Interpretable Fault Diagnosis for Industrial Processes Based on Multi-Task Meta-Feature Learning

Traditional machine learning methods often present themselves as black-box models, unable to provide interpretability for the decision-making process, which is unacceptable in industrial processes. To overcome this issue, this work proposes an interpretable fault diagnosis method based on multi-task meta-feature learning (MT-MFL). By simultaneously learning fault meta-features and fault class differences, the model can generate more expressive and interpretable feature representations. Additionally, the effectiveness and feasibility of the proposed method are demonstrated through validation experiments conducted on a three water tanks model. The experimental results show that the proposed method outperforms traditional machine learning methods significantly in terms of fault diagnosis performance and result interpretability.