Enhancing Fault Detection and Diagnosis in AHU Using Explainable AI

Heating, ventilation, and air conditioning (HVAC) systems consume a significant amount of energy in buildings and can experience faults even when maintained well, resulting in increased energy consumption and maintenance costs. To address this issue, it is important to develop an efficient fault detection and diagnosis (FDD) system. Data-driven approaches that use machine learning (ML) algorithms offer many advantages over physics-based approaches for HVAC systems since they operate in a transient state most of the time. However, the lack of transparency of ML algorithms makes stakeholders hesitant to adopt them for FDD. Therefore, this study examines the use of eXplainable Artificial Intelligence (XAI) techniques to improve FDD in Air Handling Units, to enhance their interpretability. Four machine learning (ML) techniques were evaluated based on fault detection rate and F1 score, and XGBoost algorithm was determined to be the optimal choice for the FDD model. The study utilized SHapley Additive exPlanations (SHAP) for interpreting the developed model. The SHAP summary plot and the SHAP waterfall plot are used for global and local explanations, respectively. The study successfully demonstrated the effectiveness of XAI techniques in improving the transparency and interpretability of ML models for FDD in Air Handling Units.