Reliability Analysis of an Anti-lock Braking System using Stochastic Petri Nets Kshamta Jerath Frederick T. Sheldon School of Electrical Engineering and Computer Science Washington State University, Pullman, WA 99164, USA Abstract The Reliability Analysis of an Anti-lock Braking System using Stochastic Petri Nets is a work in progress and an extension to the work presented in the paper Specification, Safety and Reliability Analysis Using Stochastic Petri Net Models [9]. The current work attempts to model the Anti-lock braking sub-system of a vehicle system using Stochastic Petri Nets. The reliability analysis is undertaken with particular focus on coincident failures of components. The model is specified in C-based Stochastic Petri Net language, the input language for SPNP. Introduction A complex system (like a vehicle) is composed of numerous components and the probability that the system survives (efficient or acceptable degraded operation) depends directly on each of the constituent components. The reliability analysis of a vehicle system can provide an understanding about the likelihood of failures occurring in the system and an increased insight to manufacturers about inherent weaknesses. In [9], the authors present Stochastic Petri Net (SPN) models of a vehicle dynamic driving regulation (DDR) system. Subsystem representations of the Anti-lock Braking system (ABS), the Electronic Steering Assistance (ESA), the traction control (TC) and a combined model are developed and analyzed for critical failures. In this study, we focus on the Anti-lock braking system and develop a stochastic Petri net model to model coincident failures of certain components, under fully operational as well as degraded operation conditions. The assumption that failures occur independently (in a statistical sense) in hardware components is a widely used and often successful model for predicting the reliability of hardware devices. However, components generally interact with each other during operation, and a faulty component can affect the probability of failure of other components. Such failures are not coincident in the sense that they occur simultaneously, but in the fact that failure of one increases the probability of the failure of another. It is this aspect of the system that we have undertaken to model in this study. The model developed includes the failure modes and effects associated with the failure rates of critical components. The program representing the model is written in CSPL (C-based Stochastic Petri net Language) and the stochastic analysis is carried out using SPNP (Stochastic Petri Net Package). SPNP is a versatile modeling tool which allows the specification of SPN reward models, the computation of steady state, transient, cumulative, time-averaged and up-to-absorption measures and sensitivities of these measures [2].

2 Anti-lock Braking System The Anti-lock braking system prevents wheel lockup during an emergency stop by modulating the brake pressure. It permits the driver to maintain steering control and stop the vehicle in the shortest possible distance under most conditions. The ABS consists of the following major components [6, 7]: Wheel Speed Sensors: These measure wheel-speed and transmit information to an electronic control unit. Electronic Control Unit (Controller): This receives information from the sensors, determines when a wheel is about to lock up and controls the hydraulic control unit. Hydraulic Control Unit (Hydraulic Pump): This controls the pressure in the brake lines of the vehicle. Valves: Valves are present in the brake line of each brake and are controlled by the hydraulic control unit to regulate the pressure in the brake lines. Under braking, the electronic control unit (ECU) reads signals from electronic sensors monitoring wheel rotation. If a wheel s rate of rotation suddenly decreases, the ECU orders the hydraulic control unit (HCU) to reduce the line pressure to that wheel s brake. Once the wheel resumes normal operation, the controls restore pressure to its brake. Depending on the system, this cycle of pumping can occur at up to 15 times per second. Anti-lock braking systems use different schemes depending on the type of brake in use: Four channel, four sensors ABS; three channel, three sensors ABS; two channel, two sensors ABS. In this study we focus on the four channel four sensor ABS [1]. Assumptions In the model developed, we assume a four channel, four sensor ABS. The model can be easily modified to represent other ABS schemes. It is assumed that on an average a passenger vehicle travels for 200,000 miles at a speed of 50 mph in its lifetime. Hence, the analysis is carried out for 50K hours, the average life span of a passenger vehicle being 40K hours. The components of the ABS are assumed to operate independent of each other, wherever coincident failures are not explicitly modeled. In order to allow a Markov chain analysis, the time to failure of all components is assumed to have an exponential distribution. This signifies that the distribution of the remaining life of a component does not depend on how long the component has been operating. The component does not age or it forgets how long it has been operating, and its eventual breakdown is the result of some suddenly appearing failure, not of gradual deterioration [10]. While this might be true for electronic components, the failure of other mechanical parts like valves might occur due to gradual deterioration. However, we assume an exponential distribution to keep the model simple. Every component operates in three scenarios: normal operation, degraded operation or loss of stability. The system is assumed to fail (failure situations resulting in absorbing states) when either more than five components are functioning in a degraded state; or more than three components are causing loss of stability; or there is a loss of vehicle. A component operating in a degraded condition causes its failure rate to increase by one order of magnitude, while a component causing loss of stability causes the failure rate to increase by two orders of magnitude. The correlation between failure rates of two related components (to model coincident failures) is consistent with the above scheme.

