Fault detection in AHU: A walkthrough for implementation in a Danish educational building

Abstract

The implementation of fault detection in research articles is relatively sparse, yet it holds significant potential to contribute to the decarbonization of our building stock. This study proposes a Fault Detection (FD) methodology that can be split into three stages. 1) FD algorithm development, testing, and validation on known datasets. 2) platform creation, data collection and curation, and method implementation. 3) FD algorithm testing on an operational system running under normal conditions with no artificially induced faults. The results showed that for step 1, using suitable evaluation metrics for realistic datasets is highly important, as otherwise wrong conclusions can be drawn. It was further found that the proposed FD methodology, as intended, led to choosing an algorithm that did not cause many false alarms, as the emphasis was on avoiding these, but also that changing the weighting of the included terms could shift the focus to prioritize other issues.