Research of feature extraction and fault diagnosis for sensor signal

The paper presents an integrated approach of feature extraction and fault diagnosis of sensor. Taking the sensor signal energies and modulus maxima of wavelet packet decomposition in three bands as initial features, valid features are extracted according to kernel fisher transform of initial feature vector, enhancing the signal characteristics. In accordance with inter-class separability a set of binary SVM classifiers are combined to construct optimal Decision Directed Acyclic Graph. The approach is applied to the FDT/DTM device management system, which is used for pressure sensor fault diagnosis in the water cycle control system of NCS4000, a numerical experiment shows that the algorithm is effective.