Probabilistic reasoning in intelligent systems: networks of plausible inference
Probabilistic reasoning in intelligent systems: networks of plausible inference
Fault detection via factorization approach
Systems & Control Letters
The nature of the unnormalized beliefs encountered in the transferable belief model
UAI '92 Proceedings of the eighth conference on Uncertainty in Artificial Intelligence
Automatica (Journal of IFAC)
Fall Diagnosis using Dynamic Belief Networks
PRICAI '96 Proceedings of the 4th Pacific Rim International Conference on Artificial Intelligence: Topics in Artificial Intelligence
Mining Causal Knowledge from Diagnostic Knowledge
ADMA '08 Proceedings of the 4th international conference on Advanced Data Mining and Applications
| Hi-index | 0.00 |
Safe and reliable control of many processes, simple or complex, relies on sensor measurements. Sophisticated control algorithms only partially satisfy the growing demand for reliability in complex systems because they depend on the accuracy of the sensor input. Noise-ridden or faulty sensors can lead to wrong control decisions, or may even mask a system malfunction and delay critical evasive actions. Thus for optimal and robust operation and control of a process system, correct information about its state in terms of signal validation is of vital importance. To this end, process control methods need to be augmented with sensor fault detection, isolation, and accommodation (SFDIA) to detect and localize faults in instruments. SFDIA can be defined as (a) the detection of sensor faults at the earliest, (b) isolation of the faulty sensor, (c) classification of the type of fault, and (d) providing alternative estimates for the variable under measurement.In this chapter, we introduce the use of Bayesian belief networks (BBN) for SFDIA. Bayesian belief networks are an effective tool to show flow of information and to represent and propagate uncertainty based on a mathematically sound platform. Using several illustrations, we will present its performance in detecting, isolating, and accommodating sensor faults. A probabilistic representation of sensor errors and faults will be used for the construction of the Bayesian network. We limit sensor fault modes considered in this work to bias, precision degradation, and complete failure. Fuzzy logic forms the basis for our second approach to SFDIA. In decision-making related to fault detection and isolation, fuzzy logic removes the restrictions of hard boundaries set by crisp rules. The main advantage of fuzzy logic is its simplicity without compromising performance. As shown in the results, three fuzzy rules per variable are sufficient for a model-based fuzzy SFDIA scheme. A comparison of these two methods based on the experience of this application provides valuable insights.