Research On Fault Diagnosis And Life Prediction Of Rotating Machinery Components Based On Deep Learning

With the rapid development of modern manufacturing,large rotating machinery and equipment are widely used.Its reliability is of great significance for safe and orderly production.However,due to long-term continuous operation and variable operating conditions,the gear,bearing and other transmission components of rotating mechanical equipment are extremely prone to various failures.If the damaged parts are not found and repaired in time,the chain reaction of the whole mechanical system will be caused.In serious cases,the mechanical system may suddenly fall into a paralyzed state,affecting normal production and even causing serious casualties.Therefore,effective and accurate fault diagnosis and remaining useful life prediction of important parts of rotating machinery can ensure safe production,reduce maintenance costs and avoid economic losses.At present,due to the rapid development of computer science and sensor technology,artificial intelligence,especially deep learning,has attracted widespread attention.Therefore,applying deep learning to fault diagnosis and life prediction of rotating machinery components is a very promising research direction.This article studies the key technologies of fault diagnosis and residual service life prediction for key components of rotating machinery equipment based on deep learning,aiming to improve the accuracy and robustness of the model.The main research content and innovation points are as follows:Aiming at the above problems,this paper studies the key technologies of fault diagnosis and remaining useful life prediction of important components of rotating machinery based on deep learning,aiming at improving the accuracy and robustness of the model.The main research contents and innovations are as follows:(1)Because existing fault diagnosis methods require a large amount of labeled fault data to operate effectively,and the signals collected in practical work are prone to noise interference,and traditional intelligent fault diagnosis models have problems such as relatively low accuracy for composite fault diagnosis,a semi-supervised fault diagnosis method based on feature pre-extraction mechanism and improved generative adversarial network is proposed.This method transforms the original one-dimensional vibration signal into two-dimensional time-frequency map through the feature pre-extraction mechanism of wavelet transform,and then marks and inputs the limited samples into the improved generation countermeasure network model for fault diagnosis.The proposed method is verified on rolling bearings and gear boxes.The results show that the proposed method has good robustness and realizes high-precision fault diagnosis of rotating mechanical equipment.(2)In practical fault diagnosis,the specificity of time-varying rotational speeds means significant non-uniformity in time,which can lead to significant differences in the distribution of training data and test data.However,existing fault diagnosis methods are often only effective for specific operating conditions.To solve this problem,an unsupervised fault diagnosis method based on deep transfer learning is proposed.This method improves the traditional residual network and enhances the ability of the model to extract transferable features.In addition,by removing the confrontation training mechanism,the common fault diagnosis model based on the domain confrontation neural network and the global maximum mean difference is improved,making the diagnosis model simpler and more efficient.The performance of the proposed method is analyzed by two experimental cases with timevarying speed.The results show that the method can accurately diagnose faults in the target domain using source domain knowledge.(3)Aiming at the problems of poor generalization and insufficient prediction accuracy of current remaining useful life prediction methods,a high-precision prediction method based on improved convolution neural network(CNN)and gated cycle unit(GRU)is proposed.This method first solves the limitations of manual feature extraction.By integrating one-dimensional depth separable convolution neural network and twodimensional transposed convolution neural network,the extracted feature representation is more obvious,and a new loss function is introduced.Then,soft threshold and residual connection are inserted into the attention mechanism to help improve the robustness and prediction accuracy of the model.Finally,the effectiveness of the proposed method is verified by the data set of turbofan engine and IEEE phm2010.