A data-driven strategy for detection and diagnosis of building chiller faults using linear discriminant analysis

Chillers contribute to a significant part of the building energy consumption. In order to save energy and improve the performance of building automation systems, there is an increasing need for chiller fault detection and diagnosis (FDD). This paper proposes a two-stage data-driven FDD strategy which formulates the chiller fault detection and diagnosis task as a multi-class classification problem. Linear Discriminant Analysis (LDA) is adopted to project the high dimensional data into a lower dimensional space so as to achieve maximum class separation and original class information maintenance. At the first stage, a fault is detected and diagnosed if the monitoring data set is the closest to one of the predefined fault clusters and within the predefined Manhattan distance range of the corresponding fault. At the second stage, fault severity level is recognized by comparing the monitoring data set with the corresponding predefined severity level clusters. The fault is diagnosed as at a particular severity level if it is the closest to the corresponding severity level cluster. The proposed strategy is validated by the experimental data of ASHRAE Research Project 1043 (RP-1043). Results show that the data-driven FDD strategy using LDA can detect and diagnose chiller faults effectively.