Abstract
A data-driven methodology for improving product yield by integrating principal component analysis (PCA) and liner discriminant analysis (LDA) is proposed. Referred to as Data-Driven Quality Improvement (DDQI), the proposed method can cope with qualitative as well as quantitative quality variables, determine the operating conditions that can achieve the desired product quality, optimize the operating condition under various constraints, and also evaluate the validity of the results. The relationship between product quality and operating conditions can be modeled by PCR when quality variables are quantitative. On the other hand, LDA can be used for modeling when quality variables are qualitative, e.g., good or bad. For such a qualitative quality variable, the yield, that is the percentage of good products to all products, can be specified on the basis of histograms for given categories. The histograms can be obtained from operation data, and they can be drawn against the axis defined by LDA. Once the desired yield is specified, the operating condition that can achieve the desired yield can be determined. The usefulness of the proposed method is demonstrated through a case study.
