A Novel Method for Pore Morphology Prediction: Multi-Ghost Encoder Generative Network

Metal additive manufacturing technology is widely used in aerospace, defense and industrial manufacturing due to its utilization rate, customizable processing of complex structural parts and other advantages. However, due to its specific manufacturing process, there are still many quality problems in the products, especially in the repeatability and reliability of the deficiencies, seriously restricting its large-scale industrial production and application, of which pore defects for the most common defects, so there is an urgent need to develop quality control technology to improve the quality of additive manufacturing. In this paper, the complex correlation and intrinsic mechanism between pore morphology and various processing parameters in metal additive manufacturing process are deeply studied and analyzed. By proposing a multi-ghost encoder generative network, we aim to solve the problem of pore morphology prediction in the laser powder bed fusion process. An exhaustive pore morphology image dataset covering key processing parameters is used for the experiments. During the experiments we compare the performance of our method with existing methods in terms of porosity reconstruction, prediction accuracy and computational efficiency. Our method is effective in predicting and reconstructing different pore morphologies and is improved in terms of the number of parameters and training efficiency because of the Ghost module structure. The results of this study not only provide new perspectives on porosity control in metal additive manufacturing processes, but also open up new possibilities for deep learning applications in manufacturing.