A novel physical information-guided predictive maintenance method for chillers

The chiller is a critical component of the HVAC system, characterized by high energy consumption and frequent failures. Consequently, implementing predictive maintenance (PdM) is essential for improving system reliability, reducing energy consumption, and lowering maintenance costs. However, existing PdM approaches are limited in their ability to predict the remaining useful life (RUL) of complex systems over extended periods, and they often fail to account for the degradation of energy efficiency when formulating maintenance strategies. To this end, this paper proposes a PdM framework for chillers guided by physical information, aiming to improve the RUL prediction accuracy and balance energy efficiency losses with maintenance costs. First, a semi-empirical model is developed to quantify the impact of degradation on energy efficiency. Energy efficiency indicators are derived by mapping the relationship between energy efficiency and degradation features. The elbow rule is then applied to identify the critical points of energy efficiency degradation, which enables more accurate RUL estimation. Subsequently, the accuracy of RUL predictions is enhanced by incorporating physical consistency constraints. Finally, a maintenance strategy optimization model is formulated to minimize both energy waste and maintenance costs. The proposed method is validated using nearly five years of operational data from a large-scale HVAC system, demonstrating substantial improvements in prediction accuracy and maintenance strategy optimization.