Fault classification of waste heat recovery system based on support vector machine

In this paper, the fault classification problem for waste heat recovery system based on support vector machine (SVM) is investigated. Firstly, two-class SVM classification algorithm is reviewed. Then the model and six kinds of faults in waste heat recovery systems (WHRSs) are briefly described. In order to effectively isolate these faults in WHRSs, key features are extracted using principal component analysis (PCA), the multi-class classification problem is then decomposed into five two-class classifiers by using improved one-against-rest approach. Consequently, the SVM classifiers are designed to train and test samples by using collected process variables. The comparison between SVM and back-propagation neural networks applied to a WHRS is discussed. Simulation results demonstrate that SVM can obtain better fault diagnosis performance.