Research On Abnormal Valve Clearance Diagnosis Method For Diesel Engine Based On Cylinder Head Vibration Signal

The valve is one of the main moving parts of the diesel engine,and its working quality directly affects the ventilation effect of the diesel engine.Due to the frequent movement of the valve and the constant impact,it is easy to cause the valve to change due to wear,which will cause the diesel engine output power to drop.In severe cases,it may cause failure of other components and endanger the safety of entire unit and the site personnel.Therefore,the accurate and timely diagnosis of the abnormality of the valve clearance plays an important role in ensuring the working efficiency and work safety of the diesel engine.At present,the vibration signal-based diagnostic method is used by many researchers in the fault diagnosis of diesel engines because of its advantages of having signals for easy acquisition and monitoring.In particular,the cylinder head vibration signal contains a lot of effective information about the valve impact.However,the structure of the diesel engine itself is more complicated,and the path of vibration propagation is more.The vibration signal of the cylinder head generally exhibits obvious non-stationary characteristics.How to extract the sensitive fault features that characterize the valve clearance anomaly from the collected cylinder head vibration signal more effectively,and to accurately identify different fault categories is the two urgent key problems in the abnormal valve clearance fault diagnosis technology based on the cylinder head vibration signal analysis.Based on the above key issues,this paper mainly carried out the following experiments and research:(1)The fault simulation experiment of valve clearance anomaly was carried out under variable working conditions.A complete set of condition monitoring system was built for the actual unit on site.The vibration signal of the cylinder head was monitored when the valve clearance was increased,and the vibration signals of seven different valve clearance conditions were collected.In the time domain,the relative position of the impact signal in a periodic vibration signal is analyzed.In the frequency domain,the frequency range of the impact component generated by each excitation in the cylinder head vibration signal is studied.(2)Different non-stationary signal processing methods are compared and analyzed,including:wavelet packet decomposition,collective empirical mode decomposition and variational mode decomposition.On the basis of introducing the principle in detail,the parameter optimization method of variational mode decomposition and the fault feature extraction method of valve clearance anomaly are studied.According to the principle of minimum power spectrum entropy,the adaptive selection of penalty factor and decomposition layer is realized.Features such as time domain,frequency domain and singular value are extracted from the processed signal.(3)Various feature dimension reduction processing methods and machine pattern based fault pattern recognition methods are studied.Firstly,the T-SNE is used to reduce the high-dimensional features to three-dimensional,and the visual analysis is carried out.In the KNN algorithm,the k-value under the highest recognition rate is used to determine the optimal parameters.Then,the KPCA method is used to perform nonlinear dimensionality reduction.The contribution rate is chosen to select low-dimensional features that characterize 90%of the original feature information.Finally,feature selection is performed based on the feature importance scores in random forest.The wavelet packet decomposition and SVM are combined,EEMD and KNN are combined,and VMD-SVD is combined with random forest.The three fault diagnosis methods are compared and analyzed.