Chiller sensor fault detection using denoising-based improved principal component analysis

Mao Qianjun, Fang Xi, Li Guannan, Liang Zhiyuan and Hu Yunpeng

2019.07.16

Presents a chiller sensor fault detection method based on empirical mode decomposition (EMD) threshold denoising (TD) and principal component analysis (PCA). Uses EMD and TD method to enhance the data quality by removing the noise factor in the raw data set, and builds a PCA model with the data processed. Monitors the operational data of a screw chiller system of an electric factory in Wuhan to evaluate fault detection effect. Compares the proposed method with the traditional PCA method and wavelet threshold denoising PCA (Wavelet-TD-PCA) method. The results show that the EMD-TD-PCA method can effectively improve the chiller fault detection efficiency, and its fault detection results are better than those of the PCA and Wavelet-TD-PCA methods at the same bias. Especially at -1 to 1 ℃ bias fault, the fault detection results are improved.