Sensor Fault Detection and Diagnosis of Air-Handling Units Using a Condition-Based Adaptive Statistical Method

Abstract

This paper presents a robust strategy based on a multivariate statistical method, principal component analysis (PCA), for the online detection and diagnosis of sensor faults in typical air-handling units (AHU). Two PCA models are built corresponding to the heat balance and pressure-flow balance of the air-handling process. Sensor faults are detected using the Q-statistic and diagnosed using an isolation-enhanced PCA method, which combines the Q-contribution plot and knowledge-based analysis. The PCA models are updated using a condition-based adaptive scheme to follow the normal shifts in the process due to changing working conditions, where the outdoor air temperature and humidity are selected to represent the outdoor operating conditions. The condition-based adaptive scheme overcomes the shortcomings of the time-based adaptive scheme and improves the detectability of the PCA-based fault detection and diagnosis (FDD) method in detecting slowly developing faults. Rules are built to determine the time when the PCA models need to be updated. PCA models generated in the adaptive process are stored in a model database. Simulation tests and field tests in a building in Hong Kong were conducted to validate the strategy for the automatic online monitoring of sensors in AHUs.