Abstract
Additive manufacturing (AM) presents an alternative to traditional subtractive technologies for fabricating materials that are difficult to machine, such as titanium alloys. Among various AM techniques, laser metal deposition (LMD) has attracted significant research interest due to its versatility in the processing of metallic materials. The process is easily scalable to produce more prominent parts, offers high design freedom, and can produce dense parts with good metallurgical bonding. In LMD, components are fabricated additively by successive layer deposition, each layer comprising multiple individual tracks. These single tracks serve as fundamental building blocks for the final structure being fabricated. Despite its considerable benefits for industrial applications, LMD is susceptible to material defects, often originating from suboptimal process parameter settings. Therefore, a comprehensive understanding and optimisation of the deposition process should begin at the most fundamental scale, at the single-track level. Geometric characteristics such as dilution, height, and width, as well as the resulting microstructural morphology, are critical as they directly influence the mechanical performance and overall build quality of the deposited material. Due to the high cooling rates and steep temperature gradients in the LMD process, coarse and columnar grains, which lead to anisotropic mechanical properties, have been a challenge. Large dilution leads to excessive penetration and melt pool superheating, risking epitaxial growth. The introduction of ceramics, combined with careful control of process parameters, has proven to promote heterogeneous nucleation and equiaxed grain growth. In this study, the geometric characteristics of TiC/Ti6Al4V single tracks fabricated via LMD were optimised using an artificial neural network (ANN) model; the optimisation criteria were primarily based on achieving minimal dilution, good metallurgical bonding, thick heights and dense cladding, and secondarily, achieving equiaxed grain growth with homogeneous dispersion of in-situ TiC phases. The model was used to predict geometric characteristics, i.e. dilution, width, and height, using laser power, scan speed, and powder feed rate as inputs. The model achieved an MSE of 0.00211 and R2 score of 0.986 in the training phase, and an MSE of 0.00719 and an R2 score of 0.885 in the testing phase. The model was found to be effective upon testing and was subsequently deployed to generate prediction data. From which process surface plots and a processing map were created to relate process-parameter combinations to resulting geometric characteristics. Interpretation of the process plots, along with further examination of the deposited samples, revealed that those...