Additive manufacturing (AM) has progressed from rapid prototyping to the production of functional components for aerospace, propulsion, energy and other demanding engineering applications. However, one fundamental challenge remains: the relationship between manufacturing parameters, defects, microstructure and final component performance is highly nonlinear. Laser power, scan speed, layer thickness, thermal history, material composition and cooling conditions can interact to produce porosity, lack of fusion, cracking and residual stresses.

Machine learning (ML) provides a potential route through this complexity. Recent research demonstrates that data-driven models can predict material properties and identify relationships that are difficult to establish using conventional empirical approaches. A 2026 Scientific Reports study, for example, developed a multi-target ML approach for predicting the mechanical properties of FDM-printed polymer components, specifically addressing variability associated with processing and material behaviour.

The next stage is increasingly moving toward physics-informed machine learning (PIML). Rather than treating experimental or simulation data as the only source of information, physics-informed approaches incorporate known physical constraints into the learning process. Conservation laws, thermodynamic relationships, material behaviour and defect mechanisms can therefore become part of the predictive framework.

This is particularly important for aerospace additive manufacturing. NASA's AM programmes have investigated alloys and processes including laser powder-bed fusion and directed-energy deposition for demanding propulsion and space applications. These programmes have highlighted the importance of understanding not only manufacturing processes but also temperature-dependent mechanical and thermophysical properties.

Recent research is pushing the concept further. A 2026 Nature Communications study introduced a knowledge-informed graph-attention framework for designing alloys suitable for laser additive manufacturing. The researchers integrated physical metallurgy knowledge and uncertainty associated with defect distributions into ML-based alloy design and demonstrated the approach using new printable nickel and aluminium alloys.

The significance extends beyond simply predicting whether a component will contain a defect. The longer-term objective is to establish a causal chain:

manufacturing conditions → defect formation → microstructural evolution → mechanical response → failure probability.

If this relationship can be reliably learned while respecting physical constraints, additive manufacturing could evolve from a parameter-optimisation technology into a genuinely predictive manufacturing science.

References:

• Xu et al., Nature Communications (2026), “Knowledge-informed graph attention networks enable defect-free alloy design for laser additive manufacturing.”

• Naidu et al., Scientific Reports (2026), “Multi-target machine learning for predicting mechanical properties of FDM-printed polymer components.”

• NASA, Maturation of Additive Manufactured Aerospace Alloys and Development of Mechanical and Thermophysical Properties for Space Applications.