Research On Diagnosing Intermittent Faults To Reduce False Alarms Of Mechatronics BIT

The Built-in test (BIT) is defined as an on-board hardware-software diagnostic means to locate and identify faults. It is a common key technique of system design, subsystem design, condition monitoring, fault diagnosis, maintaining decision and so on. The BIT has been applied in the aviation and weapon systems since the end of 1970's. It plays an important role in improving testability, reliability and diagnostic capability of the system. But the high false alarm rate (FAR) is one of the important factors that prevent BIT from being more extensively applied. How to reduce the high FAR while retaining high fault detection rate (FDR) is a key issue of BIT that should be well solved.The intermittent fault is a main factor that temporarily disables mechatronics system and incurs the BIT false alarms. Diagnosing intermittent faults is an efficient approach to improve performance and to reduce false alarms of the BIT. The present research on diagnosing intermittent faults has two main issues that are not satisfactorily solved. One is the mechanism of intermittent faults and their influences on the BIT, which are important for diagnosing intermittent faults but are seldom investigated. The other is the shortage of diagnostic technologies. For example, the stochastic process technology cannot learn and use knowledge automatically. A lot of training samples, which are unavailable yet in reality, are needed for artificial intelligence methods such as neural network to get dependable model.Supported by Research on Design Technology of Mechatronics-BIT, and taking the high BIT FAR into account, this paper is aiming to solve the above two issues of the BIT in diagnosing intermittent faults. The mechanisms of the intermittent faults and how the intermittent faults influence the BIT are analyzed systematically. Then based on the previous work, the technology of diagnosing intermittent faults so as to reduce BIT false alarms is studied. The content of the dissertation is as follows.1. The mechanisms of how the intermittent faults influence the BIT performance are analyzed systematically and the system model is built.(1) In order to analyze how the intermittent faults influence the BIT performance and model the mechatronics system with intermittent faults, the factors that induce the intermittent faults are analyzed first, and the mechanisms of intermittent faults are investigated. Then by dividing the system into 3 states, i.e., the OK state, intermittent state and faulty state, the Markov models of the system are set up. The relation between the 3-state model and the 2-state model is studied. The qualitative analysis shows that the intermittent fault is an important factor that affects BIT with high FAR and low FDR. At last, a double-threshold based 3-state diagnostic method is presented and the diagnostic principle and the method of threshold-decision are introduced. Then by comparing the3-state diagnostic method with the 2-state diagnostic method, the influences of intermittent faults on the BIT are studied. The quantitative analysis and simulation results show, the diagnosing intermittent faults is an efficient way to reduce the BIT false alarms with high fault detection rate.(2) The system states are modeled on Markov models according to their behaviors. Then because the states cannot be observed directly and should be judged by their behaviors, the system is modeled on Hidden Markov Model (HMM). These provide a theoretic base for diagnosing intermittent faults so as to improve the BIT performance and reduce false alarms.2. The technology of diagnosing intermittent faults to reduce BIT false alarms is studied deeply.(1) In order to solve the difficult issue of diagnosing intermittent faults that occur intermittently and transitorily, the HMM based intermittent faults diagnostic methods are presented in view of the merit of the HMM that have the ability to deal with continuous dynamic signals. First, the HMM based diagnostic methods are studied. The influences of the HMM with different types and different Markov models and different parameters on diagnosis are studied systemically. Then, based on the system HMM built before, two HMM based intermittent faults diagnostic methods are presented. One is the HMM-model-based diagnostic method under the condition that intermittent faults training samples are available. The other is the HMM-threshold-based diagnostic method under the condition that intermittent faults training samples are unavailable. The experimental results show that both methods can diagnose intermittent faults efficiently. The performance of BIT is improved and the BIT false alarms that caused by intermittent faults are reduced.(2) In order to diagnose intermittent faults with small samples, a SVM based intermittent faults diagnostic method is proposed. The Support Vector Machine (SVM) is a novel powerful machine-learning method with small samples that is based on the VC dimension theory and Structural Risk Minimization (SRM) principle. First, the SVM based diagnostic method is studied. The experimental results show that the SVM based diagnostic method is better than the neural network (NN) based method under small samples condition. The BIT performance is improved. Then the SVM based intermittent faults diagnostic method is presented. With the features extracted from small samples, the SVM models are trained to diagnose intermittent faults. The experimental results show that this method can diagnose multiple intermittent faults accurately with small training samples and the BIT false alarms are reduced.(3) To further improve the BIT performance , two different HMM-SVM based diagnostic methods, combining the HMM that have the ability to deal with continuous dynamic signals and the SVM with perfect classify ability, are presented. One is theHMM and SVM connected in parallel, the other is the HMM and SVM connected in series. With the features extracted from vibration signals, HMM-SVM diagnostic models are trained to diagnose. The experimental results show that these proposal methods are better than the HMM-based or the SVM-based diagnostic methods with high diagnostic accuracy. The BIT performance is improved. At last, in order to eliminate the BIT false alarms that induced by intermittent faults, a HMM —SVM based intermittent faults diagnostic method is presented. Based on the HMM system built before, the pattern-matching-degree features are extracted. With these features SVM is used to identify system states. The experimental results show that this method can diagnose intermittent faults efficiently and the BIT false alarms are reduced.(4) In order to acquire correct diagnostic knowledge automatically from impure intermittent fault samples, the unsupervised methods based on Decision-Improved one-class SVM (1-DISVM) are presented. By modifying the decision-function of one-class SVM (1-SVM), which is able to find outliers from a dataset without any class information but seldom be applied to pattern recognition for it's algorithm limits, 1-DISVM is formed. Based on 1-DISVM, a multi-pattern classing-model and a clustering-model are constructed. Based on these 1-DISVM models, two intermittent faults diagnostic methods are studied. The experimental results show, the multi-pattern classing-model can get rid of the influence of wrong samples to achieve precise classification with small samples, and the clustering model can not only recognize the unknown fault patterns adaptively and precisely, but also learn the nature of the input-patterns from small samples and diagnose the faults successfully. By diagnosing intermittent faults, the BIT performance is improved and BIT false alarms are reduced. These unsupervised learning-methods are new ways for the BIT to acquire knowledge.3. In order to evaluate the efficiency of diagnosing intermittent faults to reduce the BIT false alarm, a Mechatronics-system BIT with the ability to diagnose intermittent faults, which are realized on many technologies such as HMM and SVM and so on, is designed. By diagnosing intermittent faults, the BIT performance is improved and false alarms are reduced.