Research On Fault Diagnosis For Fule System And Valve Train Of Diesel Engine Based On Vibration Analysis

Diesel engine is the most widely used power source of machines and plays an important role in the national economy and people’s daily life.However,due to complex structure and usually harsh working environment,faults occur frequently and often cause terrible accidents and serious economic losses.According to statistics,the fuel system and valve train faults account for a large proportion of 27% and 15.1% respectively among all diesel engine faults.In addition,the fuel system and valve train directly affect the combustion situation of the engine,and their working conditions have important influence on whether the engine can operate normally and efficiently.Therefore,aiming at the requirement of the actual demand of the diesel engine fault diagnosis,the fuel system and valve train are selected as the research object of this study,and the fault diagnosis method of diesel engine fuel system and valve train are intensively researched based on the analysis of the cylinder head vibration signals.The main research contents include the following aspects:The main excitation sources of diesel engine cylinder head are analyzed,then the diesel engine cylinder head vibration information model is established,and the time domain and frequency domain characteristics of diesel engine cylinder head vibration are studied.Fault experiments are carried out on two diesel engines,which simulates 5 typical failure of valve train and 12 common fuel system and valve train faults,respectively.These fault experiments provide data support for the subsequent research work.After the basic principle of intrinsic time-scale decomposition(ITD)is studied,a fault diagnosis method based on ITD marginal spectrum and Mahalanobis distance is proposed,and this method is applied to the fault diagnosis of diesel valve train.The diagnosis results show that the method can roughly identify the faults of diesel valve train.To overcome the error accumulation problem of the directed acyclic graph(DAG),a separability based directed acyclic graph(SDAG)is proposed,and it is used to construct the multi-classification model of relevance vector machine(RVM).In addition,a general framework for multi-classification of RVM named probability-based error correcting output codes(PECOC)is also proposed,this framework has achieved the organic combination of RVM probability output and the error correcting output code multi-classification method.To overcome the defects of ITD,such as interpolation method,termination condition,etc.An improved intrinsic time-scale decomposition(IITD)method is proposed,and it is combined with SDAG-RVM for the fault diagnosis of diesel valve train.Experimental analysis results show that this method can obtain higher fault diagnosis accuracy,and is better than other conventional methods.After the mode mixing problem of IITD is detailed analyzed,a complete ensemble improved intrinsic time-scale decomposition(CEIITD)is proposed,in which the positive and negative noise is added in pairs to the original signal before each proper rotating component is obtained.Therefore,the residual noise in the finally decomposition results can be eliminated and the final averaging difficulty caused by the number of PRCs are not equal in different realizations can also be overcome.The simulation results show that the mode mixing problem of IITD method is well solved by the CEIITD method,and CEIITD is better than other conventional methods.To overcome the cross-term interference of Wigner distribution,this paper proposes an adaptive cross-term interference processing method based on IITD,called self-adaptive Wigner distribution,and a simulation signal is used to verify the effectiveness of the proposed method.Simulation results show that this method not only can eliminate the cross-term interference,but also can guarantee the time-frequency figure has higher resolution and time-frequency clustering.Targeting the problem that the original fast correlation-based Filter(FCBF)algorithm has not considered the redundant degree between the candidate feature and the selected feature subset,an improved FCBF algorithm is proposed.Then a new fault diagnosis approach for diesel engine fuel system and valve train based on self-adaptive Wigner distribution,improved FCBF and PECOC-RVM is proposed,by which the fault diagnosis problem can be transformed into an image classification problem,and the features,including moment invariants,gray statistical characteristics,textural features and differential box-counting fractal dimension,that have been widely acknowledged in image classification can be directly used as the fault features.Therefore,the difficulty and workload of feature extraction is reduced.Experimental analysis results show that the fault diagnosis approach of diesel engine fuel system and valve train based on adaptive Wigner distribution,improve FCBF and PECOC-RVM can obtain higher fault diagnosis accuracy,and is better than other conventional methods.In view of the performance a single unitary pattern recognition method decreases sharply when there are too many fault types,a fault diagnosis approach of diesel engine fuel system and valve train based on multistage Adaboost-RVM is proposed.The approach using KFCM algorithm to decompose the complex classification problem,which contain too many fault types,into several simple classification problems,and the output results of the complex classification problems can be obtained by the integration of the output of each simple classification problem.In addition,to further enhance the diagnostic ability of classification algorithm,the AdaBoost technique is adopted to enhance the weak classifiers.Experimental analysis results show that the proposed approach has obvious advantages in the fault diagnosis problem with many fault types,and is superior to other conventional methods.