Intelligent Fault Diagnosis and Resilient Control of Safety-Critical Industrial Systems Using Explainable AI
Main Article Content
Abstract
Industrial systems that are safety-critical have very high reliability and safety criteria, and the effects of a failure may be disastrous, both economically, environmentally, and in terms of human lives. The conventional fault diagnosis and control methods cannot usually cope with the uncertainty, nonlinearity, and cyber-physical complication of the modern industrial environments. Recent progress in artificial intelligence has made it possible to perform fault diagnosis with data and sustainable control, but due to the characteristics of the black-box of most artificial intelligence models, it cannot be applied in controlled and risky areas. The following paper suggests an integrated intelligent fault diagnosis and resilient control framework based on the Explainable Artificial Intelligence (XAI). The framework integrates real-time data acquisition, fault detection and classification using AI and explainability layer, among other means, to make decisions about the transparent one and the fault-aware control strategies. The presented approach is related to the high level of operator confidence, regulatory compliance, and enhanced system resilience by incorporating explainability into the diagnostic and control loop. The framework is tested in a simulation-based case study of situation of an industry process that is important to safety in several conditions of faults such as sensor fault, actuator fault and process fault. The outcomes indicate that the accuracy of fault detection, recovery time, and system stability are improved when using the traditional and black-box AI-based methods. Further, the explainability layer offers predictable and interpretable results on the causes of fault and the influential system variables, enabling proper human-AI interaction. The paper provides the prospects of resilient control facilitated by XAI as a major enabler of credible industrial AI implementation.