A model-based fault detection and diagnostic methodology based on PCA method and wavelet transform

Abstract

Building automation systems (BASs) are widely used in modern buildings and large amounts of data are available on the BAS central station. This abundance of data has been described as a data rich but information poor situation and has given an opportunity to better utilize the collected BAS data for fault detection and diagnostics (AFDD) purposes. Air-handling units (AHUs) operate in dynamic environment with changing weather conditions and internal loads. It is challenging for FDD method to distinguish differences caused by normal weather conditions change or by faults. Principle Component Analysis (PCA) has been found to be powerful as a data-driven model based method in detecting AHU faults. Wavelet transform is a promising data preprocess approach to solve the problem by removing the influence of weather condition change. A combined Wavelet-PCA method is developed and tested using site-data. The feasibility of using wavelet transform method for data pretreatment has been demonstrated in this study. Comparing to conventional PCA method, Wavelet-PCA method is more robust to the internal load change and weather impact and generate no false alarms.