Aluminium alloys process optimization in laser powder bed fusion using machine-learning algorithms

In this study, machine learning algorithms (MLAs), including gradient boosting and random forest, were used to optimize the process parameters of AlSi10Mg alloy in the laser powder bed fusion (L-PBF) process, significantly reducing substantial experimental effort and time. Through MLAs, the dimensions and geometries of the melt pool measured from experiments were categorized into various types of melt pool to further predict the depth‒width ratio and the statuses (i.e. lacking, balling, conduction and keyhole) of the melt pool. Unlike traditional optimization methods, which require extensive experimentation to develop a process map, MLAs enabled the creation of accurate process maps with far fewer experiments (e.g. 32 vs. 128 in conventional approaches). The accuracy of the MLA-created process map was validated by comparing the process map plotted from experimentally data, which their consistency presents the high feasibility and efficiency of creating process maps with limited data to shorten the time needed for process optimization.