A team from the University of Notre Dame has demonstrated a machine learning-assisted 3D printing process to fabricate thermoelectric materials with unprecedented performance. By combining extrusion-based printing with Bayesian optimization, they produced bismuth antimony telluride (BiSbTe) structures exhibiting a room-temperature zT of 1.3, the highest value reported for printed thermoelectrics.
Published in the Journal of Materials Chemistry A, the study introduces a data-driven strategy to rapidly identify optimal ink formulations and printing parameters. The team employed Gaussian Process Regression (GPR) to predict thermoelectric power factors and Support Vector Machines (SVM) to enforce printability constraints, ensuring high geometric fidelity. The optimized inks, comprising 83 wt% BiSbTe particles, 0.5 wt% Xanthan gum in water, with 1.4 mm filament spacing and 1.0 mm standoff distance, enabled the creation of intricate 3D structures like curved tubes and lattices, critical for applications such as waste heat recovery on irregular surfaces.
Machine learning accelerates thermoelectric development
The team’s Bayesian optimization framework addressed a critical challenge in thermoelectric printing: the competing demands of particle loading (which boosts conductivity but reduces printability) and rheological modifiers (which improve shape fidelity but increase porosity). By modeling the four-dimensional parameter space, the algorithm identified optimal conditions that human intuition might have missed.
The Gaussian Process Regression (GPR) model showed strong predictive accuracy for power factors, where predicted and experimental values are closely aligned across multiple optimization rounds. Bayesian optimization reduced the experimental iterations needed to identify optimal parameters, though the paper does not quantify the exact time savings.”
Scalable and shape-conforming devices
The researchers demonstrated printing of complex heat-exchanger inspired architectures, including 60° inclined tubes capable of conforming to curved surfaces, hexagonal honeycomb lattices for enhanced mechanical stability, and multi-layer spiral structures optimizing surface-area-to-volume ratios.
Post-processing via hot isostatic pressing (HIP) at 480°C under 200 MPa pressure preserved the printed geometries while enhancing material density. SEM/EDS analysis revealed tellurium segregation at grain boundaries, a microstructural feature contributing to the high zT. The paper notes that HIP achieved comparable thermoelectric performance to cold-pressed samples while better maintaining structural integrit, though it does not specify exact dimensional deviation or density percentages.
The water-based ink system offers potential cost advantages over organic solvent-based alternatives by eliminating expensive solvents and simplifying processing. The use of aqueous chemistry and Xanthan gum suggests material savings compared to conventional approaches.
Machine learning and 3D printing optimization
Machine learning is increasingly being used to optimize 3D printing processes across materials and applications. Researchers at the University of Toronto recently introduced a machine learning framework to optimize the laser parameters in metal additive manufacturing, improving quality and reducing material waste. In the energy sector, Bambu Lab collaborated with researchers to develop an AI model that forecasts wind turbine blade production based on 3D printing performance metrics.
Elsewhere, a separate study demonstrated how machine learning can trace the origin of 3D printed parts with 98% accuracy, offering new capabilities for quality assurance and traceability. These advancements highlight how AI tools are becoming essential for predictive control and smart manufacturing in additive processes.
What 3D printing trends should you watch out for in 2025?
How is the future of 3D printing shaping up?
To stay up to date with the latest 3D printing news, don’t forget to subscribe to the 3D Printing Industry newsletter or follow us on Twitter, or like our page on Facebook.
While you’re here, why not subscribe to our YouTube channel? Featuring discussion, debriefs, video shorts, and webinar replays.
Featured image shows printed 3D complex structures. Image via Song et al./Journal of Materials Chemistry A.
