A novel explainable fault diagnosis model for homogenization process using probabilistic Boolean network

Ensuring the stability of homogenization processes (HS) is crucial in industrial production, thereby heightening the urgency for effective fault diagnosis. Existing fault diagnosis methods suffer from high complexity and low interpretability of the processing steps. Boolean networks (BNs) possess unique advantages in addressing the aforementioned issues. Firstly, system complexity is reduced by representing the device state as a binary node. We establish the HS dynamic model by integrating Boolean logic with probability parameter estimation, which are employed for fault diagnosis of industrial processes. The normal state transitions of HS processes are constructed by applying the attractor cycles principle of BNs. Subsequently, a novel concurrent system is established by using the bijective property of semi-tensor product (STP) and probabilistic Boolean networks (PBNs) for fault diagnosis. The interpretability of diagnostic results is improved by integrating the graph theory and state transition mechanism. Finally, experiments validate the effectiveness of the proposed method.