Machine-learning assisted optimization of process parameters for controlling the microstructure in a laser powder bed fused WC/Co cemented carbide

Abstract

A Single laser scan was performed on a sintered WC/Co composite and the process conditions were optimized to realize a WC/Co two-phase microstructure in a WC/Co composite fabricated by laser powder bed fusion ( L -PBF). Laser irradiation on the sintered WC/Co composites generated two different microstructural regions. One is composed of WC and Co phases (WC/Co two-phase region), and the other includes W 2 C and W 3 Co 3 C phases (WC decomposition region). In the L -PBF-fabricated WC/Co composites, these two microstructural regions are distributed in a complex manner and their fractions varies depending on the laser parameters. Using a convolutional neural network model that is trained with the micrographs of single-laser-scanned WC/Co composites, the fractions of the two microstructural regions in the WC/Co composites fabricated by the L -PBF under various laser conditions were quantified. Multiple linear regression analyses revealed that the laser power and spot size contribute more than the scan speed, which has been verified by numerical analysis based on the moving heat source model. A support vector machine (SVM) suggested that low laser power and low scan speed are required for suppressing the WC decomposition in the consolidated samples. The WC decomposition has been successfully suppressed in the samples fabricated under SVM-recommended conditions. • Parameter optimization of laser powder bed fused WC/Co composite was performed. • The parameter optimization was carried out for controlling the microstructure. • A convolutional neural network (CNN) was used to segment the microstructure. • The micrographs of single laser-scanned samples were trained to the CNN. • A process window was explored by a support vector machine.