Thermal-node-link fast-FE model for deformation suppression in LWAM of thin-walled structures with machine learning assisted parameter optimization and prediction

Abstract

Coaxial laser Wire Deposition Additive Manufacturing (LWAM) offers a cost-effective, high-deposition-rate solution for fabricating complex thin-walled aerospace and automotive components but suffers from deformation induced by thermal residual stresses. Conventional finite element models (FEM) for deformation prediction are computationally prohibitive (>5 h/100 layers), limiting industrial adoption. This study proposes a thermal-node-link Fast-FEM model for high-speed thermo-mechanical simulation of LWAM processes. By representing printed structures as simplified nodal networks with conductive links and incorporating a double ellipsoidal heat source, radiation, and conduction boundary conditions, this approach reduces computational time by 17.1× (thin walls) to 19.2× (ring geometries) while maintaining accuracy. Simulations of thin-wall and circular ring geometries demonstrate <1.9% deviation in temperature distribution and thermal deformation predictions compared to experimental data. To suppress deformation, it implements a strategy combining periodic cooling cycles and adaptive power modulation during deposition. An artificial neural network (ANN) trained on 64 experimental datasets optimizes control parameters (start/final layer power, cooling duration, inter-layer interval), predicting peak melt-pool temperature within 6.2% error and deformation with 8.1% accuracy (ANN: 0.34 mm vs experimental: 0.37 mm). Experimental validation confirms strong alignment between accelerated Fast-FEM simulations, ANN predictions, and test results of the samples. This integrated framework reduces computational costs by 90% while achieving up to 41% deformation suppression in benchmark geometries, enabling high-precision deposition of LWAM components.