Prediction of surface roughness of laser selective metallization of ceramics by multiple linear regression and artificial neural networks approaches

Abstract

Laser selective metallization of ceramics (LMC) has been widely applied to prepare high-quality electrical patterns on ceramic circuit carriers. As the dominating factor, surface roughness after laser treatment is controlled mostly by the laser technical parameters, such as laser power, scanning velocity, and laser frequency. Therefore, establishing an accurate relation between the surface roughness and the laser parameters will be of great benefit to the LMC specimens. In the present research, machine learning is used to simulate the LMC process on an alumina-copper oxide ceramic. The effect of each laser parameter and the interaction between them are revealed by the multiple linear regression method. The artificial neural network model trained with the Levenberg–Marquardt function provides the best estimation of the surface roughness after laser treatment compared with the Bayesian-regularization function and the scaled-conjugate-gradient function. The result can be used as a practical prediction and reasonable guideline for the optimization of LMC processes.