Fault Detection in AHUs Using Deep Learning and Isolation Forests for Semi-Labeled Data

Air Handling Units (AHUs) are key components of Heating, Ventilation, and Air Conditioning (HVAC) systems, responsible for air circulation and conditioning in buildings. Faults in AHUs can reduce energy efficiency, affect air quality, and increase maintenance costs. Detecting and diagnosing these faults quickly is essential. Traditional machine learning methods struggle with new fault types due to limited and imbalanced data. This study presents a method for AHU fault detection using deep learning and Isolation Forest algorithms with semi-labeled data. The model employs Convolutional Neural Networks (CNNs) to analyze operational data and detect both known and new faults, while the Isolation Forest algorithm identifies anomalies. As new faults are detected, they are added to the training data, improving the model over time. Tested on AHU data from an office building in Kuala Lumpur, the model identified nine fault types, including airflow, valve, Supply Air Temperate (SAT), and Return Air Temperature (RAT), despite being trained on only five. The CNN model achieved a precision of 0.94, recall of 0.91, accuracy of 0.95, and an F1-score of 0.92. The Isolation Forest model also performed well with a precision of 0.91, recall of 0.89, accuracy of 0.92, and F1-score of 0.90. In conclusion, this study demonstrates the effectiveness of deep learning with semi-labeled data for improving AHU fault detection and HVAC system efficiency.