Deep learning driven prediction method for roughness and hardness of laser in-situ forging additive manufacturing Ti-6Al-4V

Laser powder bed fusion suffers from surface quality, thermal stress, defects, and poor fatigue performance in Ti-6Al-4V components due to unstable melting/solidification. Based on this, this research proposes laser in situ forging additive manufacturing (LIF-AM). By using an ultrafast laser to impact the molten layer in situ, the performance of additive manufacturing components can be improved. Addressing complex parameters and unclear mechanisms, this research constructs a deep learning model based on a CNN (Convolutional Neural Network)-Transformer to predict the surface roughness and hardness of LIF-AM components. The CNN’s dynamic sliding kernels extract local interactions, while the Transformer’s self-attention captures global dependencies. This resolves the nonlinear coupling and long-range correlation challenges in high-dimensional parameter space. In the experimental verification, the coefficient of determination R2 is used to measure the degree of fit between the model's predicted values and the actual values. For predicting roughness and hardness, the model achieves R2 values of 0.922 and 0.920, respectively, with mean absolute errors of 0.095 and 3.824, respectively. This research provides a data-driven intelligent solution for the optimisation of the LIF-AM process, significantly reducing the traditional trial-and-error cost and promoting intelligent manufacturing of high-performance titanium alloy components in fields such as aerospace.