Research And Development Of Real-Time Condition Monitoring And Fault Diagnosis System For Rolling Bearings Based On Machine Learning

Mechanical failures seriously threaten the safe and reliable operation of large-scale mechanical equipment such as wind power generation equipment,aeroengines,and highend CNC machine tools.When the mechanical equipment is in normal operation,necessary measures are taken to monitor the operating status of the equipment,predict possible mechanical failures and give timely warnings,can greatly reduce the cost of production and improve production safety.The traditional bearing fault diagnosis is often to classify the faults of the bearings that have failed.In this regard,based on the research on bearing fault classification algorithms and fault prediction algorithms,this thesis develops a set of rolling bearing fault real-time diagnosis system that can be deployed on edge equipment,which can monitor,identify and predict rolling bearing faults in real time.This thesis focuses on the intelligent fault diagnosis and prediction of rolling bearings,and completes the following research:(1)The fault analysis method based on the characteristic frequency of bearing faults is studied by using the Fourier transform(FFT)algorithm combined with the envelope spectrum.In the time domain,frequency domain and time-frequency domain,the multidomain eigenvalues of the vibration signal are extracted through the time-domain feature index,wavelet packet algorithm,and short-time Fourier(STFT)algorithm.Use variance filtering and boxplot analysis for time-domain features for feature screening.At the same time,this thesis proposes a wavelet packet feature dimensionality reduction method based on MIC correlation analysis to filter features with low correlation with classification labels and reduce the space size of feature data sets.(2)Based on the wavelet packet feature extracted by the wavelet packet algorithm and the time-spectrum feature extracted by the STFT algorithm,this thesis proposes a fault classification method using the XGBoost model and the GoogLeNet model.Compared with other current intelligent fault diagnosis algorithms such as neural network and support vector machine,XGBoost algorithm has obvious advantages in time complexity and space complexity of the algorithm,and is very suitable for deployment on edge devices.The fault signal can be further diagnosed based on time-spectrum features combined with GoogLeNet network.This thesis also proposes an XGBoost bearing fault classification method based on TPE search algorithm optimization.Further improve the classification accuracy.(3)The traditional bearing failure prediction or life prediction is usually based on the classical stochastic degradation model,but the actual working conditions are often accompanied by complex noise signals,which affect the accuracy of the model prediction.In this thesis,an intelligent fault prediction method based on the time series forecasting model LSTM is studied.Compared with traditional model-based methods,the prediction error is greatly reduced.(4)Based on the research on bearing fault classification and fault prediction algorithm,this thesis develops a real-time diagnosis system for rolling bearing faults.The system includes three main modules: bearing fault monitoring,bearing fault prediction and bearing fault analysis.The system development architecture is developed as a web lightweight application based on the MVC architecture,and is deployed on edge devices based on Docker containerization technology.