Hybrid-Ensemble Learning Based Fault Detection and Diagnosis of Chillers*

Most proposed fault diagnosis models for chillers use a single classification approach, which limits their efficacy and ability to improve. To tackle this problem, this study suggests that ensemble learning models (ELMs) are created to integrate diverse base models. In addition to a single steady-state condition and 7 fault situations, ELMs are simulated for steady-state and dynamic conditions. The findings of study reveal that ELMs (soft voting) outperform other machine learning systems in both diagnostic performance and defect detection of chillers with accuracy reaching up to 97.82%. ELMs are much better at finding true positives and true negatives, as well as rejecting false positives and negatives, compared to other ensemble classification models. They are also significantly robust in terms of noise or outliers.