Equipment Fault Diagnosis And Prediction Technology Based On Temporal Convolutional Network And Self-attentio

In present,mechanical equipment is developing towards intelligence,complexity,and precision.The failure of any component will lead to a catastrophic failure of the system,which highlights the importance of system reliability.Once the system breaks down,it will cost expensive maintenance costs and suffer production losses due to downtime.In the worst case,it may cause equipment parts to fall off and disintegrate,causing serious casualties.Therefore,the prognostics and health management(PHM)of the equipment is very important.This paper aims to study the key technology in PHM: the fault diagnosis and the remaining useful life prediction technology.First of all,considering that the default weight coefficient of the residual connection in temporal convolutional network(TCN)is too large,when the residual block is shallow,the feature extraction ability of the residual block will be inhibited.This paper proposes an adaptive residual coefficient assignment(ARCA)unit,establish an adaptive residual coefficient allocation temporal convolution network(ARCATCN).The ARCATCN is used for fault identification feature extraction and deep degradation feature extraction to enhance the identification of fault features and highlight the trend of degradation features.Secondly,in the current fault diagnosis research,only the vibration signal collected from a single location is used for fault diagnosis,which contains incomplete fault information.This paper proposes a multi-view feature fusion fault diagnosis method based on ARCATCN and self-attention mechanism.For the first time,the acceleration signals collected from different locations are used as the source of multi-view sources,which can make the fault features contain more comprehensive fault information.ARCATCN is used to extract deep-level fault features from different view information.Considering the contribution differences of different view features to the diagnosis results,the proposed method utilizes self-attention mechanism for adaptively assigning weights to different views,thereby enhancing the contribution of important views to RUL prediction.Finally,the mapping relationship between multi-view fusion features and labels is realized through the classification function.Finally,in the current remaining useful life(RUL)prediction research,the contribution differences of different channels and different time steps to RUL prediction are not considered,and only use deep learning features or handcrafted statistical features for prediction.These limitations can lead to inaccurate prediction results.To solve these problems,this paper proposes an RUL prediction method based on multi-layer self-attention and ARCATCN.The MLSA is designed to adaptively assign feature weights.The ARCATCN is used to extract deep learning features.Dilated convolution and residual connection are adopted in ARCATCN structure.Dilated convolution is an efficient way to widen receptive field,and the residual structure can avoid the gradient vanishing problem.Besides,we propose a feature fusion method to fuse deep learning features and statistical features.And Finally,the mapping between fusion features and RUL is realized through the regression layer.The fault diagnosis method and the RUL prediction method proposed in this paper are respectively verified using public datasets,multiple comparative experiments are carried out,and the model prediction process is visualized.The results show that the proposed fault diagnosis method can enhance the identification of fault features and improve the accuracy of fault diagnosis.The proposed RUL prediction method can highlight the trend of degradation features and enhance the accuracy of RUL prediction.