We combine thermal-system physics and operating data to support faster design, diagnosis and control. Our methods span surrogate-assisted optimization, digital twins, energy baselines and model predictive control.
For compact R290 heat pumps, we combine system operating conditions, heat-exchanger models and CFD with surrogate-assisted geometry screening and inverse design. Heat transfer, pressure loss, refrigerant charge and manufacturability are evaluated together; physics-informed methods are investigated to reduce repeated analysis cost.
Related papers
In preparation
A heat-pump neural network and desiccant-rotor Fourier neural operator form a fast prediction model for waste-heat-regenerated EV HVAC. A supervisory receding-horizon controller coordinates compressor, fan and rotor operation under changing heat availability, cabin-comfort and humidity constraints.
Related papers
Real-Time Supervisory Control Based on Physics-Model-Trained Neural Surrogates for an Electric-Vehicle HVAC System with a Waste-Heat-Regenerated Desiccant Rotor
Under review
A shared physics-based heat-pump digital twin is adapted to individual vehicles through state and parameter estimation. We investigate refrigerant-condition and degradation diagnosis, vehicle-specific predictive control and learned occupant preferences, with a modular architecture for sensor feedback and future software updates.
Related papers
In preparation
We develop normal-operation baselines, uncertainty-aware prediction, anomaly detection and virtual-fault data for industrial utilities. Factory energy management systems (FEMS) combine these data-driven methods with physical energy balances to identify inefficient operation and evaluate plant-wide energy-saving options.
Related papers
Objective-Dependent Optimal Recovery Temperature for Condenser-Side Waste-Heat Recovery in a Semiconductor Cleanroom Chiller Plant
In preparation
Dynamic digital twins combine system physics with sensor measurements to estimate component condition. In lunar-habitat thermal-control studies, hybrid initialization and model-based inference help separate fault severity from changing thermal loads and operating conditions.
Related papers
Online System-Level Fault Severity Estimation for Lunar Habitat Active Thermal Control Systems via Sensor-Digital Twin Coupled Hybrid Initialization
In preparation