Work

Additive Gaussian Process Regression for Predictive Design of High‐Performance, Printable Silicones

Year: 2026

Type: article

Abstract: Silicones are valued for durability, elasticity, low reactivity, and availability of biocompatible formulations, making them suitable for a wide range of applications. Additive manufacturing (AM), or 3D‐printing, enables production of complex, customized parts, but the limitations of AM‐compatible silicones restrict application‐based property tuning. This work introduces a material design framework for predictive formulation of UV‐curable polydimethylsiloxane (PDMS), a type of silicone, suitable for vat photopolymerization (VPP) and other UV‐based 3D‐printing processes. Additive Gaussian process regression models trained on custom PDMS formulations predict ultimate tensile strength and elongation at break of the cured polymer along with viscosity of the precursor resin. These properties serve as application and manufacturing driven screening criteria for computational formulation design. The PDMS formulations contain diverse combinations and quantities of siloxane base polymers, crosslinkers, and reinforcing fillers, while satisfying polymer network design constraints. Model structure is chemistry‐aware, using polymer network interactions to inform variable definitions and additive model components. The resulting models provide predictions of formulations of interest and support exploration of a diverse property range for UV‐cured and VPP PDMS.

Source: Advanced Engineering Materials

DOI: 10.1002/adem.202502575

Authors: Roxana Carbonell, Martha Leach, Hongtao Song, Carolyn Seepersad

Institution: Georgia Institute of Technology

Language: English

Topic: Machine Learning in Materials Science

Subfield: Materials Chemistry

Field: Materials Science

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: hybrid

Estimated APC paid: $4,520