KKCV-GA-Based Method for Optimal Analog Test Point Selection

Abstract

A novel diagnosis oriented method for optimal analog test point selection is proposed. The method uses a kernel density estimation on $K$ -nearest neighbors-based cross-validation to evaluate the diagnostic capability of the selected test points, and then employs a genetic algorithm to search the quasi-optimal solution with the maximum diagnostic capability under a limited number of test points. Experimental results show that better solutions in terms of the diagnostic accuracies of the advanced intelligent classifiers can be obtained, and the knowledge of critical test points is revealed, which supports further improvement of the optimization, not only to increase the probability of reaching the global optimal solutions but also to reduce the time costs. In addition, with the precise estimation of the diagnostic capability, the proposed method can also be used to verify the results produced by other methods in this field.