Machine learning-based prediction of ultrafast laser ablation in nickel using early-stage transient density evolution

R Haq and T Islam and MM Hayder and S Chowdhury and KA Rahman, MODELLING AND SIMULATION IN MATERIALS SCIENCE AND ENGINEERING, 33, 085011 (2025).

DOI: 10.1088/1361-651X/ae1e1b

This study presents a multi-layer long short-term memory neural network framework for predicting ultrafast laser ablation outcomes using only early-stage spatiotemporal density evolution and a dimensionless scalar encoding relative fluence. Unlike conventional models that depend on fixed laser and material parameters, the proposed approach enables accurate prediction of ablation depths and 2D crater profiles from a single reference simulation. Systematic evaluation across three temporal windows (30, 50, and 100 ps) reveals that the 50 ps model offers the optimal trade-off between accuracy and efficiency, capturing key ablation physics, from stress-induced spallation to phase explosion, with R-2 > 0.97 and mean absolute percentage error < 1.2%. The framework achieves nearly two orders of magnitude reduction in computational cost while enhancing spatial resolution and showing excellent agreement with experimental crater profiles in nickel over a wide fluence range of 425-2600 mJ cm(-2). By leveraging fluence-modulated transient density as a universal physical observable, this method offers a scalable path toward material-agnostic, real-time modeling tools for laser-based process design in advanced manufacturing.

Return to Publications page