Research On Predictive Maintenance Method Of Rotating Machinery Based On Deep Learning

With the development of computer intelligence and modern automation,industrial equipment systems have gradually become larger,more sophisticated and automated.Rotating machinery is a common large-scale mechanical equipment system.Because the structure of these mechanical systems is becoming more and more complex,the cost required for equipment operation is also getting higher and higher.If the system fails to detect the failure in time,it will bring great Time loss and economic loss.In addition,excessive equipment maintenance will not only affect equipment operation time,but also generate excess costs.Predictive maintenance refers to the use of equipment real-time operation status information,environmental information and other data to judge,analyze,and predict the current state of machinery and equipment through data collection and condition monitoring of equipment,to determine the current type of equipment failure,Fault location,fault degree and cause of the fault,etc.,and then indicate the next development trend of the fault and the remaining service life of the equipment.Therefore,it is very important to identify mechanical failures early and apply this information to achieve rapid troubleshooting.Predictive maintenance of rotating machinery can effectively prevent mechanical performance failure and ensure the normal operation of mechanical equipment.Based on the deep learning theory,this thesis carried out a series of in-depth studies on predictive maintenance methods for rotating machines.The main work is as follows:(1)Aiming at the problems of long data processing time,poor real-time performance and poor accuracy of traditional fault diagnosis methods,a rotating machinery fault diagnosis method based on wavelet denoising deep convolutional neural network(WDCNN)was proposed.Firstly,the fault diagnosis model of rotating machinery based on WDCNN was established,and the fault diagnosis classification of rotating machinery was studied.Then,deep convolutional neural network(DCNN)is used to mine the fault features more deeply.Finally,Softmax activation function is applied in the output layer of the neural network as a feature classifier for training.According to the accuracy of training recognition and the fitness of the model,the optimal fault diagnosis structure is determined by continuously adjusting the DCNN network structure parameters,Dropout regularization parameter values and activation function selection,etc.The simulation results show that the proposed fault diagnosis method based on WDCNN can achieve high accuracy in fault diagnosis classification.(2)Aiming at the problems of slow convergence speed and low generalization ability of model training when Softmax activation function is used as classifier,a fault diagnosis method of rotating machinery based on WDCNN-SVM is proposed by integrating deep learning and support vector machine(SVM)algorithm.After deep feature extraction using deep convolutional neural network,deep feature is input into support vector machine classifier for training,and the accuracy of fault classification and model training time are measured as indexes.Simulation results show that compared with many other fault diagnosis algorithms,the proposed fault diagnosis method is superior to other fault diagnosis algorithms in the accuracy of fault classification and model training time,which improves the efficiency and stability of fault diagnosis.(3)Aiming at the problems of single model,poor data processing and large loss of life prediction in the residual life prediction method of rotating machinery,a PCA-LSTM residual life prediction method of rotating machinery based on multi-layer grid search was proposed.Firstly,the principal component analysis(PCA)algorithm is used to remove the redundancy of the index data and form new uncorrelated index data to reduce the complexity of the data.Then,the long and short time memory network(LSTM)model is used to predict the complete time series life data of mechanical equipment.Finally,the parameters of the LSTM model are optimized by using the multi-layer grid search algorithm.The simulation results show that the proposed method is superior to other life prediction algorithms in terms of accuracy and model training time,and improves the accuracy and stability of life prediction.(4)On the basis of the theory and algorithm research,the online predictive maintenance system for rotating machinery is built,which realizes the online fault state judgment and residual service life prediction of rotating machinery equipment,which is of great significance for the continuous,stable and reliable operation of rotating machinery.