Enhanced geometrical control in cold spray additive manufacturing through deep neural network predictive models

Abstract

Cold spray additive manufacturing is a deposition technique that facilitates the fabrication of large metal components with limited thermal effects, making it suitable for a wide range of industrial applications. Despite its potential, achieving precise geometrical control remains a bottleneck, hindering cold spray's establishment as a competitive additive manufacturing technology. This study introduces a computationally efficient framework that combines an adaptive slicing algorithm and process-specific toolpath planning strategies, designed to optimise deposit accuracy and material efficiency with respect to the Standard Tessellation Language (STL) model of the part to fabricate. Central to this approach is the integration of predictive models for cold spray deposition, which utilise deep neural networks trained on data from physics-based analytical models. These models offer rapid and accurate predictions of single-track cross-sections and full 3D shapes. The adaptive slicing algorithm dynamically adjusts layer thickness based on local curvature variations, ensuring improved geometrical fidelity while minimising material waste. Additionally, the toolpath planning methodology ensures continuous deposition, effectively addressing challenges such as surface waviness and edge losses. Validated against experimental data, the framework demonstrates significant improvements in efficiency and accuracy over conventional approaches, paving the way for broader adoption of cold spray additive manufacturing in complex industrial applications.