"Good" parts - "Bad" parts discrimination: An ARX modeling approach

This investigation focuses on the formulation and assessment of a strategy for the discrimination of good and bad manufactured pads. The proposed approach is based on an exogenous autoregressive structural model of the good parts and on the error of this model in predicting the measured responses of good and bad parts. The discrimination strategy relies on the expectation that the modeling error of good parts will be smaller than the one of bad parts since the model is representative of one or several good parts. Six techniques were then introduced to reduce the modeling error, a time-dependent vector, into a few overall discrepancy measures which quantify in different ways the "magnitude" of the error vector and thus can each be used as a basis for the discrimination of the good parts from the bad ones. The results of a preliminary assessment with simple one and four degree of freedom models of the parts clearly indicates that the present error-based discrimination strategy performs very well - it is generally either optimum or close to it.