Research On Rotating Machinery Fault Diagnosis Method Based On Multi-source Data Fusion

Rotary machinery plays an increasingly important role in industrial production,such as aeroengines,centrifugal compressors,industrial gas turbines,wind turbines,transmissions and steam turbines,especially in petroleum,chemical,coal,electric power,equipment manufacturing and other industries.As a widely used equipment in the industrial system,the failure of rotating machinery will result in the restriction of enterprise production efficiency,the reduction of product quality,the decline of enterprise market competitiveness,huge economic losses.and even irreparable casualties.With the continuous progress of modern science and technology,rotating machinery in industrial systems is becoming increasingly large,complex,multi-functional and intelligent.Besides,the working environment of rotating machinery is usually harsh and complex.In the industrial system with harsh operating environment and high coupling,it is difficult to achieve high accuracy fault diagnosis of rotating machinery through a single signal source.Therefore,the fault diagnosis method based on multi-source data fusion is widely used in rotating machinery.In this paper,for the background of rotating machinery fault diagnosis based on multi-source data fusion,a fault diagnosis method based on self-feature(feature of single source data)fusion and mutual-feature(feature of multi-source data)fusion is studied.The main research content of this paper is summarized as follows:(1)Aiming at the problem that it is difficult to fully extract self-features of rotating machinery in the single sensor scenario,a self-feature fusion fault diagnosis method based on convolutional neural network is proposed.Based on one-dimension and two-dimension convolutional neural networks,the different networks is used to extract features from a single signal source data.The method combines the advantages of two kinds of network structures,which can extract one-dimensional features of signals and correlation features of signals in non-adjacent regions.Then the features are fused for fault diagnosis.The performance of the proposed method is verified using bearing fault data set.Experimental results show that the proposed method can effectively extract the one dimension and two dimension features of vibration data,and the diagnosis accuracy is greatly improved.(2)Aiming at the problem that mutual-features of rotating machinery are difficult to be fully extracted in multi-sensor scenario,a new fault diagnosis method based on mutual-features fusion is proposed,which is multi-dimensional concatenated convolutional neural network based on attention mechanism.In this method,one-dimension and two-dimension convolutional neural networks based on attention mechanism are used to extract the complementary fault features and the correlation between different features.The attention mechanism can assign different weights to features in order to emphasize important features and suppress non-important ones.The bearing fault data set was used to verify the performance of the method.The experimental results show that there are complementary fault features between vibration data and torque data,and the proposed method can extract and effectively fuse these complementary fault features,which greatly improves the performance of bearing fault diagnosis.(3)Aiming at the problem of insufficient fault information mining in multi-source data of rotating machinery,a fault diagnosis method based on self-feature fusion and mutual-feature fusion is proposed,and a bilinear fusion model based on mutual attention mechanism isestablished.Firstly,the features extracted from different modal(time domain and time-frequency domain)data of single signal source interact with each other through mutual attention mechanism,and then these features are concatenated together to achieve self-feature fusion.Then,the bilinear model is used to fuse the features extracted from different sensors,in which the element multiplication method is used to realize the fine-grained feature fusion to complete the mutual-feature fusion.The combination of self-feature fusion and mutual-feature fusion can fully mine fault features in multi-source data and perform fine-grained fusion.The performance of the proposed method is verified on the bearing fault data set.The results show that the proposed method can effectively extract fault features from multi-source data and fuse them,which greatly improves the performance of fault diagnosis.(4)Aiming at the problem of weak complementarity of features extracted from multi-source data of rotating machinery,a multi-head attention mechanism based differentiation feature extraction and fusion fault diagnosis method is proposed.A dual-input and dual-output network structure was adopted in this method.Firstly,multi-head attention mechanism improved by one-dimension convolutional neural networks is used to extract deep features of multi-source data,and self-feature fusion of single-source data was realized by multi-head feature concatenation.Then the features of the multi-source data are input to the feature source classifier to distinguish the source of features,so that the extracted features contain complementary fault information as much as possible.Finally,the bilinear model is used to fine-grained fuse these features to realize the mutual-feature fusion of multi-source data and input them into the fault classification classifier to complete the fault classification.In this method,the complementary fault features were extracted by classification task,instead of relying only on feature extraction network,which made the extracted features more complementary.The performance of the proposed method is verified on bearing fault data sets.The results show that the proposed method can effectively extract complementary fault features from multi-source data and fuse them,which greatly improves the accuracy of fault diagnosis.