Techniques for Improving the Performance of Automated Fault Detection and Diagnosis Modules through Hyperparameter Tuning

The energy used in buildings accounts for approximately 30% of the total energy consumption worldwide.HVAC (Heating, Ventilation, and Air Conditioning) systems are essential for creating indoor environments tailored to the building's purpose and maintaining the occupants comfort.However, in commercial buildings, HVAC systems consume between 30% and 50% of the building's total energy usage.Therefore, many studies aim to reduce this energy consumption.This study focuses on Automated Fault Detection and Diagnosis (AFDD), which can detect and diagnose issues in HVAC systems early, whether due to faults or aging, to reduce wasted energy.Recent trends in FDD research reveal a focus on data-driven approaches that align with AI or automatic control keywords, evaluating or enhancing fault diagnosis performance using machine learning techniques.In this study, we developed an FDD module using an Artificial Neural Network (ANN) model, known for its high predictive accuracy and capability to handle high-dimensional data.We analyzed which hyperparameters need to be optimized to improve diagnostic performance and the extent of performance improvement achieved through optimization.As a results, by analyzing the optimal performance through adjustments of hyperparameters such as number of hidden layers and train function of the ANN, we were able to develop an AFDD module with classification performance of over 96%.