Sensor fault diagnosis and data reconstruction based on MSPCA

Because of the disadvantage of conventional MSPCA based on wavelet transform when detecting fault with high frequency, the method with MSPCA based upon wavelet packet decomposition was proposed and applied into sensor fault diagnosis and data reconstruction in this paper. Firstly, the sensor data was decomposed as orthogonal wavelet packet transform to achieve the best-tree for decomposition. Modals were established at each scale corresponding to the coefficients of the best-tree. Sensor fault was detected by the square prediction error in the residual subspace of the main principal space, and the faulty sensor was discriminated via sensor validation index. After the reconstruction of the PCA model that had detected and identified the faulty sensor, it was reconstructed by reverse wavelet package transform. Finally, the result of diagnosis and data reconstruction for cyclic failure of the sensors in the ground testing bed illustrates the effectiveness of the modal established above.