Mining Production Data for Quality Control

Traditionally, industrial engineers use a set of statistical test and inference procedures, known as Six Sigma techniques; these work well for a few variables and a few hundred observations when the underlying production processes are well understood. However, many production processes often have hundreds of control and quality measurement variables, involving chemical, mechanical, electrical and magnetic processes. These processes are generally controlled by complex nonlinear relationships, often not completely understood by operators and plant engineers. Furthermore, the collected data is noisy. In such situations, data mining offers an alternative approach to relate quality with process variables. In this study, we describe the results of a pilot project of applying data mining to the manufacturing of coated glass. The glass coating involves mechanical, chemical, electrical and magnetic processes with over one hundred variables. The project was aimed at understanding the influence of different process variables on the quality of the glass coating to minimize the trial and error process of variable setting. The neural network and CART approaches were applied to model the relationship and analyze the coating process. The initial results show the promise of using data mining techniques to solve the complex production problem.

Please contact this expert for more information.

To see the resume of the expert associated with this case study, see the link below.