Abstract Fault diagnosis is crucial for complex system health management. The intelligent fault diagnosis model based on belief rule bases (BRB) effectively handles small sample data and uncertain information. However, due to the limitations of expert knowledge, simply utilizing this knowledge to construct initial rules may not fully capture the highly nonlinear relationships between inputs and outputs in complex systems, resulting in reduced model accuracy. While optimization algorithms can dynamically adjust the parameters and structure of the rule base to improve model accuracy, the iterative process of seeking an optimal solution can inevitably compromise the model's interpretability. To address the aforementioned issues, a fault diagnosis model based on interpretable BRB with fault tree analysis (FTA) has been proposed. In this model, initial rules are constructed through a structured representation of fault relationships using FTA. Additionally, interpretability constraints are introduced during optimization to limit the range of parameters, ensuring that the optimization results align with real-world situations. A case study on a CNC milling machine was conducted to validate the proposed method. The results indicate that the model identifies fault states in complex systems more effectively and accurately than existing models, maintaining a balance between accuracy and interpretability throughout the diagnosis process.
Liu et al. (Sat,) studied this question.