Research On Diesel Engine Fault Diagnosis Method Based On Information Fusion Technology

With the continuous development of industrialization process, more and more attention is paid to supervision and fault diagnosis technology for mechanical equipment. Diesel engine is considered as a typical power machinery. Both reciprocating motions and rotational motions are contained in diesel engines. Owning to the complex internal structure of them, it is difficult for the fault diagnosis to be accurately performed by single sensor thus the information fusion among multiple sensors is required. Hence, aimed at solving key technical problems in fault diagnosis of diesel engines, a research on diagnosis methods is conducted based on information fusion, in which, an effective multiple sensors information fusion algorithm is put forwarded. Thus diesel engine fault detection, prediction and decision analysis can be solved. The main work of this thesis is listed as follows:Firstly, the correspondence between diesel faults and supervision information of multiple sensors is established and the feature extraction of typical signals is conducted by amplitude analysis. Characteristic variables which can judge fault type are constituted by typical characteristic variables and other monitoring information. As it is considered, in clustering analysis, the effect of importance difference of characteristic variables on classification cannot be determined. Therefore, subjective and objective fusion weighted method is proposed.Secondly, in the case of unclear correlation of characteristic variable of faults by a complex of monitoring information of diesel engine, effective features of faults cannot be calculated neither through Mahalanobis distance nor though Euclidean distance. However, Fusion distance measurement method is put forwarded. Based on Fusion distance and Taguchi method, Fusion distance-Taguchi method for fault diagnosis of diesel engine is given. Optimization of data are fused from multiple sensors into a single system level performance metric using Fusion distance in the proposed scheme. And fault clusters based on Fusion distance values are set up. Fusion distance thresholds derived from the clustering analysis are introduced for fault detection. Simulation results show the effectiveness of Fusion distance-Taguchi method. When strong related issues exist in characteristic variables, M-P generalized inverse matrix can be used to calculate Mahalanobis distance, through which, the problem that strong correlation among characteristic variables makes it impossible to calculate Mahalanobis distance is solved.Thirdly, as it is considered, in clustering analysis, the effect of importance difference of characteristic variables on classification cannot be determined by Mahalanobis distance. Therefore, the definition of weighed Mahalanobis distance is proposed. Based on Mahalanobis distance and Taguchi method, weighted Mahalanobis-Taguchi method for fault diagnosis of diesel engine is given. Optimization of data are fused from multiple sensors into a single system level performance metric using weighted Mahalanobis distance in the proposed scheme. And fault clusters based on weighted Mahalanobis distance values are generated. Weighted Mahalanobis distance thresholds derived from the clustering analysis are used for fault detection and prediction. Simulation results show the effectiveness of weighted Mahalanobis-Taguchi method.Fourthly, neural network and D-S evidence theory are introduced to diagnosis of decision layer fusion. Diagnosis results of multiple sensors are fused, through which fault decision fusion can be effectively analyzed. Aimed at independent premise of evidence combination and issues related to evidence, combination method of related evidences is proposed. The influence of repeated using of the same information on the evidence combination is removed. So the results by combination method of related evidences are more accurate and reasonable. Decision layer fusion of related evidence is resolved. Simulation results show the effectiveness of combination method of correlative evidence.Fifthly, with regard to conflicts among evidences cannot be calculated through D-S evidence theory, weighted evidence theory is proposed, evidence conflict function and reliability function are built and the conflicting evidence is applied as information, through which evidence damage is avoided and decision layer fusion of conflicting evidence is resolved. Interference information of test results by fault sensor are excluded through weighted evidence theory.Simulation results show the effectiveness of combination method of conflicting evidence.Sixth, comprehensive diagnosis strategy based on information fusion is built up and fault diagnosis simulation system is developed through Matlab/Simulink. A unique solution for diesel engine fault detection, prediction and decision analysis is provided in this process, eliminating the need for developing a tool for each separately. The process is application independent. It can be applied to a wide variety of multivariate problems provided that normal operation and abnormal operation data are available. Then fault simulation test is given by the real data of diesel engine. Simulation results show the effectiveness of diagnosis algorithm in this thesis.Finally, this thesis comprehensively concludes and analyzes the content herein and points out the research direction in the future.