Issue: A large Southwest US pharmaceutical and medical device company had a small percentage of lower-quality parts reaching their markets. They were interested in an in-line test method to insure that only the highest quality devices move beyond the beginning stages of production.

Solution: Through the use of partial least squares-discriminate analysis modeling, InvenGen was able to isolate product configuration factors that were strongly correlated. A novel method was devised to combine different types of data streams and create models of the product that explained 94% of the model's variability.

Result: The resulting models were able to classify non-conforming product from production with exceptional accuracy. The models demonstrated excellent ability in discriminating profile and surface features. Even though the results represent the efforts of this project's first stage, estimates project hundreds of thousands of dollars of savings in inspection labor, scrap reduction through further process improvements and increased customer satisfaction.