Research on Online Monitoring Technology of Refrigerant Charge in Airborne Liquid-Cooled - Evaporative Refrigeration System

Liquid-cooled evaporation refrigeration systems are an efficient thermal management solution widely employed in aircraft applications. The refrigerant charge amount significantly influences the system's performance and operational stability. Therefore, research on online monitoring technology for refrigerant charge quantity is of significant importance. This study proposes an online monitoring model based on multiple machine learning algorithms for the prediction and evaluation of refrigerant charge in liquid-cooled evaporation refrigeration systems, ensuring their efficient and stable operation under diverse operating conditions. A one-dimensional model of the aircraft liquid-cooled evaporation refrigeration system was developed on the AMESim platform using numerical simulation methods. Based on this model, the influence of varying refrigerant charge levels on key performance indicators was systematically analyzed, and the optimal charge level was identified. A series of simulation experiments were conducted to collect data, resulting in a comprehensive dataset comprising 798 fault-related samples. After data preprocessing, feature selection was performed, leading to the identification of an optimal feature subset comprising five variables, thereby enhancing both the computational efficiency and accuracy of the model. Subsequently, four regression models for refrigerant charge estimation were developed using the optimal feature subset, and these individual machine learning models were integrated via the Stacking ensemble method. The final trained Stacking ensemble model achieved a significant reduction in prediction error. The results demonstrate that the online monitoring model developed using the Stacking ensemble method can be effectively applied to the task of monitoring refrigerant charge in aircraft liquid-cooled evaporation refrigeration systems, thereby ensuring long-term operational stability.