Research On Mechanical Bearing Diagnosis Method Based On Deep Learning And Information Fusion

As one of the essential components in rotating machinery,rolling bearings have been widely used in daily life and production.Usually,rolling bearings are installed in rotating machinery to support their rotation,and their health status is directly related to the working efficiency and safety performance of mechanical equipment.Due to the complexity of the actual working environment of rolling bearings,faults often occur.If faulty bearings cannot be detected and replaced in a timely manner,it may cause unimaginable damage to the entire equipment.Therefore,for mechanical equipment,it is crucial to establish an algorithm that can explore the complex mapping relationship between the raw signals and fault modes of rolling bearings.Early fault diagnosis methods relied too much on expert prior knowledge and required manual feature extraction,followed by machine learning to identify and classify the fault modes of rolling bearings,resulting in low fault recognition accuracy.In response to the above issues,this paper combines the convolutional neural network model in deep learning with information fusion strategy,which enables the network model to have stronger representation learning and information processing capabilities.The main research contents include:(1)Firstly,the research background and significance of this paper are explained,and the commonly used rolling bearing fault diagnosis related algorithms are summarized and analyzed,and divided into two modules of shallow learning and deep learning.The structure,fault types,and vibration signal characteristics of rolling bearings are introduced and analyzed.In addition,this paper points out some of the main problems in the current diagnostic methods and proposes corresponding solutions.(2)In response to the problems of the single model structure,ineffective feature extraction of fault characteristics,and low diagnostic accuracy in deep learning,a dualchannel deep learning network is considered to extract fault characteristics,and combined with a shallow learning network for classification to improve diagnostic accuracy.The model has two parallel CNN structures,an SSA-optimized SVM classifier,and a feature fusion strategy.The dual-channel CNN automatically learns and extracts features from the original vibration signal.The feature fusion strategy is used to input the fused features into the SSA-SVM classification layer for classification.The strong global search capability of SSA is used to optimize SVM to ensure the reliability of the model.Compared with some classical feature extraction and classification algorithms,the model has high fault recognition accuracy without relying on prior expert knowledge for feature extraction.(3)In response to the problem of traditional CNN network structures relying on a single input feature and having difficulty in obtaining comprehensive and detailed fault information,improvements have been made based on the model in Chapter 3.A fault diagnosis method based on time-frequency dual-channel CNN is proposed.In the1 DCNN,the original one-dimensional signal is replaced by the FFT frequency spectrum obtained through FFT,which has faster processing speed and does not require artificially set related parameters.In the 2DCNN,the input two-dimensional data is transformed from a grayscale image formed by matrix arrangement into a wavelet timefrequency image obtained through wavelet transform,which can more comprehensively extract feature information.The t-SNE dimensionality reduction technique is used to display the model classification effect,and the feasibility and effectiveness of the model are verified from the perspectives of the loss function and recognition accuracy of different methods.Finally,it is concluded that the algorithm model has good classification performance for rolling bearing faults.