Application of an LM-neural network for establishing a prediction system of quality characteristics for the LGP manufactured by CO2 laser

The light guide plate (LGP) is a part of a backlight module, which evenly spreads light sources in a liquid crystal display (LCD) to eliminate total reflection. Light is transmitted into the LGP, where it is reflected, scattered, and refracted due to the microstructure, which allows light to uniformly enter the panel. Therefore, the design of an LGP micro-structure and processing mode is important for light transmission in an LCD. This study used a CO2 laser to fabricate a polymethyl methacrylate (PMMA) LGP, applied a Taguchi orthogonal array to set up the experiment, and utilized the data to establish a prediction system. Backpropagation (BP) neural network and the Levenberg–Marquardt (LM) algorithm were integrated to establish a prediction system for LGP processing by CO2 laser, with the controlling factor as the input parameter, and quality characteristics as the output parameters. After learning and training the network, the prediction error rate of the system was controlled within 5%, demonstrating good predictive validity.