| Diesel engines are being widely employed in many fields as a common power source.To ensure normal operation of diesel engine and mechanical system,accurate and efficient fault recognition and condition evaluation method based on vibration signal without disassembling becomes a research hotspot.Due to complex structure,weak power feature in early failure stage is easily submerged in strong background noise.So analyzing original vibration signal in time and frequency domains,extracting and selecting features from it,and recognizing faults and states of diesel engine by efficient classifier are significant.As frequent faults,fuel injection mechanism and fuel supply failure and valve and its seat failure are taken as objects,important parts of modern faults detection and state recognition programme,such as time-frequency analysis,feature extraction and selection,information fusion and pattern recognition,are researched.Main results are given in the following:(1)The key parameters of variational mode decomposition(VMD)are researched to realize decomposing signals adaptively.For unstable accuracy caused by manually selected composition level K,two optimized approaches are proposed.The first approach is decomposing signals several times,and the results are evaluated by correlation coefficient and energy ratio to abtain the best one.The second approach is analyzing the frequency component of singnals by scale-space representation of power spectrum to provide decomposing parameters for VMD.Between the two approaches,the first one is suitable for decomposing small quantities of signals with complex frequency,the second one is suitable for decomposing large quantities of signals with several obvious frequency peaks.For there is no acceptable selection criteria of quadratic penalty ?,square root amplitude is employed to evaluate the results under different ? and ?=6400 is obtained as the optimal value.Results show the quadratic penalty could filter noise better.By optimizing VMD comprehensively,feature components are extracted accurately,which is laid the foundation to subsequent state recognition.(2)A novel single channel state recognition method is proposed by combining the optimized VMD and kernel fuzzy C-means clustering algorithm(KFCM).A feature selection method of multi evaluation criteria based on correlation test,chi square test and relief F algorithm is researched.The high correlation components of VMD result are extracted firstly,then the three methods are used to evaluate these features integratedly.Finally,adaptive kernel fuzzy C-means clustering(AKFCM)is proposed as pattern recognition algorithm.To solve the problem in kernel functions and their parameters selecting,five kernel functions are compared.An adaptive kernel parameter,which changes adaptively with the Euclidean distance between the sample point and the cluster center,is proposed.Based on that,the accuracy of KFCM algorithm is greatly improved.It is proved that the recognition rates of fuel injection,rail pressure and valve clearance based on the optimized VMD-AKFCM method are97.4%,95.1% and 94.2% respectively.Results show the optimized method has ability to recognize parts of states and has advatages in high efficiency and simple structure.(3)Combining with genetic algorithm(GA)and expectation maximization(EM),a diesel engine state recognition method based on multi-channel vibration signal feature fusion is proposed.Correlation-based feature subset selection(CFS)is taken as fitness function for GA to search feature subset globally and obtain the best subset.EM is taken as classifier,and cross validation(CV)is used to optimize it to select cluster number adaptively and improve its accuracy and generalization.Based on thses,valve clearance failure,abnormal rail pressure,abnormal injection quantity,abnormal injection timing and two kinds of compound failures are recognized,the recognition rates are 90.4%,99.0%,98.3%,88.0%,99.0%,93.4% respectively.Reults show this method could detect enegine states by multichannel vibration signals,and has adavantages in recognition rate and recognize types.(4)Based on research of instantaneous speed collection and information fusion,a diesel engine state recognition method of fusional instantaneous speed and virabtion is proposed.Measurement method for instantaneous angular speed based on analog-to-digital conversion(ADC)is improved.Based on K-means clustering method and Fourier series model,a method to accurately identify outliers and define angular displacement values corresponding to outliers under periodic fluctuant speed is proposed.Experiment proves the reliability and effectiveness of this method,which can provide a solution for on-site measurement for diesel engine instantaneous speed.The valve clearance fault,abnormal rail pressure,abnormal fuel injection quantity and abnormal fuel injection timing are recognized by fusional instantaneous speed and virabtion features.The recognition rates are 98.5%,99.2%,98.3% and 92.0%respectively.Reults show this method could detect enegine states by fusional instantaneous speed and vibration signal,which has advantages in sensitivity and recognition rate for abnormal combustion state.In conclusion,based on vibration signal,this paper proposed three strategies for diesel engine state recognition: single channel vibration signal,multichannel vibration signal feature fusion and fusional speed and vibration signal.Several algorithms,such as VMD,KFCM,GA,EM and so on,and instantaneous speed collection method are researched.A method system for diesel engine state recognition for different condtions is bilut.The recognition results for faults and unnormal condtions porve its accuracy,and efficiency.The work of paper has guiding significance for the research and engineering application of engine state recognition. |
PDF Full Text Request