Fault detection and diagnosis for the screw chillers using multi-region XGBoost model

• Published in: HVAC&R research Vol. 27; no. 5; pp. 608 - 623
• Main Authors: Zhang, Shuai, Zhu, Xu, Anduv, Burkay, Jin, Xinqiao, Du, Zhimin
• Format: Journal Article
• Language: English
• Published: Philadelphia Taylor & Francis 28-05-2021; Taylor & Francis Ltd
• Subjects: Air conditioning; Chillers; Clustering; Fault detection; Fault diagnosis; Model accuracy; Strategy; Support vector machines; Thermal comfort
• ISSN: 2374-4731; 2374-474X
• DOI: 10.1080/23744731.2021.1877966

Chillers play essential roles in the heating, ventilation and air conditioning (HVAC) systems to ensure the required thermal comfort. To reduce the operational risk such as faulty operation or energy waste, it's essential to develop the robust and effective fault detection and diagnosis (FDD) strategy for the chillers. This paper presents a novel hybrid reference model called multi-region XGBoost model that integrates parameter optimized XGBoost model with mean shift clustering method. Based on the reference model, an FDD strategy, including two stages, is proposed. The experiments are carried out on a screw chiller, on which three thermal faults are investigated. The indicative characteristic quantities are selected as the model inputs for detection and diagnosis purpose. The FDD result of the multi-region XGBoost model is compared with that of support vector machine (SVM) model and XGBoost model without clustering. In terms of the performance of fault detection, the multi-region XGBoost model detects 97.26% faulty samples while correctly identifies 99.10% fault-free samples. As for fault diagnosis, the multi-region XGBoost model possesses the highest fault diagnosis accuracy of 96.89%. Besides, the hybrid model also shows the best generalization ability. The FDD result reveals that the multi-region XGBoost model has the reliable efficiency for the screw chiller application.