<![CDATA[Chillers in government buildings located in tropical climates often operate without a building automation system (BAS), making performance degradation difficult to detect through routine monitoring. This study reanalyzes the 2024 field measurement dataset for a 500 TR (R134a) water-cooled centrifugal chiller in a city government office without new data collection. The Isolation Forest model interpreted with permutation-based SHAP attribution marks instances of zero flow rate and surrounding reading changes, with residuals against local trends as the primary explanatory feature. Extended flatline segments following zero events are overlooked, complementing rather than replacing manual screening. The marking proportion ranges from 6.2% to 6.5% per session, adhering to a contamination parameter of 0.06. Pipe cross-section examinations restore stable diameters at three locations: 240 mm, 97.7 mm, and 450 mm, correcting a labeling error. Recalculated results show evaporator heat of 825.5 ± 79.7 kW and an actual COP of 3.75 ± 0.36 compared to a design COP of 5.76 at a measured load of 47.2% capacity. The energy balance deviation reaches 11.6%, exceeding the 5% threshold. This imbalance reflects combined uncertainty, with specific power at 0.94 kW/TR and a temperature difference of 1.3 °C, consistent with additional thermal resistance yet to be verified.]]>
Copyrights © 2026
