Research On Diesel Engine Anomaly Detection And Fault Diagnosis Technology Based On Deep Feature Extraction Of Auto-encoder

Diesel engine is an important source of power for many large-scale machinery and equipment,and has been widely used in industrial,marine,electric,military,and other fields.However,due to the harsh working conditions,complicated mechanical structures,multiple vibration sources,and high speeds,it is very prone to component wear and even mechanical fault,which threaten the health of the machine.Moreover,due to the complexity and strong coupling of the vibration signal transmission paths,it is difficult for traditional fault diagnosis methods to meet the needs of online fault diagnosis in actual industrial sites.Fortunately,with the rapid progress of deep learning technology,automatically learning fault features through a large amount of data and performing online fault diagnosis has become an effective solution.This article focuses on how to introduce deep learning into the field of anomaly detection and fault diagnosis of diesel engines,and the following researches have been done:(1)Aiming at the problem of insufficient fault samples in practice,an anomaly detection model based on a one-dimensional convolutional auto-encoder is proposed.The error between the state model and the observation model is output through a one-dimensional convolutional auto-encoder,and the error is evaluated by the boxplot method,and the threshold for anomaly detection is determined.Finally,the experimental data is applied to verify the algorithm,and the results show the effectiveness of the proposed method in diesel engine anomaly detection.(2)The feature extraction performance of the stack auto-encoder is researched,and comparative analysis of various feature extraction methods is carried out by multiple evaluation indicators,verifying the superiority of the deep feature automatically extracted by the stack auto-encoder.In addition,with the help of Dropout technique,a fault diagnosis model for diesel engines based on stacked auto-encoder is built,and experimental results show that the proposed fault diagnosis method has higher accuracy than other traditional methods.(3)Aiming at the selection of hyper-parameters of auto-encoder networks,an improved variational auto-encoder is proposed.The harmony search optimization algorithm is introduced into the variational auto-encoder to realize the automatic optimization of model hyper-parameters.Then,the.improved variational auto-encoder is applied in diesel engine fault diagnosis under variable operating conditions.The experimental results show that the proposed improved method performs better than the original stack auto-encoder in non-steady operating conditions.And the accuracy is also higher than many other typical fault diagnosis algorithms.(4)The valve clearance fault data is collected by building up a diesel engine fault simulation experimental platform,and the features of valve faults on vibration signals are obtained by analyzing the vibration data.Other typical diesel engine fault features and corresponding diagnostic methods are also studied by analyzing actual engineering cases.