A Divide-And-Conquer Information Entropy Algorithm for Dependency Matrix Processing

Processing the system's dependency matrix is a core procedure for system diagnosis. However, the present algorithms for this job always try to maximize its fault isolation capability, which is thus not only unnecessary for the system that cannot be repaired in the field like weapons, but also generates low-efficient test sequence. To this end, the present work proposes a new processing algorithm for testability D-matrix, named Divide-And-Conquer Information Gain (DIG), targeting to those systems without strong fault isolation requirement. It combines the advantage of the classic algorithm Information Gain (IG) and Weight index for Fault Detection (WFD) by introducing a new entropy computing method considering the weight index for fault detection. To verify the advantage, generality and flexibility of the new algorithm, a D-matrix from a real system and random D-Matrixes are tested in the experiments, and measured by test sequence length, actual test cost and expected test cost. The result shows that DIG algorithm is 15.7% and 14.3% better than that of IG and WFD algorithm on the expected test cost metric, respectively.