Multi Source High Frequency Time Series Data Analysis For Equipment Fault Diagnosis

In recent years,due to the continuous expansion of the scale of industrial equipment and the continuous increase in the magnitude of the signals collected by sensors,equipment fault detection and fault diagnosis have gradually entered the era of big data.In the multidimensional and massive industrial data collection,based on traditional statistical analysis methods It is difficult to effectively identify equipment failures.The deep learning algorithm has a strong ability to process data.Among them,Convolutional Neural Network(CNN)has become a common algorithm in fault diagnosis because of its special hierarchical structure of convolution and pooling,as well as powerful data mining capabilities and data fusion capabilities.However,for high-frequency timing signals,there are a large number of similar redundant signals in the signals,which affect the identification of fault information,and the CNN network has a limited sensing range of timing signals,and cannot effectively sense the timing characteristics in the signal,and most of the current research is based on A single signal source is used as the input of the diagnostic model,and it is difficult to comprehensively reflect the status of the device.In response to the above problems,this paper improves the traditional CNN to improve the accuracy of fault diagnosis.The main work of the thesis is as follows:(1)A CNN fault diagnosis method based on timing compression enhancement is proposed.Aiming at the problems of a large amount of redundant information in high-frequency time series signals and the inability of CNN to effectively extract time series features,this method first compresses and reduces the dimensions of the signal through cascaded pooling,removes redundant information in the signal,and then uses the first layer of large size The convolution kernel increases the network’s perception range of time series signals,and compresses longer time series signals as input.It can not only extract the data characteristics of the signal itself,but also sense more time series characteristics,increasing the accuracy of recognition.Finally,it is verified by experiments that the method has a higher recognition accuracy.(2)A CNN fault diagnosis method based on multi-channel weight adaptive is proposed.Aiming at the problems that it is difficult to fully extract the device status information under a single source signal,and CNN cannot effectively achieve accurate recognition and classification based on the importance of multi-channel features,this method first normalizes different data sources as different channels of CNN Using the characteristics of the convolutional network to realize the feature fusion and extraction of different data sources.Then the SENet module is embedded in the CNN network,and the weight of each channel is learned adaptively to improve the accuracy of classification.Finally,through comparative experiments,it is proved that this method can further improve the accuracy of fault recognition.