Hybrid AI-Mathematical Modeling Approach for Predictive Maintenance in Rotating Machinery Systems

Abstract

Industrial and industrial environments place increasing pressures on rotating machine systems to be operational efficient and reliable — which drives increasing focus on predictive maintenance (PdM) strategic utilization. The motor, turbine, pump, and compressor systems are subjected to continuous mechanical stresses and are subject to similar wear and performance degradation and failures. In this paper a novel hybrid form works synergistically blending physics based mathematical modeling with the most advanced artificial intelligence (AI) for optimizing prediction maintenance. First a coupled second order differential equations that describe vibration dynamics and torque transmission as well as thermal interactions are developed for a comprehensive dynamic model of the rotating machinery under different operational loads. The physical model answers to what the system will behave like and how it should show up based on our baseline. At the same time, an AI module based on data, which employs a bidirectional long short term memory (BiLSTM) network to learn temporal pattern from real time vibration and temperature sensor data, is developed in parallel. A co-simulation strategy is used to achieve the hybrid model, wherein the outputs from the physical model are used as residual inputs for the AI network so that it can detect early anomalies and predict failures. The approach proposed is validated through the simulation studies and on an industrial real world employment in the thermal power plant with the systems of centrifugal pump. Experimental results indicate that with substantial improvement in fault detection accuracy, remaining useful life prediction and early warning capabilities to conventional physics (only) or AI (only) methods. The results of this research not only show the superiority of hybrid model for predicting the failures of rotating systems for predictive maintenance, but this also lays the groundwork for future developments of next generation digital twin frameworks for intelligent industrial asset management