Font Size: a A A

Study On Fault Prognostic And Health Management For Electronic System

Posted on:2010-07-06Degree:DoctorType:Dissertation
Country:ChinaCandidate:L J XuFull Text:PDF
GTID:1118360275480094Subject:Detection Technology and Automation
Abstract/Summary:PDF Full Text Request
In recent years, the failures in electronic system's essential module parts or the keycomponents often lead to the disastrous accidents occur, which cause the vast loss of themanpower, the material resource and the financial resource et al. All governmentsurgently need to realize Condition-Based Maintenance (CBM) on electronic systems,which is based on the fault prognostic and health management technique, and thistechnique can avoid the overmuch maintenance of traditional fixed-time maintenance orthe huge loss of subsequent maintenance. Because lacking accurate state judgment andhealth analysis on electronic systems, much nonessential maintenance have been donefrom the security angle which leads to much increase of the run cost. If we carry on thesubsequent maintenance, the losses can not be avoided in time. CBM owns someremarkable advantages such as the small logistics support scale, the good economywithstanding and the performance of avoiding significant incidents et al and has thegood prospect. It requests to be able to monitor and distinguish the faults of electronicsystem as early as possible, along with the ability of managing system's health andpredicting its state. Traditional fault diagnosis technology can not satisfy the actual need,just based on this reason study on fault Prognostic and Health Management (PHM)arouses the domestic and foreign researchers' enormous interest. At present the researchresults in this domain are very few especially the domestic research just is in start stage,therefore PHM will be the key research direction in the future.PHM mainly includes some key points such as state monitoring and healthmanagement, module-level or component-level fault diagnosis and state prediction et al.Based on the above reasons, the main works of this dissertation are shown as follows:1. Study on state monitoring and health management for electronic system. Statemonitor and health management is the premise of deciding whether electronic systemneeds to be maintained and is very important in the whole system. Electronic systemsdo not show obvious faulty features during this period. How to extract state features isvery essential if we want to monitor incipient faults as early as possible. Becauseelectronic systems own various characters, we should make different methods to extract their state features. The dissertation presents an analog circuit (which represents simpleelectronic system) and a radar transmitter (which represents complex electronic system)as examples respectively: using LDA to extract normal state features from the formerand using wavelet technology to the latter, then the processed features are used to formthe observation sequences sent to HMM, where an improved algorithm is presented totrain DHMM. HMM is used as the state monitor to calculate the KL distance ofunknown state, which shows that the proposed method can convert the unconspicuouschange of incipient fault process into the obvious change of KL distance successfully,based on which we can estimate the health status of electronic system accurately andprovide basis for CBM. This study answers question such as "whether should electronicsystem be repaired?".2. Study on module-level fault diagnosis for electronic system. If electronic systemneeds to be repaired, there usually exists an anfractuous relation among its sub-modules,so it is very difficulty to understood electronic system's fault propagation mechanism.Especially for some complex electronic systems, their fault trees or multi-signal modescan not be built successfully, and then Bayesian network is the best choice. For "blackbox" system, the fault model can be built and the faults can be diagnosed successfullythrough learning both Bayesian network's structure and its parameters. The dissertationbrings forward a new structure learning algorithm which inserts both the cross operatingand the mutation operating into PSO algorithm. The experimental results show that theproposed algorithm has good precision and excellent efficiency, which provides thepossibility for Bayesian network to apply to diagnose faults in complex electronicsystems. Taking a radar transmitter as the example, the dissertation gives detailed designsteps and makes corresponding simulation and the experimental results show thatBayesian network is very effective to diagnose the faults in complex electronic system.This study answers question such as "where is the fault module?".3. Study on component-level fault diagnosis for electronic system.It is necessary torecognize incipient faulty components while the replacement modules or spare parts areinsufficient, and the dissertation puts forward a novel method based on LDA and HMMto diagnose the incipient faults in analog circuit, where the performance of LDA isimproved through overcoming the shortcomings existing in the original LDA. Throughcomparing with BP network and some other methods, the experiment results show that the novel method has the best recognition capability. The dissertation also makesdetailed experimental analysis on selection of HMM's parameters, its types and itsstructures. Considering a kind of feature containing less fault information, thedissertation presents a fault diagnosis method based on feature fusion. Different kinds oforiginal feature vectors are extracted from analog circuit simultaneously, and then LDAis used to reduce the dimensions of the original feature vectors and remove theirredundancy together aiming at achieving their fusion skillfully. Finally HMM is used asthe classfier to accomplish the diagnosis of the incipient faults. The experimental resultsshow that the proposed method provides higher recognition rate compared to that of anykind of feature combined with HMM. This study answers question such as "what is thefaulty component?".4. Study on state prediction for electronic system. It is necessary to predictelectronic system's state if it needs not to be repaired. State prediction is a highermonitoring technology compared to fault diagnosis. State prediction usually makes useof electronic system's historical information to estimate its future state and tendencyaiming at avoiding disastrous faults. The dissertation takes both a radar transmitter andan analog circuit as examples and an improved GM (1, 1) is used as the state predictorthrough analyzing their key testing signals' characters, where a metabolism method ispresented to make the model parameters on-line change and PSO algorithm is used toobtain the best forecast dimension. The improved model is tested and the experimentalresults show that the improved model has good precision and performance. This studyanswers question such as "when will the failure occur?".
Keywords/Search Tags:state monitor, fault diagnosis, Hidden Markov Model, Bayesian network, linear discriminant analysis
PDF Full Text Request
Related items