Determining manufacturing parameters to suppress system variance using linear and non-linear models

Determining manufacturing parameters for a new product is fundamentally a difficult problem, because there has little suggestion information. There are several researches on this topic, and most of them focus on single specific model or the engineer's experience. As to other approaches, the usage of multiple models may be an alternative approach to help determining the parameters. This research proposed an aggregation of multiple regression and back-propagation neural network to find the manufacturing parameter's limits (upper and lower limits). A real-problem of a new product parameter setting model in the real Thin Film Transistor-Liquid Crystal Display (TFT-LCD) manufacturing company is demonstrated, where three forecasting models are applied, and t test is used to judge which models are the suitable ones. Finally, we average the computed parameter values from the chosen models to suppress the system variance. The empirical results show that the proposed method is successful in suppressing the system variance and improving the production yields.