A hybrid data-driven simultaneous fault diagnosis model for air handling units

Simultaneous faults are situations where two or more faults occur at the same time, which are difficult to be diagnosed by simple and stand-alone standard machine learning methods as a multi-label problem. Simultaneous faults for HVAC systems are not given enough attention under the challenges of insufficient sensors, coupled faults, and sophisticated mathematical models. A novel simultaneous fault diagnosis model based on a hybrid method of classifier chains integrated with random forest (CC-RF) is proposed in this study. On-site experiments involving six single fault cases and seven simultaneous fault cases for an air handling unit (AHU) system are conducted to verify this model. The results demonstrate a satisfactory performance with the test accuracy of 99.50% and F1 score of 99.66% for the fault diagnosis model. The model is proven to be neither underfitting nor overfitting and can be scalable with a reasonable training time. Through online analysis, the proposed method demonstrates a good competence of diagnosing not only single faults but also simultaneous fault. The CC-RF method has a better performance compared with classifier chains with logistic regression and support vector machine. Besides, the proposed method of classifier chains outperforms binary relevance due to the benefitting of label relevance. (c) 2021 Elsevier B.V. All rights reserved.

Automated summary

A novel simultaneous fault diagnosis model, CC-RF, is proposed and validated with on-site experiments, achieving high accuracy and performance in diagnosing both single and simultaneous faults. The model is proven to be scalable with reasonable training time and shows good competence in online analysis.