Fault diagnosis method for multi-source heterogeneous data based on improved autoencoder

In response to the difficulties in feature extraction and insufficient diagnostic accuracy of traditional fault diagnosis methods when facing complex multi-source heterogeneous data, this paper proposes a multi-source heterogeneous data fault diagnosis method based on convolutional autoencoder (CAE)-gated autoencoder unit (GAU). This method combines the advantages of CAE and GAU (CAE-GAU). Firstly, the multi-source data is preprocessed, including data cleaning, transformation, standardization, and normalization. Then, CAE is used to extract spatial features of the data. The input data is compressed into low dimensional hidden representations through convolutional and pooling layers. GAU further processes the hidden representations using gating mechanisms to highlight important features and suppress unimportant ones. Finally, the extracted features are fused with feature weighting, and the self attention mechanism is used for weight allocation to obtain the final data features. Through case analysis of multi-source datasets, the reliability and robustness of this method are verified. Besides, compared with various existing intelligent fault diagnosis methods, it can perform better. At the same time, it has stronger generalization ability and lower sensitivity to data distribution when dealing with multi-source heterogeneous data.