A Method of Dimensionality Reduction of Analog Circuit Fault Feature

To deal with dimensionality reduction of nonlinear analog circuit fault features, a method named local linear discriminant analysis (LLDA) was proposed. First, considering the local neighborhood of each data, the projection direction in the local space is found by making the neighbor samples of the same class as compact as possible and the neighbor samples of different classes as distant as possible. The final objective function is reconstructed by an alignment algorithm on the entire data space. Finally, a projection matrix is obtained by performing the standard eigenvalue decomposition for dimensionality reduction. The algorithm not only makes full use of the local discriminant information of data manifold to process the nonlinear data, but also keeps the class separate. The two-stage four-op amp low-pass filter circuit demonstrate the effectiveness of LLDA.