Digital Twin-Assisted Predictive Calibration of Generator Sets Under Variable AI Workload Profiles in Mission-Critical Data Centers

Abstract

Mission-critical data centers typically rely on generator sets that operate for a small percentage of their total lifecycle. Accurately calibrating these generator sets is crucial to meeting operational reliability and efficiency while reducing replacement costs and maximizing asset lifetime. However, conventional methods do not integrate lifecycle data from the generator sets or the synchronization of calibration and control. This work proposes a digital twinassisted predictive calibration solution that incorporates sensor streams from generator sets, telemetry from a preventive maintenance system, associated weather predictions, and the demand profile of Artificial Intelligence-supported workloads running on the data centers. The research concludes by analysing the robustness of the prediction of the generator-set parameters when the response of the Data Centre is exposed to variation from the normal workload profile for which the Data Centre was designed.