Research On Fault Diagnosis Method For Rotating Machinery Based On 1-D LCNN

Traditional fault signal-based rotating machinery fault diagnosis methods are mostly "feature extraction + fault identification" mode.In recent years,deep learning technology led by Convolutional Neural Network(CNN)which relies on its automatic extraction of signal features,has achieved great success on the field of image recognition speech recognition.Therefore,this dissertation will draw on the successful application of CNN in these fields,from the perspective of single channel fault diagnosis,multi-channel fault diagnosis and fault diagnosis of variable working conditions.Combined with convolutional neural networks,the coping methods in these three scenarios are proposed.Main tasks as follows:(1)Aiming at the problem that two-dimensional convolution kernel in the classic convolution model LetNet-5 is not suitable for processing one-dimensional vibration signals,one-dimensional convolution neural network is proposed to process one-dimensional vibration signals.Secondly,according to the convolution network design principle that the larger convolution kernel size,the larger the receptive field is,one-dimensional large-size convolution kernel is used;In addition,since the hyperparameters in convolution network have a great influence on the network performance,so it is proposed to use the genetic algorithm to optimize the hyperparameters.Simulation analysis and experimental analysis show that the proposed method is better than the traditional fault diagnosis method.The hyperparametric optimization of genetic algorithm is better than the random selection of hyperparameters;the onedimensional large-size convolution kernel has better feature extraction effect than the twodimensional convolution kernel and one-dimensional small-size convolution kernel;(2)Aiming at the problem of low accuracy and low credibility of single channel information source for diagnosis of complex working conditions,a fault classification method combining 1-D LCNN and D-S evidence theory is proposed.First,using multiple channels of data to train multiple 1-D LCNN.Since multiple 1-D LCNNs are trained at the same time,the Adam optimization algorithm is used to train the network.Since the learning rate of the network has a great influence on the network training efficiency,the variable learning rate mechanism is introduced.In the test phase,the diagnosis result of multiple channel networks is used as multiple evidence bodies,and the evidence body is synthesized by D-S evidence synthesis rules,and synthetic evidence body is considered as the final diagnosis.The results of the experimental analysis indicate that the proposed method can obtain higher diagnostic credibility than the diagnosis by only a single channel;In the research of the noise immunity of the method,the method still has high fault diagnosis accuracy under the background of strong noise;In the comparison with other shallow network information fusion methods,the proposed method has better diagnosis results.It shows that the fusion of deep learning method and multi-information method has certain using value;(3)In view of the low diagnostic accuracy caused by the large difference in vibration data distribution under different working conditions,a fault diagnosis method for variable working conditions combining MMD and 1-D LCNN is proposed.In the network training phase,the Maximum Mean Discrepancy(MMD)is used to measure the data distribution difference between the source domain and the target domain,and this difference is regarded as one of the training objectives of the network.The sum of the Loss of the source domain convolutional network and the MMD item is taken as the total network training target,and the Gradient Descent(GD)algorithm is used to optimized the overall goal.The experimental results show that the proposed method has higher diagnostic accuracy than the direct use of 1-D LCNN.The visual results show that the MMD values in between source domain and target domain are constantly shrinking during the network training phase.The adaptation method is suitable for fault diagnosis of variable working conditions.