Automated calibration of semiconductor fabrication HVAC models using energyplus-isight integration for digital twin fidelity enhancement

초록

Accurate building energy simulation is essential for implementing a reliable digital twin (DT), especially in semiconductor fabrication (FAB) facilities, where heating, ventilation, and air-conditioning (HVAC) systems are both highly energy-intensive and complex. This study aims to enhance the fidelity of a DT-based energy model by developing an automated calibration framework that integrates EnergyPlus with Isight. The initial EnergyPlus model was developed based on actual facility data, including zoning configurations, envelope U-values, internal heat gain, and specifications for makeup air handling units (MAUs), chillers, and cooling towers. Electricity consumption data from a FAB HVAC system were used for calibration. The Design of Experiments (DOE) module in Isight, using Optimal Latin Hypercube Sampling, generated parameter combinations, while the Data Matching module applied the Pointer Method to minimize discrepancies between simulated and measured results. Key calibrated parameters included internal heat gains, system capacities, chiller coefficient of performance (COP), design water flow rates, fan power, pressure conditions of pumps, and cooling coil loads of MAUs, among others. Automated calibration enhanced model accuracy, increasing the coefficient of determination (R2) from 0.84 to 0.98 and decreasing the coefficient of variation of the root mean square error (CVRMSE) from 8.17% to 1.22%. Moreover, calibration time decreased from over one month using the manual method to less than one day. In conclusion, the proposed automated approach improves model accuracy and efficiency by minimizing manual trial-and-error procedures and mitigating user subjectivity. This framework is well-suited for semiconductor FABs and scalable to other industrial energy systems requiring high-fidelity DT models.

키워드