Research On Fault Diagnosis Algorithm Of Mine Equipment Bearing Under Small Sample

Mining equipment is a key factor in ensuring the safe and efficient extraction of coal.As one of the most important components of mining equipment,bearings are easily prone to failure.Existing bearing failure diagnosis work mainly depends on sufficient sample data,but in the complex mine environment,useful fault signals for model training are difficult to capture,and there are problems such as insufficient or imbalanced sample data,resulting in inability to perform precise diagnosis.In this article,a bearing failure diagnosis model for mining equipment based on small-sample learning and transfer learning is constructed to improve the diagnostic performance under small-sample data.The main research work includes the following:(1)To address the issue of insufficient fault samples in a single class of mine equipment bearings,a small sample bearing fault diagnosis model based on the improved siamese network is proposed,which transforms the complex classification problem into a similarity difference problem by converting the traditional convolutional structure into a multi-scale convolutional structure.On the basis of the siamese network,multi-scale convolution is introduced to enhance the perception ability of convolution layers.First,constructed pairs of similar and dissimilar fault samples are input into a shared parameter feature extraction network to be mapped to a low-dimensional feature space.Second,the Euclidean distance is used to calculate the similarity between sample features,further achieving the classification of test samples.The results show that the proposed model has a higher accuracy and a smaller total number of parameters under small samples.(2)To address the imbalanced sample problem of fault samples in mining equipment with multiple types of bearings data sets,an improved small-sample bearing fault diagnosis model combined with transfer learning based on Relation Net is proposed,in which the similarity of each layer of feature vectors is measured in the relationship network to enhance the classification effectiveness of the model.First,fault data samples from the source domain and target domain are constructed and input into the relationship network for feature extraction and similarity calculation,producing enhanced similarity scores.Second,the source domain and target domain are aligned in the Hilbert space,and the distance between feature vectors is calculated using the maximum mean difference to achieve domain transfer.The results show that the proposed model has good transfer and classification effectiveness in the absence of target domain samples.In summary,this thesis proposes a mine equipment bearing fault diagnosis model suitable for small samples based on improved siamese network and Relation Net,and designs and develops a mine equipment bearing fault diagnosis prototype system to validate the effectiveness of the proposed model.This thesis has 38 figures,16 tables,and 82 references.