3 Since the model is an abstraction of a real world problem, predictions based on the model must be validated against actual measurements collected from the real phenomena. A poor validation may suggest modifications to the original model [10]. The ABS Model A Petri Net (PN) is a bipartite directed graph whose nodes are divided into two disjoint sets called places and transitions. Directed arcs in the graph connect places to transitions (called input arcs) and transitions to places (called output arcs). A marked Petri net is obtained by associating tokens with places. In a graphical representation of a PN, places are represented by circles, transitions are represented by bars and the tokes are represented by dots in the places. The firing of a transition is an atomic action in which one or more tokens are removed from the input place of the transition and one or more tokens are added to each of the output place of the transition. By requiring exponentially distributed firing times, we obtain stochastic Petri nets (SPN). Stochastic Reward nets are SPNs augmented with the ability to specify output measures as reward-based functions, for the evaluation of reliability for complex systems [3]. In our SRN model, the ABS is represented as a combination of all the important components it consists of, as shown in Figure 1. It represents the operation of the ABS under normal, degraded and lost stability conditions. Loss of vehicle, extreme degraded operation and extreme loss of stability signify critical failures and determine the halting condition for the model. The model is instantiated with a single token in the start place. When the central_op controllerop controller controllerfail failedcontroller controllerdegradedop controllerlosop controllerlovop controllerdegraded controllerlos degraded_operation loss_of_stability loss_of_vehicle Figure 2: SPN model of an ABS component central central_op start braking and the axle_op transitions fire, a token is deposited in each place that represents a component of the ABS. The operation of each component is now independent of every other component. The model of a component of the ABS is shown in Figure 2. The component depicted here is the controller. Every component either functions normally as shown by the controllerop transition or fails as shown by the controllerfail transition. A failed component may either cause degraded operation, loss of stability or loss of vehicle. The probability of any one of these three transitions occurring is axle axle_op mbrakecyl controller tubing piping axlecentral FRWheel RLWheel RRWheel FLWheel degraded_operation loss_of_stability loss_of_vehicle Figure 1: The ABS model

4 different for each component. When the failure causes either degraded operation or loss of stability, the component continues to operate, though the failure rate increases by one and two orders of magnitude respectively. Coincident failures are modeled in a similar manner. The function that calculates the failure rate of the transition controllerfail is shown in Figure 3. It is assumed that malfunctioning tubing affects the operation of the controller. Hence, while calculating the failure rate of the controller, the normal rate is increased by one order of magnitude if the tubing has failed causing degraded operation (indicated by a token in the tubingdegraded place). While modeling other coincident failures like loss of controller itself affecting the failure rates of the hydraulic pump, if the failure of the controller causes loss of stability, the failure rate of the hydraulic pump increases by two orders of magnitude. Only a few coincident failures have been represented in the model. However, coincident failures between other components can be easily modeled by suitably modifying the failure rate function of the component in question. The model is easily extensible to include other components deemed relevant to the ABS. Results The Stochastic Petri Net Package (SPNP) allows the specification of SRN models, the computation of steady state, transient, cumulative, timeaveraged, up-toabsorption measures and sensitivities of these measures. Steady-state analysis of SRNs is often adequate to study the performance of a system, but timedependent behavior is double controllerrate() { double controller_rate = ; if (mark("controllerlos") > 0) return controller_rate * 100; if ((mark("controllerdegraded") > 0) (mark("tubingdegraded") > 0)) return controller_rate * 10; return controller_rate; } Figure 3: Variable rate to model coincident failures 1 Reliability of ABS Time (in hrs) Without coincident failures With coincident failures MTTF (w/o) = hrs. MTTF (with)= hrs. Figure 4: Reliability analysis results sometimes of greater interest: instantaneous availability, interval availability, reliability, response time distribution, and computational availability. The reliability of the system at time t is computed as the expected instantaneous reward rate at time t [3]. Transient analysis of the ABS model developed was carried out and the reliability was measured between 0 and 50K hours (representing average lifetime of a passenger vehicle). The

5 expected values of reliability at various time instances was determined and plotted as a function of time. The measure was predicted at 169 points along the range. The interval between the points did not remain constant along the entire time range; instead the time range was divided into four segments. Each of these segments has a different time interval. In Figure 4, the Y-axis gives the measure of interest - the reliability; while the time range (0 to 50K hours) is shown along the X-axis. The shape of the curve is not a property of the system but of how the data was collected from the Petri net model. Conclusion and Future Work In this study, we have shown how to model coincident failures in the Anti-lock Braking system of a passenger vehicle using Stochastic Reward Nets. In order to specify the system, we had to make some system assumptions. The stochastic Petri net modeled a few coincident failures possible in a four channel four sensor ABS. The model, however, is easily extensible to model other schemes of ABS. Other coincident failures between components can be easily modeled by suitably modifying the failure rate function of the component in question. In order to specify the system and carry out the reliability analysis, we used SPNP. The goal of future work is two-fold. First, specify and analyze the model developed using UltraSAN, a software tool for model-based performance, dependability and performability evaluation of computer, communication and other systems [8]. We would like to compare the results for the reliability analysis of the model from both SPNP and UltraSAN tools. Second, extend the model to include other systems that operate in conjunction with the ABS sharing some components e.g. Acceleration Slip Regulation (ASR) and Electronic Steer Assist (ESA). References [1] Bosch, R. Automotive Handbook, Bentley Pubs. [2] Ciardo, G.; Muppala, J.; Trivedi, K. SPNP: Stochastic Petri Net Package. Proc. 1st Int. Workshop on Modeling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS'93). [3] Ciardo, G.; Muppala, J.; Trivedi, K. Stochastic Reward Nets for Reliability Prediction. Communications in Reliability, Maintainability and Serviceability 1(2): [4] Ciardo, G.; Muppala, J.; Trivedi, K. SPNP User Manual Version 6. [5] Dugan, J. B.; Ciardo, G. Stochastic Petri Net Analysis of a Replicated File System. IEEE Transactions on Software Engineering 15(4): [6] Kolsky, M. ABS: Understanding Anti-Lock Brakes. [7] Nice, K. How Anti-Lock Brakes Work. [8] Sanders W. UltraSAN User s Manual version [9] Sheldon, F. T.; Greiner, S.; Benzinger, M. Specification, Safety and Reliability Analysis Using Stochastic Petri Net Models. ACM International Workshop on Software Specification and Design. [10] Trivedi, K. Probability and Statistics with Reliability, Queuing and Computer Science Applications, Prentice-Hall.