Efficient optimization of process parameters in shadow mask manufacturing using NNPLA (neural network partial least squares) and a genetic algorithm

A shadow mask, the primary component of a cathode ray tube (CRT), is used to prevent the outer edges of electron beams from hitting incorrect phosphor dots. It is fabricated by means of a photo-etching process consisting of a few hundred/thousand process parameters. A primary concern in the management of the process is to determine the optimal process parameter settings necessary to sustain the desired levels of product quality. The characteristics of the process, including a large number of process parameters and collinear observed data, make it difficult to accomplish the primary concern. To cope with the difficulties, a two-phase approach is employed that entails the identification of a few critical process parameters, followed by determination of the optimal parameter settings. The former is obtained through the operator's domain knowledge and the NNPLS-based prediction model built between process parameters and quality defects. The latter is obtained by solving an optimization problem using a genetic algorithm (GA). A comparative study shows that the proposed approach improves product quality greatly in the shadow-mask manufacturing process.