Data Resource Library for AHU of HVAC/ACMV Condition Monitoring: A Case Study Using AI-Driven Intelligence Approach

Résumé

Air Handling Units (AHU) classic leakage of outdoor-air dampers generates energy increments in the cooling-system and this energy increments is used for threshold-based Fault Detection and Diagnosis (FDD). For tropical buildings equipped with gas-fired reheats the thermodynamic effect on heating and reheat load resulting from excess outdoor air enthalpy leads to loss of such characteristic energy signatures, making it impossible for FDD systems based on fan electric power to identify the fault. This work generates a public-accessible labeled data library on a particular type of fault occurring in a tropical climate gas-reheat VAV AHU. Energy Plus 23.2 and OpenStudio 3.7.0 simulation, with modifications of the ORNL Flexible Research Platform for Kuala Lumpur climate and with fault levels of severity between 20% and 50%, yield 16,680 hours analyzed based on Fault Impact Ratio calculation, sensitivity ranking, and Random Forest classification. The fault manifests as an ambiguous signal pattern: no fan electricity sensitivity (0.00%), 2.5% growth in cooling coil electricity consumption, up to 85.8% reduction in hot water pump electricity, and up to 88.66% decrease in boiler natural gas consumption, rendering both fan-based and energy threshold FDD ineffective. Electrical-signal based FDD attains 71.47% balanced accuracy while employing the practical signal group yields 95.49%. This dataset is introduced as an open access FDD  dataset for Southeast Asia. Thermal and latent energy signatures of the hot water loop and cooling coil, respectively, can be used to diagnose the fault without any direct airflow measurements.