Presents a methodology that uses a combination of techniques - thermodynamic modelling, pattern recognition, and expert knowledge to determine the "health" of a reciprocating chiller and to diagnose selected faults. The system is composed of three modules. The first one deals with the detection of faults that are more discernible when the chiller is off, such as sensor drift. The second module detects faults during start-up and deals with those related to refrigerant flow characteristics, which are generally more apparent during the transient period. Finally, the third module detects deterioration in performance followed by diagnosis when the unit is operating in a steady-state condition. The approach has been experimentally tested on one laboratory unit and results presented. It is emphasised that further data are required to establish the repeatability of the emerging patterns and validate the applicability of the approach to reciprocating chillers in general.

KEYWORDS: year 1996, performance, monitoring, failure, detectors, chillers, fluid flow, properties, R22, refrigeration