Novel Actuator Fault Diagnosis Framework for Multizone HVAC Systems Using 2-D Convolutional Neural Networks

Abstract

Heating, ventilation, and air conditioning (HVAC) systems are used to condition the indoor environment in buildings. They can be subjected to malfunctioning since they are the most extensively operated buildings' components that account alone for almost half of the total building energy usage. Therefore, fault diagnosis (FD) of the HVAC system is important to maintain the system's reliability and efficiency and provide preventive maintenance. This article presents a supervised FD strategy for single actuator faults in HVAC systems given that actuators, such as dampers and valves, are mostly prone to faults resulting in thermal discomfort and energy inefficiency in buildings. The proposed approach is based on 2-D convolutional neural networks (CNNs) using an efficient 1-D-to-2-D data transformation performed on the time-series signals acquired from the HVAC system. The performance of the CNNs is ensured by an optimal tuning of its significant hyperparameters using the Bayesian optimization algorithm toward maximizing the classification accuracy. The proposed 1-D-to-2-D data transformation approach is computationally efficient and eliminates the use of advanced signals preprocessing. It is performed in two schemes: the static and dynamic schemes to analyze the correlation between the system's variables and consider the temporal effects of the time-series signals without compromising the detection time. The proposed approach is developed and validated using simulation data collected from a three-zone HVAC system simulator using Transient System Simulation Tool (TRNSYS). It demonstrates improved performance compared to the 1-D CNN-based approach and the other standard data-driven approaches for actuator FD in HVAC systems.