The International Association for Automation and Robotics in Construction
Publications / 2025 Proceedings of the 42nd ISARC, Montreal, Canada
Feature Sensitivity Analysis for Enhanced HVAC Fault Detection and Diagnosis
Automated Fault Detection and Diagnostics (AFDD) is a data-driven approach that enables the timely detection of faults, their types, and severity in Heating, Ventilation, and Air Conditioning (HVAC) systems. Machine learning models can be used to develop and implement AFDD models for HVAC at the system, sub-system, and equipment levels. However, there is a discrepancy between experimental facilities and actual buildings in practice. This study implements sensitivity analysis to determine how variations in the features used in the case study under investigation affect the performance of the AFDD models. The analysis includes techniques such as correlation analysis and machine learning models (Support Vector Machine (SVM) and Artificial Neural Network (ANN)) to assess the sensitivity of the models to changes in input features. The results depict how sensitive the AFDD models are to different features and the extent to which variations in these features can impact the models' performance. These findings can facilitate the selection of robust features for machine learning-based Fault Detection and Diagnostics (FDD) models of HVAC in buildings by building operators, including facility managers, asset managers, and owners.
