Research On Data Driven Fault Diagnosis Method Based On Machine Learning

With the complication and enlargement of modern equipment,the abnormal detection and fault diagnosis of the system has always been the focus of academic attention.If system failure cannot be discovered and dealt with in time,it will cause huge economic losses and casualties.If an early fault can be detected and the alarm is isolated,the abnormal accident will be effectively avoided.Therefore,reasonable fault diagnosis of complex systems has become a key tool.At present,the research hotspot of fault diagnosis is mainly the data-driven fault diagnosis method based on intelligent learning method.However,the traditional intelligent learning method can not fully exploit the hidden fault feature information in the data,and there is a problem of insufficient approximation accuracy,and the parameters in the fault diagnosis model Uncertainty exists in a large amount,resulting in large fluctuations in fault diagnosis accuracy and insufficient accuracy.To this end,this paper proposes the implicit fault feature extraction method based on machine learning algorithm and the fault diagnosis model integrated with multiple machine learning algorithms from the perspective of fault diagnosis feature extraction and intelligent fault diagnosis model construction,in order to improve the accuracy of fault diagnosis.Based on the data set in UCI machine learning database,the above methods and models are analyzed and verified,and the effectiveness of the proposed method is verified by the example of fault diagnosis of injection molding machine.The main work of this paper is as follows:(1)An implicit fault feature extraction method based on XGBoost algorithm is proposed.The deep feature extraction ability of this method is studied.The accuracy of the diagnostic prediction on the test data are analyzed by using the SVM,neural network and random forest fault diagnosis model constructed before and after the extracted implicit fault features,in order to verify the effectiveness and versatility of the method.(2)A fault diagnosis model based on LightGBM algorithm is proposed.This paper studies the classification and recognition ability of the model in the context of complex data,and compares the fault diagnosis performance of the diagnostic model with several traditional fault diagnosis models and deep learning models.Under the different parameters,the change of the fault diagnosis and recognition ability based on LightGBM algorithm is studied.Considering the influence of parameter uncertainty on the diagnostic performance of the model,the Gaussian process and acquisition function of Bayesian optimization algorithm are improved,and the improved BOA-LightGBM fault diagnosis model is proposed.The diagnostic prediction performance of improved BOA-LightGBM fault diagnosis model and grid search LightGBM fault diagnosis model and random search LightGBM fault model are discussed.(3)The machine learning algorithm is applied to the fault diagnosis of the tie rod and the XGB-BOA-LightGBM model is used to diagnose and identify the fault,in order to solve the real-time fault diagnosis of the tie rod.The diagnosis effect of the tie rod fault under different data characteristics are discussed and analyzed.The real-time diagnosis and prediction effect of the XGB-BOA-LightGBM fault model on the tie rod fault is analyzed.