HVAC Smart Predictive Maintenance Using Machine Learning and Bayesian Network

Abstract

Smart Predictive Maintenance (SPM) features for the building's Heating, Ventilation, and Air Conditioning (HVAC) system are crucial for reducing energy consumption, improving scheduling, and detecting potential problems. Popular approaches, such as Machine Learning (ML) and probabilistic methods, are employed for SPM. These methods can be considered forward inference. However, since numerous interdependent HVAC components are involved, SPM requires not only forward but also backward inference (diagnostic capabilities). Given that such abilities have been underexplored, the present study proposes an SPM-based HVAC monitoring system that combines ML and Bayesian Network (BN). While ML is used to predict the status of the HVAC components, BN performs the diagnostic tasks. A case study was conducted at the Sydney Aquarium in Australia to demonstrate the implementation of the proposed approach. The ML model, trained using the Simple Logistic Regression (SLR) method, achieved an accuracy of 0.99, higher than the 0.92 obtained by using the Decision Tree (DT) and Logistic Regression (LR) methods. Furthermore, the BN was used to diagnose and estimate the probability of a component's performance degradation if another component was problematic. Among the key benefits of this proposed system is its potential to enhance operators' understanding of problems with HVAC systems.