An inquiry into the predictability of failure events in chillers and boilers

This paper explores the predictability of failure events in large boilers and chillers by using data inherent in sensor and actuator data from four boilers and five chillers of a central heating and cooling plant (CHCP) in Ottawa, Canada are extracted. The plant operators are interviewed to understand how they handle failure events, and their logbooks are reviewed to extract the date and time of the recorded failure events. The sensor and actuator data up to two weeks prior to each of these failure events are used to develop regression tree models that predict the remaining time-to failure. The results indicate that about half of the modelled failure events could be accurately predicted by looking at the data available in the distributed control system. Further, rules that define operational conditions leading to failure events were derived by using the regression tree models. It is argued that this regression tree modelling and rule extraction method can guide controls technicians to reconFigure the setpoints and control loop parameters. Lastly, future work recommendations are developed to study prognostics in CHCP equipment by using more comprehensive datasets and to demonstrate the use of prognostics in operational decision-making.