2013 18th IEEE European Test Symposium (ETS) (2011)
May 23, 2011 to May 27, 2011
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/ETS.2011.31
Mahalanobis distance is one of the commonly used multivariate metrics for finely segregating defective devices from non-defective ones. An associated problem with this approach is the estimation of a robust mean and a covariance matrix. In the absence of such robust estimates, especially in the presence of outliers to test-response measurements, and only a sub-sample from the population is available, the distance metric becomes unreliable. To circumvent this problem, multiple Mahalanobis distances are calculated from selected sets of test-response measurements. They are then suitably formulated to derive a metric that has a reduced variance and robust to shifts or deviations in measurements. In this paper, such a formulation is proposed to qualitatively screen product outliers and quantitatively measure the reliability of the non-defective ones. The application of method is exemplified over a test set of an industrial automobile product.
Outliers, Analogue, Test, Reliability, Mahalanobis distance
H. G. Kerkhoff and S. Krishnan, "A Robust Metric for Screening Outliers from Analogue Product Manufacturing Tests Responses," 2011 16th IEEE European Test Symposium (ETS), Trondheim, 2011, pp. 159-164.