Combined support vector novelty detection for multi-channel combustion data
Multi-channel combustion data, consisting of gas pressure and two combustion chamber luminosity measurements, are investigated in the prediction of combustion instability. Wavelet analysis is used for feature extraction. A SVM approach is applied for novelty detection and the construction of a model of normal system operation. Novelty scores generated by classifiers from different channels are combined to give a final decision of data novelty. We compare four novelty score combination mechanisms, and illustrate their complementary relationship in assessing data novelty. © 2007 IEEE.