Automated diagnostics for AHU-VAV systems using pattern matching

Abstract

Faults in commercial HVAC systems can result in energy waste of up to 30% of the total usage. This demonstration exhibits a novel application for automated fault detection and diagnosis (AFDD) for air handling units (AHUs) and variable air volume units (VAVs) in building HVAC systems. The application has been designed to facilitate low-cost implementation and fault monitoring, while maintaining high diagnostic accuracy and low false alarm rates. This is accomplished via the use of unsupervised machine learning methods and analytic redundancies, combining pattern-matching methods, principal component analysis (PCA) methods, and Bayesian network analysis into a single AFDD application. The benefit of using these automated methods is the ability to adapt to the built up custom systems found with AHU-VAV installations, thereby minimizing the engineering effort required for implementation.