Abstract:
This work proposes a machine-learning framework for developing digital twins of evaporators, crucial components within heat pumps, by training nonlinear regression models with highfidelity unsteady CFD data. This is applied to a plate-Tube type heat evaporator within a single-stage heat pump utilizing R134a refrigerant to recover heat from waste water. The independent variables are chosen as the refrigerant and water inlet velocity and temperature values, and the R134a evaporation pressure under transient conditions. We identify the dynamic heat transfer process in the evaporator with three objective functions i.e., dependent variables: heat transfer coefficients on both the water and refrigerant sides, and dryness fraction at the outlet of the refrigerant flow domain. The Latin hypercube data-sampling method is employed to generate a data space along the selected independent variables. Reynolds-Averaged Navier-Stokes filtered flow equations, coupled with the Volume of Fluid method for modeling phase change in the 3D heat exchanger domain, are solved for each scenario in the data space to generate the training data. Gaussian process regression, trained using 80% of the generated data and validated by the remaining set, is utilized to identify surrogate models for the objective functions at each time interval during operation. Β© 2024 by ASME.
