Fault detection and diagnosis for chiller based on feature-recognition model and Kernel Discriminant Analysis. (April 2022)

• Record Type: Journal Article
• Authors: Bai, Xi; Zhang, Muxing; Jin, Zhenghao; You, Yilin; Liang, Caihua
• Is Part Of: Sustainable cities and society. Volume 79(2022)
• Publication Date: 2022-04
• Subjects: Chiller -- FDD -- Kernel Discriminant Analysis -- Feature-recognition model; Sustainable urban development -- Periodicals; Sustainable buildings -- Periodicals; Urban ecology (Sociology) -- Periodicals; 307.76
• DOI: 10.1016/j.scs.2022.103708
• Languages: English
• ISSNs: 2210-6707

Abstract

Highlights: A chiller FDD method combining model-based and data-based methods was proposed. The deviation was used as a diagnostic parameter, and its contribution to FDD was verified. The effect of sample size on the accuracy of the feature recognition model was explored. Compared with the PCA, FDA, and SVM, the proposed method had excellent performance. Abstract: Reliability of chillers is of great significance to maintaining sustainability buildings and reducing carbon emissions. The cause of chiller performance degradation was found in early fault detection and diagnosis technology, and the measure can be taken to save energy. This paper proposed an automatic diagnosis technique for Chiller based on the feature-recognition model and Spectral Regression Kernel Discriminant Analysis (SRKDA). Feature-recognition model would be used to calculate diagnostic parameters, deviations( D ) between normal and fault data. At the same time, SRKDA would be applied for mapping original nonlinearly separable feature space to the separable features that are linear and improving the computational speed. For one thing, compared with principal component analysis (PCA) and Fisher Discriminant Analysis (FDA), the proposed method has the lowest false alarm rate and the highest detection rate for fault detection. For another, compared with the FDA and Support vector machine (SVM) for fault diagnosis, the proposed method has excellent accuracy and training time performance. In addition, experiments show that model-based data processing improved the separability of original data and further improved FDD accuracy.