Research On Fault Diagnosis Method Of Rotating Machinery Based On Improved Deep Residual Network

Fault diagnosis and life prediction of rotating machinery are of great significance for ensuring the safe operation of machinery and equipment.Vibration signals of rotating machinery have typical non-stationary and non-stationary characteristics.In the strong background noise working environment,the noise is easy to drown the weak fault features,which increases the difficulty of fault diagnosis.In this paper,based on deep learning technology,the intelligent recognition of complex fault modes and the prediction of remaining useful life of rotating machinery are realized by using the improved deep residual network.The main contents of this paper are as follows:Firstly,an improved residual network model is proposed to solve the problem that the deep learning-based fault diagnosis accuracy and computing efficiency speed decrease caused by the interference of strong background noise and the proliferation of neural network layers.On the one hand,a multi-scale feature fusion block is designed in the first layer of the network model,which can extract multi-scale fault feature information from the signals and enhance the capability of model feature extraction.On the other hand,the depthwise separable convolution and the optimization of residual blocks are used to reduce the computational burden of the network.Through experimental analysis,it is verified that the proposed network model has higher recognition accuracy and higher operating efficiency.Secondly,a fault pattern recognition method for one-dimensional multiscale residual networks is proposed.In this method,one-dimensional time-domain signals are used as input,and through four sets of convolution kernels of different scales,deeper fault feature information can be mined,and the feature information can be fused together in the residual block for unified learning.In addition,the multi-scale feature fusion technology is used in the residual block to further improve the ability of multi-scale information extraction.An experimental study on complex fault pattern recognition of a gearbox under multiple load conditions was carried out,and the anti-noise and generalization performance of the proposed method were verified.Finally,the residual block of the one-dimensional depthwise separable convolutional optimization temporal convolutional network was proposed.Based on this,the remainning life prediction method of rolling bearings was established based on the multi-scale temporal convolutional network.Taking the temporal convolutional network as the basic model,the dilatation rates of different scales in the convolutional kernel were set inside the residual block,and then the multi-scale time-series feature information in the learning signal was fused,finally the remaining life prediction of rolling bearings was realized.The accuracy of the proposed method for bearing life prediction was verified by the analysis of PHM2012 accelerated life experiment data of rolling bearings.