Concurrent Learning-Based Fault Detection in Closed-Loop HVAC Systems with Inaccessible Control Inputs

Heating, Ventilation and Air-Conditioning (HVAC) systems are critical components that consume a significant percentage of the energy in the building sector. Faults in the sensing and actuation equipment of HVAC systems can create uncomfortable indoor conditions, as well as cause significant waste of energy. State-of-the-art fault detection techniques typically require control input data that maybe inaccessible in practice, since HVAC systems in large-scale buildings are controlled and monitored by proprietary Building Management Systems (BMS). This paper proposes a learning-based fault detection approach for closed-loop HVAC systems with inaccessible control information. The learning-based adaptive estimation scheme aims to estimate the temperatures within each zone-air handling unit pair and the unknown control gains that have been selected by the manufacturer. The proposed scheme enables the design of less conservative detection thresholds, which enhances the performance of the fault detection procedure for the closed-loop HVAC system. Simulation results illustrate the effectiveness of the proposed method in a large-scale HVAC system using the EnergyPlus software. Copyright (C) 2022 The Authors.