Abstract
Metallic additive manufacturing (AM) technologies can be utilized to manufacture components with tailored and desired microstructure and properties. However, this is just a promise. Fast and accurate computational methodologies are required to optimize the process parameters to achieve the desired microstructure and eliminate the trial-and-error process optimization methodologies. In this paper, a surrogate machine learning model is implemented to design the additive manufacturing (AM) process parameters, including scan speed, laser power, and scanning strategies, to achieve desired microstructure during AM of Ti-6Al-4V. Temperature gradient and solidification rate are the factors that control the target solidification structure, i.e., columnar to equiaxed transition (CET) of grains, to achieve desired mechanical properties. A multi-track finite element model is established and validated to simulate the temperature field during AM process using different scan strategies within a layer. As a result of this process simulation, a dataset containing temperature gradient (G), maximum temperature (Tmax), and solidification rate (R) is extracted from a collection of temperature distribution graphs by using an automated signal processing technique. Next, various machine learning methods (e.g., Boosting methods, multilayer perceptron, support vector regression, and random forest) are evaluated to design process parameters through a cross-validation procedure. The surrogate model is then validated against experimentally obtained microstructures during the solidification path.