A machine learning framework for detection and severity estimation of faults for chillers and air handling units in HVAC systems

Abstract

Heating Ventilation and Air Conditioning (HVAC) systems are omnipresent in modern-day industrial plants, office spaces, and residential complexes. Chillers and Air Handling Units (AHU) are the critical HVAC sub-systems that perform vital operations and consume maximum energy. Any fault in these sub-systems can thus adversely affect any organization's operations, making fault detection pivotal for effective maintenance. This study proposes a machine learning-based chiller and AHU fault detection framework that aims to reduce the risk of Type I Error and detect the fault severity. We first discuss detecting specific chiller and AHU faults using the Neyman-Pearson principle, controlling the Type I Error rate under 5%, with high probability. We then propose an ensemble-based feature selection method, which yields a set of essential features that can accurately estimate the severity level of specific faults. Further, Random Forests and k-Nearest Neighbors-based classifiers are employed to detect fault severity. We evaluate the performance of the proposed framework on two benchmark data sets, ASHRAE-RP-1043(Chiller) and ASHRAE-RP-1312(AHU), respectively, which show promising results.