Deep Learning Fault Diagnosis Based On Multi-Source Heterogeneous Data Fusion

Fault diagnosis is one of the key technologies to ensure the safe and efficient operation of key equipment in intelligent manufacturing process.In the absence of precise mechanism model,the fault diagnosis of key equipment in intelligent manufacturing process can be realized by analyzing the equipment operation status data.Because of its powerful feature representation ability,deep learning has attracted extensive attention of experts in the field of data-driven fault diagnosis research.The data acquired in actual industrial monitoring often include 1-D signal sequence,2-D image and 3-D video.The potential features of heterogeneous data collected by these different sensors may be related and complementary.Only using the same kind of data for deep learning in fault diagnosis will inevitably result in the waste of information contained in other data,which will affect the fault diagnosis accuracy.This paper designs the fusion mechanism of multi-source heterogeneous data under the framework of deep learning,realizes the data-level fusion and feature-level fusion of multi-source heterogeneous data,and fault diagnosis based on the features extracted after fusion,effectively improves the accuracy of fault diagnosis algorithm of deep learning,and realizes the innovation of fully utilizing data.The main work and innovations are as follows:(1)A deep learning fault diagnosis method based on data level fusion is proposed.According to the characteristic that the maximum eigenvalue of the data matrix can represent almost all the key information after the maximum singular value decomposition,a 1-D representation method of 2-D data is designed.The 1-D feature and the original 1-D signal sequence are fused at the data level.Then a deep neural network model is constructed based on the results of data level fusion to realize deep learning fault diagnosis of multisource heterogeneous data.(2)By designing an alternating optimization mechanism,a deep learning fault diagnosis method based on feature level fusion is proposed to realize feature extraction of heterogeneous data fusion.The initial features of 2-D data and 1-D data are extracted by convolutional neural network and stacked autoencoder deep network respectively,and the features extracted independently from two kinds of heterogeneous data are fed into the designed fusion network.Then,given the loss function,the feature extracted from the fusion network is taken as the initial value to further optimize the parameters of the convolutional neural network and the deep network of stacked autoencoder.The features extracted from the optimized two networks are then extracted through the fusion network.Repeat this alternately optimized fusion feature extraction process to achieve the goal of two types of feature extraction and fusion of global optimum,realize common feature extraction of multi-source heterogeneous data,and improve the accuracy of deep learning fault diagnosis.(3)The graphical user interface(GUI)of the deep learning fault diagnosis system based on multi-source heterogeneous data fusion is developed in Python environment,which provides the possibility for engineering test of fault diagnosis algorithm.