A district cooling system (DCS) is a system that distributes thermal energy through chilled water from a central source to residential, commercial, or industrial consumers, designated to air conditioning purposes. It is one of the most important part of a heating, ventilation, air conditioning and refrigeration systems (HVAC), because a DCS is composed of: Cooling towers, central chiller plant, water distribution systems and clusters of consumer buildings. This research is focused on the central chiller plant, due to it accounts for a substantial portion of the total energy consume of DCS and HVAC systems. The performance of central chiller plant is often affected by multiple faults which could be caused during installation or developed in routine operation. These non-optimal conditions and faults may cause 20–30% waste of energy consumption of HVAC&R systems. Automated fault detection and diagnosis (AFDD) tools have potential to detect an incipient fault and help to reduce undesirable conditions and energy consumption, and optimize the facility maintenance. We propose an online data driven fault detection strategy for district cooling system. The main objective is to develop an automated fault detection tool based on historical process data, which can be applied in transient operation. The proposed hybrid strategy is based on unsupervised and supervised learning techniques, and multivariate statistic techniques. Its aim is to identify the operating states of the chiller and evaluate the fault occurrence depending of its current operating state. This strategy uses the K-means clustering method, Naive Bayes classifier and Principal Component Analysis (PCA). The developed strategy was evaluated using the performance data of a 90-ton water-cooled centrifugal chiller (ASHRAE RP-1043) and also evaluated using a dynamic model of a chiller (Simscape™.) under similar conditions. The results show the advantages of novel early fault detection technique compared to Conventional PCA method in terms of sensitivity to faults occurrence and reduction of missed detection rate.

This content is only available via PDF.
You do not currently have access to this content.