Deflection coefficient maximization based cooperative multi-sensor feature extraction for fault diagnosis

In this paper we address two feature extraction approaches schemes, namely the maximum eigen-vector based algorithm (Max-EV) and the multiple eigen-vector (Mul-EV) based algorithm. They serve to detect the state characteristic parameters over rotating machinery using multi-sensor. A fusion centre (FC) is operating to collect the raw observations data from the multisensor in the network and make the final decisions over the information fusion. The optimal weights of the proposed multi-sensor feature information sensing methods are the multiple eigenvectors of the signal sample autocorrelation matrix or the maximum eigenvector. The proposed algorithms demand no a prior knowledge of the noise power and the mechanical signal. Theoretical analysis and simulation results show the proposed methods are robust against the noise power uncertainty and require less sensing data to yield the same performance, compared with the conventional feature extraction methods.