Research On Typical Fault Diagnosis And Unstable Conditions Evaluation Of Diesel Engine

As an important power machine,diesel engine has the characteristics of many parts,complex movement and harsh working environment,which will directly affect the smooth operation of the task.Once a breakdown occurs,economic losses will be caused by the suspension of production,and serious casualties will result.Therefore,research on the fault diagnosis technology and condition assessment methods of diesel engines is of great significance to ensure the normal operation of diesel engines.In the actual work of diesel engines,due to noise interference and changing working conditions,traditional feature extraction and pattern recognition methods need to manually extract and screen features.It has been difficult to adaptively solve the problem of fault diagnosis.Based on this,this thesis takes diesel engines as the object,aiming at solving the problems of lack of fault data and inaccurate fault-sensitive feature extraction.Convolutional neural network is introduced to extract features adaptively,data augmentation and transfer learning are used to make up for the shortage of data,and then typical fault diagnosis methods are studied.In addition,for the lack of diesel engine unstable conditions assessment methods and other issues,relevant evaluation studies were conducted.Finally,these methods are verified on the actual engineering data.The main content of the paper is as follows:(1)The mechanism-based data augmentation and improved deep convolution network method for diesel engine fault early warning and diagnosis are studied.Firstly,based on the characteristics of the fault mechanism,vibration images of misfire and valve faults are generated respectively.Then,the improved ResNet is used to train and make the fault database.Finally,the threshold method and SVM method are used for verification respectively.(2)The diesel engine fault diagnosis method based on transfer learning is studied.Firstly,the pre-trained AlexNet model is used to train and diagnose faults under 5%?80%data.Then,it is compared with CNN and EMD+SVM to verify the effectiveness of the method.Finally,the rationality of transfer learning is explained through visualization.(3)The start-up evaluation method of diesel engine based on grey evaluation is studied.Firstly,the Teager operator and impact location are used to adaptively capture the high-pressure gas impact region of each cylinder in each cycle during the start-up phase.Then,the feature that can characterize the working force of the high-pressure air are extracted.Finally,the grey evaluation method is used for comprehensive evaluation,which is verified by actual test results.