Robust model-based fault diagnosis for air handling units

Abstract

Fault detection and diagnosis (FDD) for heating, ventilation and air conditioning (HVAC) equipment significantly impact energy consumption in both residential and commercial buildings. In most modern building management systems (BMS), HVAC historical data logs in large quantity and high resolution are recorded and available for further online or offline analysis, including automated FDD. In this paper, a model-based fault diagnosis method is developed by applying support vector machine (SVM) techniques to model parameters recursively calculated by an online estimator. The estimator presumes an autoregressive time series model with exogenous variables (ARX). A real-world air handling unit (AHU) dataset containing process variables measured at regular intervals is pre-processed by the online estimation algorithm. Each data vector in the original dataset (measured), together with a small number of appropriately selected lags, is converted to a parameter vector representing the state at the same instant. The set of parameter vectors is sub-divided into classes by SVM, enabling fault classification. Validation via experimental data demonstrates that the proposed hybrid approach produce superior performance measured by F-measure scores compared to alternative methods. Robustness to model uncertainty is also established.