Research On Typical Fault Feature Extraction And Intelligent Diagnosis Methods Of Diesel Engine

Diesel engine is the power device of modern national defense equipment and industry.It is widely used in the fields of vehicles,industrial uses,ship transportation,military equipment and engineering machinery,which itssecurity and reliable operation has irreplaceable position and function for the implementation and completion of tasks.The harsh working environment such as high temperature,high pressure and high speed makes the internal mechanism of diesel engine easy to produce failure.However,due to the inherent characteristics such as compact and complex structure,multiple impact coupling and variable working conditions,the fault symptoms are relatively hidden and the causal relationship is complex.Traditional fault diagnosis technology needs expert experience,which is difficult to meet the diagnosis requirements of current intelligent manufacturing.With the proposal of new manufacturing strategies such as "German industry 4.0 strategy","American industrial Internet" and "made in China 2025",artificial intelligence monitoring and diagnosis has been listed as an important key technology of intelligent manufacturing.Therefore,how to effectively extract fault characteristics and more convenient intelligent diagnosis is a hot research direction and difficulty in various fields.In order to provide a solid theoretical basis for intelligent diagnosis system,based on the vibration monitoring data of diesel engine,this paper combines the modern signal processing method with artificial intelligence technology to study the fault feature extraction and intelligent diagnosis of diesel engine.The main contents of this paper are as follows:In view of the deficiency of monitoring and diagnosis relying on keyphasor signal for whole cycle signal acquisition,a method of whole cycle signal acquisition without keyphasor triggering of diesel engine is proposed.Since the vibration and shock characteristics of cylinder head are not obvious and there are many interferences,an adaptive threshold Teager energy operator threshold is designed,which can enhance and effectively identify the transient impact characteristics.At the same time,a new time-frequency analysis method,synchronous compression generalized S-transform,is proposed based on the gearbox vibration to estimate the crankshaft speed.Finally,the whole cycle signal without keyphasor is obtained by adaptive correlation analysis.The experimental results show that compared with the keyphasor acquisition method,the proposed method can achieve the whole cycle vibration signal acquisition under different working conditions.Aiming at the nonlinear,nonstationary and multi-transient characteristics of diesel engine vibration signal,a novel feature extraction method of diesel engine vibration signal based on adaptive signal decomposition is proposed.In order to solve the problem that traditional PSO algorithm is prone to fall into local optimal position and slow or non-convergence,a dynamic adaptive chaotic particle swarm optimization algorithm is proposed.On this basis,intelligent optimization of key performance parameters combination of VMD is carried out.In addition,a new feature extraction methodof diesel engine is designed based on improved MFCC of speech recognition.This method can extract fault characteristics under different working conditions and different fault conditions.In order to solve the problems that it is difficult toidentify condition identification using single domain characteristic and the complex and tedious feature engineering may abandon the important working condition information,different condition identification methods of diesel engine are proposed.Firstly,the VMD is used to decompose the vibration signals adaptively,and multi domain features are extracted from IMF components to ensure that different spatial information of each working condition is expressed.Aiming at the dimension disaster and redundant information interference in multi-domain feature fusion,a feature fusion method of mRMR-SKCCA is designed and KNN is introduced and improved to realize working condition recognition.Secondly,an intelligent working condition recognition based on one-dimensional deep canonical convolutional neural network is proposed.The automatic feature extraction function of CNN and the complex nonlinear feature correlation mapping ability of DCCA are combined to realize theend toend intelligent condition recognition.The experimental results show that the classification accuracy of the proposed method is better than other classifiers and traditional deep models.Aiming at the problem that traditional deep learning model fails under the operating conditions of variable working conditions,a new deep convolution neural network model MBA-CNN is proposed,which imitates the human visual attention selection mechanism.It endows the model with bionic attention ability,that is,only focusing on the signal region and convolution characteristics related to the fault category.Secondly,in view of the problem that there is not enough labeled fault data in practical engineering application and many data samples in normal working condition,a new intelligent diagnosis method based on deep transfer learning is proposed.Taking the normal data of different working conditions as training sample,deep convolution neural network can learn the characteristics of the change working condition and normal state.Then,a small number of faults are used to fine tune the model,so as to realize the intelligent fault diagnosis under variable working conditions.The results show that the method can accurately identify the various faults under the variable working conditions,and its performance is better than other models.