Model-based Performance Monitoring: Review of Diagnostic Methods and Chiller Case Study

The paper commences by reviewing the variety of technical approaches to the problem of detecting and diagnosing faulty operation in order to improve the actual performance of buildings. The review covers manual and automated methods, active testing and passive monitoring, the different classes of models used in fault detection, and methods of diagnosis. The process of model-based fault detection is then illustrated by describing the use of relatively simple empirical models of chiller energy performance to monitor equipment degradation and control problems. The CoolTools™ chiller model identification package is used to fit the DOE-2 chiller model to on-site measurements from a building instrumented with high quality sensors. The need for simple algorithms to reject transient data, detect power surges and identify control problems is discussed, as is the use of energy balance checks to detect sensor problems. The accuracy with which the chiller model can be expected to predict performance is assessed from the goodness of fit obtained and the implications for fault detection sensitivity and sensor accuracy requirements are discussed. A case study is described in which the model was applied retroactively to high-quality data collected in a San Francisco office building as part of a related project (Piette et al. 1999).