Techniques to interpret T² control chart signals

A common fault of multivariate control charts is that these charts can signal a change in a process, but do not indicate which variable or group of variables have caused the problem. The literature contains many approaches to solve the problem, from graphical methods to more analytic approaches. One of the more successful approaches consists in decomposing the T ² statistic, however, the effectiveness of this approach in terms of the correct classification percentage of the variables (shift, no shift) has not been previously analyzed. We perform these calculations and compare the results with those obtained using a neural network approach. We show that the results obtained using the neural network method are in general better than those obtained using the decomposition approach.

Acknowledgements

This work was supported by the Ministry of Education and Science of Spain under grant DPI2002-03537 which included European FEDER funding. The authors wish to thank the referees for their thoughtful suggestions that improved the final version of this paper.