Integrated approach for AlSi10Mg rapid part qualification: FEM, machine learning, and experimental verification in LPBF-based additive manufacturing process

An integrated approach combining finite element modelling, machine learning, and experimental verification was proposed for developing process maps to optimize the LPBF process for AlSi10Mg alloy. A transient thermal simulation model was validated to predict single-layer melt pool size by modifying laser beam power, scan rate, feedstock bed depth, and preheating of feedstock. Using the verified model, a pool of data was generated to develop a backpropagation neural network to predict melt pool dimensional ratios indicating printing defects. It was found that beyond process parameters, powder bed thickness and preheating temperature impacted defect formation. Excessively high preheating temperatures increased the lack of fusion defects by transforming melt pool dynamics from conduction to keyhole mode. Optimal combinations were identified as 30.0 μm thickness with 90.0 and 120.0 °C preheating and 50.0 μm thickness with 120.0 °C preheating. By reducing iterative physical testing, the integrated process mapping approach enables accelerated qualification of LPBF parameters for AlSi10Mg alloy.