(108e) A Physics-Informed Machine Learning Framework for Autonomous Additive Manufacturing of Functional Materials

Additive manufacturing (AM) is a transformative approach for fabricating complex geometries with enhanced design flexibility, material efficiency, and scalability [1]. Among AM techniques, direct ink writing (DIW) is particularly notable for its ability to construct multifunctional 3D structures at meso- and microscale resolutions in a single step [2,3]. Its compatibility with a broad range of materials—including polymers [4,5], ceramics [6,7], and bio-based composites [8,9]—positions DIW as a versatile platform for applications requiring high precision and functional customization. However, DIW presents challenges due to the nonlinear and stochastic behavior of ink flow and deposition, which significantly affect the printed geometry, microstructure, and functional properties of printed materials. Conventional trial-and-error-based process optimization is inefficient and cost-prohibitive. To address these limitations, this work presents a physics-informed machine learning (PIML) framework for autonomous optimization and real-time control of DIW processes.

The framework integrates experimental, computational, and data-driven methods. A robotic DIW platform is developed using a syringe-based dispensing system integrated with a programmable robotic arm to enable precise material deposition. Polydimethylsiloxane (PDMS) ink is utilized to investigate the effects of key process parameters—including deposition speed, ink pressure, and nozzle-substrate distance—on line morphology, thickness, and resolution. Experimental observations reveal that reduced deposition speed improves interlayer adhesion and uniformity, while excessive speed introduces defects associated with flow instability. Maintaining an optimal ink pressure is critical for balancing flow rate and structural fidelity; deviations often result in discontinuities or over-deposition. A hybrid multiscale modeling framework is developed to describe the physical behavior of the system, combining microstructural insights with continuum-scale simulations. This framework captures the spatiotemporal evolution of ink characteristics such as viscosity, diffusion, and phase transition under varying printing conditions.

To improve process prediction and adaptability, a deep recurrent neural network (DRNN) is trained using experimental data and physics-based simulation results. The DRNN model captures time-dependent deposition dynamics and enables the estimation of intermediate variables that are difficult to model analytically. This predictive capability underpins a Bayesian optimization (BO) algorithm, which iteratively refines the PIML model using new observations to efficiently explore the parameter space and identify optimal operating conditions. The integrated PIML-BO approach functions as a digital twin of the DIW process, facilitating autonomous monitoring, real-time prediction, and optimization. This framework enhances defect mitigation, dimensional fidelity, and functional performance, contributing to the advancement of scalable and precise additive manufacturing of functional materials.

References:

[1] Huang, S. H.; Liu, P.; Mokasdar, A.; Hou, L. Additive Manufacturing and Its Societal Impact: A Literature Review. Int. J. Adv. Manuf. Technol. 2013, 67 (5–8), 1191–1203.

[2] Lewis, J. A. Direct Ink Writing of 3D Functional Materials. Adv. Funct. Mater. 2006, 16 (17), 2193–2204.

[3] Smay, J. E.; Gratson, G. M.; Shepherd, R. F.; Cesarano, J.; Lewis, J. A. Directed Colloidal Assembly of 3D Periodic Structures. Adv. Mater. 2002, 14 (18), 1279–1283.

[4] del Campo, A.; Arzt, E. Generating Micro- and Nanopatterns on Polymeric Materials. Gener. Micro- Nanopatterns Polym. Mater. 2011, 369.

[5] Maguire, A.; Pottackal, N.; Saadi, M. A. S. R.; Rahman, M. M.; Ajayan, P. M. Additive Manufacturing of Polymer-Based Structures by Extrusion Technologies. Oxford Open Mater. Sci. 2020, 1 (1).

[6] Minas, C.; Carnelli, D.; Tervoort, E.; Studart, A. R. 3D Printing of Emulsions and Foams into Hierarchical Porous Ceramics. Adv. Mater. 2016, 28 (45), 9993–9999.

[7] Muth, J. T.; Dixon, P. G.; Woish, L.; Gibson, L. J.; Lewis, J. A. Architected Cellular Ceramics with Tailored Stiffness via Direct Foam Writing. Proc. Natl. Acad. Sci. U. S. A. 2017, 114 (8), 1832–1837.

[8] Diederichs, E. V.; Mondal, D.; Patil, H.; Gorbet, M.; Willett, T. L. The Effect of Triglycerol Diacrylate on the Printability and Properties of UV Curable, Bio-Based Nanohydroxyapatite Composites. J. Mech. Behav. Biomed. Mater. 2024, 153, 106499.

[9] Chen, J. X. M.; Chen, T.; Zhang, Y.; Fang, W.; Li, W. E.; Li, T.; Popovic, M. R.; Naguib, H. E. Conductive Bio-Based Hydrogel for Wearable Electrodes via Direct Ink Writing on Skin. Adv. Funct. Mater. 2024, 34 (40), 2403721.