As industrial production equipment and systems become increasingly complex, the safety risks and economic losses caused by system failures have escalated. Consequently, ensuring the stable operation of industrial systems has become a critical issue that requires urgent resolution. In recent years, deep learning-based fault diagnosis methods have garnered widespread attention due to their outstanding performance. However, such methods often exhibit “black-box” characteristics, lacking sufficient interpretability, which limits user’s understanding and trust in the decision-making process of the model. To address the aforementioned issue, this study proposes a Fault Logical Perception (FLP) model and utilizes this model to learn the Disjunction Normal Form Rule Set (DNFRS), thereby achieving interpretable classification of industrial faults based on logical rules. The FLP is characterized by its lightweight design and combination with differentiable \(L_{0}\) regularization. It automatically extracts concise and effective logical rules during training, which enhances the model’s transparency and practicality. Additionally, considering the characteristics of the sequence fault classification task, we propose an innovative data preprocessing and reconstruction method that performs feature augmentation on the original data, thereby further improving the model’s generalization performance. Experimental results on the Tennessee Eastman benchmark process demonstrate that the DNFRS learned by FLP effectively classifies faults while achieving a good balance between accuracy and interpretability, thereby validating its potential for application in industrial intelligent diagnosis.