To maintain standards, all manufactured automotive items are subject to strict quality control. Quality control processes test items against agreed upon specification limits. Items that “pass” these tests are "good enough" for commercial use, while items that “fail” are discarded. Quality control testing for large numbers of manufactured products can be costly; as a result, manufacturers might implement inexpensive test processes. Less expensive test processes may result in lower accuracy measurements. Utilizing lower accuracy measurements can increase both the number of discarded items and the number of returned (or recalled) items. We have created a multi-dimensional statistical ellipse model, formed from both lower and higher accuracy measurements, with which the impact of decisions based on lower accuracy test procedures can be assessed.
A common method used to improve production yield is to shift the specification limits, increasing the process’s capability index. However, this one-dimensional approach excludes the impact on returned items. While a small shift in limits may increase production yield, it could unintentionally increase returned yield. Our multi-dimensional statistical ellipse model uses higher accuracy measurements to assess the cost differential of returned and discarded items produced by specification changes. A sensitivity analysis is performed on the lower accuracy measurements, resulting in optimal resource allocation. The generic mathematical formulation is completely defined and a Matlab implementation is available from the author.