Exploiting Multiple Mahalanobis Distance Metrics to Screen Outliers From Analog Product Manufacturing Test Responses

Mahalanobis distance is commonly used for fault classification in analogue testing. However, it cannot guarantee a robust mean value and covariance matrix, which makes it an unreliable metric in the presence of outliers. In this case study the authors therefore work with a multi-variate classifier based on multiple Mahalanobis distances from selected sets of test-response measurements. For an industrial automotive product they show that their classifier can both qualitatively screen product outliers and quantitatively measure the reliability of the non-defective ones.