Deep Data-driven RUL Prediction for Optimizing Safety-constrained CBM Strategies in Plant Modernization

Systems in commercial utility facilities and advanced manufacturing industries suffer from gradual deterioration due to environmental factors, operational stresses, and aging effects. As failures caused by these degradations could result in serious financial, safety, and environmental consequences, critical systems must implement advanced artificial intelligence (AI) techniques for managing, performing, and modernizing maintenance tasks in the plant. Specifically, self-diagnosis and health monitoring technologies must be integrated into plant maintenance strategies to improve the overall facility’s operating safety, availability, and performance. InfoBeyond advocates the deep data-driven AI technique for learning, modeling, and predicting Remaining Useful Life (RUL) for optimizing safety-constrained condition-based maintenance (CBM) policies to address the above technical gap in modernizing plant maintenance capabilities. The proposed solution results in a non-destructive CBM policy optimization with dynamic-scheduled inspection and plant safety constraints by interacting with a deep-learning RUL prediction algorithm. Specifically, attention-based deep RUL learning and prediction algorithm integrate with classic data-driven feature engineering technique improve the learning accuracy and further effectively determines the combination of time steps, which has a larger influence on the final RUL prediction. Additionally, the safety-constrained CBM optimization and strategy model is developed that considers constraints of plant safety, imperfect measurement, and dynamic scheduling inspection for demonstrating the CBM decision-making and policy optimization. Use Cases based on real-world machine sensory datasets are developed to demonstrate the conceptual effectiveness and practicability of the proposed solution in real-world RCM (reliability-centered maintenance) analysis.