Artificial intelligence (AI) based predictive maintenance of waste heat recovery system

Abstract

Decarbonization and sustainability urge the deployment and utilization of distributed energy systems for high-efficiency gains. The dispatchable devices (gas turbines or diesel engines) are integrated with a waste-to-energy system to harness the energy lost or waste heat and support heat and cold loads. This paper investigates the characteristics of a waste heat recovery system and its performance degradation mechanism to assess its maintenance necessity and optimize maintenance frequency and the associated maintenance and downtime costs. The effectiveness of the waste heat recovery system (WHRS) is regularly estimated using the measured inlet and outlet parameters (flow and temperature) to identify the need for maintenance. The effectiveness changes not only with degradation but also with inlet conditions that deviate from the design conditions. Therefore, the operators are instructed to operate this system at the rated inputs and gauge its actual effectiveness. However, this approach did not provide much information on the root-cause parameters, i.e., the fouling formation and thickness in the shell and tube sides, which are quite important to decide the type of maintenance and the associated cost and duration. This paper studies the performance characteristics of the waste heat recovery system with reference to all critical and influential parameters (i.e., fouling thickness, heat transfer coefficients, and off-design inlet conditions) using a rigorous physics-based model. An AI model was developed using the derived performance characteristics to predict the fouling thickness estimation. The developed prediction model is able to accurately estimate the fouling thickness on the gas and water sides, and the error or deviation is within ±0.3 mm. By deploying this prediction model, the critical parameters can be monitored in real-time, and the performance degradation trajectory paves the way to understand degradation status and estimate the right maintenance time frame to schedule maintenance proactively, considering the maintenance cost and downtime effects.