A Deep Learning Method For Rolling Bearing Fault Diagnosis Using Feature Mining And Feature Fusion

Fault diagnosis plays a key role in ensuring the safety of equipment operation and avoiding loss of personnel and property.The data-driven fault diagnosis method can extract deep features from the historical working data of the device without accurate mechanism model.As a powerful and effective data feature extraction method,deep learning has been widely favored by researchers in the field of data-driven fault diagnosis.The safety protection mechanism of the actual industrial site determines that it is difficult to collect a large amount of real fault data.At the same time,these fault data may be multi-source fault data collected by different sensors.If the single-source fault and multi-source fault data are not used effectively,it will inevitably result in a waste of a large amount of effective information in the fault data,which will affect the accuracy of fault diagnosis.This paper is devoted to fully using the effective information in the fault data,and carried out research on deep learning fault diagnosis methods based on fault data feature mining and multi-source data feature fusion,in order to achieve the full use of fault data on the feature level.The main work and innovations of this paper are as follows:(1)A deep learning fault diagnosis method based on data feature mining is proposed.This method realizes the feature mining of rolling bearing fault data by building a new data heterogeneous form dynamic waveform sequence,which makes the time series features and spatial neighborhood features appear simultaneously.Then according to the needs of feature extraction,CCLSTM fault diagnosis model is designed.Through this new heterogeneous form of feature extraction,deep learning fault diagnosis for fault data feature mining is realized.(2)A deep learning fault diagnosis method based on multi-source feature fusion is proposed.This method uses a progressive alternating-feature fusion method to achieve the fusion of multi-source data features.At the same time,a progressive training method for alternating fusion networks is proposed.A specific training strategy is used at different stages of alternating training in the fusion network to ensure that alternating training converges while accelerating the speed of network training.By using the asymptotic alternate feature fusion method to extract common features between multi-source fault data,deep learning fault diagnosis of multi-source fault data feature fusion is realized.(3)Design and implement deep learning bearing fault diagnosis software based on feature mining and feature fusion,and then introduce the design ideas,implementation framework and main functions of each functional module.This software makes it possible to apply deep learning fault diagnosis methods based on feature mining and fusion to reality.