Bridging gray-box modeling and machine learning: A digital twin approach to refrigeration system identification and predictive maintenance

This paper investigates the application of digital twins for refrigeration systems, focusing on system identification and predictive maintenance. The work explores the potential of thermodynamic modeling, using real physical parameters including thermal resistance, thermal capacity, and heat evaporation coefficient. The model’s ability to detect changes in system parameters with minimal data is confirmed. The methodology includes a thermodynamic digital twin model developed in MATLAB, built for refrigeration system data. Traditional machine learning techniques were used for system identification and robust results validation. By embedding the compressor’s hysteresis‐based control logic, the closed-loop digital twin delivers accurate, high resolution real-time temperature predictions and paves the way for advanced disturbance detection and model-based control. Furthermore, the model successfully detected anomalies during an experiments with insulation degradation through cyclic door openings. These insights provide a foundation for advancing the use of digital twins in refrigeration system monitoring and maintenance