Prediction and inverse design for ultrafast laser colorization on aluminum via color-morphology coupled neural network

Laser surface colorization has been regarded as a promise technology for inkless printing due to its advantages such as versatile, stable and environmentally friendly. However, the further applications of laser colorization in real industrial manufacturing are limited by inevitable trial and error procedure in experiment. Here, a prediction and inverse design of laser surface colorization process are realized to address this challenge by using a Color-Morphology Coupled Neural Network (CMCNN). The forward prediction is conducted by a Physical-guided Multilayer Perceptron (PG-MLP) with improved prediction accuracy. For solving the non-uniqueness of inverse design procedure, a Convolution Neural Network (CNN) based on the input physical image is utilized. This enables the combining of structural color and morphology information in input physical image for processing parameters planning. The network performance is verified through the fabrication of portrait and artwork image. The corresponding relationship between target color and experimental results demonstrate the validity of proposed model. This approach may open new possibilities for the fast and precise surface colorization, and facilitate the development of laser structure fabrications in practical engineering applications.