A robust fault detection and diagnosis strategy for centrifugal chillers.

摘要

Building Heating, Ventilating, Air-conditioning and Refrigerating (HVAC&R) systems usually fail to satisfy performance expectations envisioned at design stage for a variety of reasons. FDD (Fault Detection and Diagnosis) methods are needed to identify the faults and help put the systems back into normal operation. However, the performance of FDD methods applied to HVAC&R systems require sensor measurements, which are also susceptible to various faults. Although a great deal of FDD research has been carried out on component faults or sensor faults in HVAC&R systems, not much research has tackled these two different kinds of faults when they occur simultaneously in HVAC&R systems.; This thesis presents a robust FDD strategy for centrifugal chillers, which are the core of building HVAC&R systems. The strategy consists of a basic chiller FDD scheme and a PCA-based sensor FDD&E (Fault Detection, Diagnosis and Estimation) scheme, which target chiller component faults (called chiller faults hereafter for conciseness) and sensor faults, respectively. In implementing the strategy, it is important that the measurements of the sensors crucial to the chiller FDD are validated by the sensor FDD&E scheme before the chiller FDD scheme is carried out to detect and diagnose chiller faults.; The basic chiller FDD scheme employs six performance indexes which have strong thermophysical meaning to depict the health status of a typical centrifugal chiller. The residual for each performance index is generated by comparing the measured value with its benchmark predicted by a simple reference model. Once residuals of one or more performance indexes are outside the range defined by the corresponding fault detection thresholds, the chiller will be considered to be faulty. Particular faults are diagnosed by a fault diagnostic classifier, which is based on the deviation pattern of the performance indexes when a chiller fault occurs. Robustness of the scheme is achieved by the adoption of an adaptive estimator for the fault detection threshold. The estimator can set reasonable fault detection thresholds by taking account of essential influencing factors. (Abstract shortened by UMI.)