Process-structure multi-objective inverse optimisation for additive manufacturing of lattice structures using a physics-enhanced data-driven method

Abstract

Additive manufacturing (AM) has become a practical solution for fabricating lightweight and high-strength metallic lattice structures. The inverse optimisation of process-structure parameters to achieve high performance and minimised trial-and-error experiments has presented a persistent challenge. To address this problem, an inverse optimisation methodology has been proposed for coping with multiple conflicting performance objectives, consisting of mechanical properties and lightweight extent under AM-constraints. In the pursuit of greater accuracy, a physics-enhanced data-driven algorithm, i.e. encoding-stiffness-analysis multi-task Gaussian process regression, has been developed. This empowers us to precisely analyse how process-structure parameters impact the properties of AM-formed lattice structures. As an emerging machine learning method for AM, the physics-enhanced data-driven algorithm exhibits strong fitting capability and extrapolation performance, due to the interpretability provided by physical information. It has been applied as a surrogate model within the multi-objective genetic algorithm, facilitating the efficient design of parameters and the expansion of objective space. Notably, a deviation of less than 15% has been observed between predictive and experimental results, providing solid confirmation of our methodology's reliability. This confluence of physical insights and data-driven modelling holds substantial promise for accelerating the development of highly efficient designs.

Keywords

• additive manufacturing
• machine learning
• physics-enhanced gaussian process
• multi-objective optimisation
• inverse design