An application of hidden Markov model on determination of root cause frequency and locations

Abstract

Manufacturers are often interested in deciding frequency of an abnormal status causing a nonconformity of a certain quality attribute. In this article, analytical solutions for determining the frequency of an anomalous quality characteristic guided by an underlying root cause are illustrated. Using the independent mixture model and the hidden Markov model, we were able to measure the probable occurrence of past discrepancies. In fact, both of our proposed models draws a similar conclusion in terms of the proportion of incidence. However, the hidden Markov model provides additional information about the transition probabilities and the time positions of abnormal data points, which would be favorable for the manufacturer to take the decision about the required investment to eliminate the source of the root cause.