A novel efficient SVM-based fault diagnosis method for multi-split air conditioning system's refrigerant charge fault amount. (5th September 2016)

• Record Type: Journal Article
• Authors: Sun, Kaizheng; Li, Guannan; Chen, Huanxin; Liu, Jiangyan; Li, Jiong; Hu, Wenju
• Is Part Of: Applied thermal engineering. Volume 108(2016:Sep.)
• Page Start: 989
• Page End: 998
• Publication Date: 2016-09-05
• Subjects: Fault diagnosis -- Max-relevance and min-redundancy -- Refrigerant charge amount -- Support vector machine -- Wavelet de-noising; Heat engineering -- Periodicals; Heating -- Equipment and supplies -- Periodicals; Periodicals; 621.40205
• DOI: 10.1016/j.applthermaleng.2016.07.109
• Languages: English
• ISSNs: 1359-4311

Abstract

Highlights: Proposed a hybrid model for diagnosing refrigerant charge faults. Only original equipment manufacturer (OEM) sensors are modeled. Wavelet de-noising improves the fault diagnosis efficiency greatly. Correlation analysis of features is implemented for feature selection. Proposed method can handle noisy data having vast numbers of features. Abstract: For the multi-split variable refrigerant flow (VRF) system, the key of efficient operation is to achieve the appropriate refrigerant charge amount (RCA). However, it is difficult to achieve because of the complexity of VRF systems. To overcome the difficulty, this paper presents a hybrid RCA fault diagnosis model combined support vector machine (SVM) with wavelet de-noising (WD) and improved max-relevance and min-redundancy (mRMR) algorithm. WD is responsible for improving the quality of collected VRF experimental data. In addition, mRMR is firstly used to rank all the variables in descending order in terms of their importance for identify RCA faults. After top-ranked variable is determined, correlation analysis of features is implemented for further feature selection removing the redundant variables in linkage to the variable at the top. Finally, a subset of seven features are selected to develop the SVM model. Results indicate that fault diagnosis accuracy of the seven-feature SVM model decreases only 2.14% compared with the initial eighteen-feature model. The proposed wavelet de-noising-max-relevance and min-redundancy-support vector machine (WD-mRMR-SVM) model shows good fault diagnosis performance for RCA faults.