PCA-SVM-Based Automated Fault Detection and Diagnosis (AFDD) for Vapor-Compression Refrigeration Systems

Abstract

To improve the classification accuracy and reduce the training and classifying time, a novel automated fault detection and diagnosis (AFDD) strategy is proposed for vapor-compression refrigeration systems, which combines principle component analysis (PCA) feature extraction technology and the “one to others” (binary-decision-tree-based) multiclass support vector machine (SVM) classification algorithm. Eight typical faults were artificially introduced into a refrigeration system in the laboratory, and tests for normal and faulty conditions were carried out over a −5°C~15°C (23°F~59°F) evaporating temperature and a 25°C~60°C (77°F~140°F) condensing temperature. The data obtained for 16 variables are first preprocessed by PCA to get four comprehensive features (principle components) that account for over 85% of the cumulative percent value (CPV); the new sample data are then randomly split into training (70%) and testing (30%) sets as the input of an eight-layer SVM classifier for AFDD. Results show that the proposed PCA-SVM strategy has better detection and diagnosis capability with more satisfying FDD accuracy and is less time consuming compared to other approaches, such as SVM without PCA, back propagation neural network (BPNN) and the “one to one” and “one to rest” multi-class SVM algorithms. In this sense, this study provides a promising AFDD strategy for vapor-compression refrigeration system application.