A data-centric approach to generative modelling for 3D-printed steel.

Area of Science:

• Materials Science
• Mechanical Engineering
• Statistical Modeling

Background:

• Additive manufacturing (AM) enables complex metallic components, disrupting traditional engineering.
• Metallic AM parts show significant geometric and mechanical property variations, hindering widespread adoption.
• Current extensive post-manufacture testing is required to ensure safety standards for AM components.

Purpose of the Study:

• To develop a generative statistical model for predicting the quality of additive manufactured steel designs before production.
• To quantify and understand the intrinsic variations in geometric and mechanical properties of metallic AM materials.
• To reduce the need for extensive post-manufacture testing by enabling pre-manufacture quality prediction.

Main Methods:

• An interdisciplinary approach combining probabilistic mechanics and uncertainty quantification.
• Development of a generative statistical model to describe intrinsic variation in AM steel.
• Characterization of geometric variation using an anisotropic spatial random field with oscillatory covariance.
• Modeling mechanical behavior with a stochastic anisotropic elasto-plastic material model.

Main Results:

• Intrinsic variation in AM steel can be accurately described by a generative statistical model.
• Geometric variations are modeled as an anisotropic spatial random field.
• Mechanical behavior is captured by a stochastic anisotropic elasto-plastic model.
• The generative model was validated on an independent experimental dataset.

Conclusions:

• A generative statistical model can predict the quality of AM steel designs pre-manufacture.
• Combining statistical and physics-based modeling is crucial for characterizing new AM steel products.
• This approach can mitigate barriers to adopting metallic AM by reducing uncertainty and testing requirements.