Machine learning-enabled feature classification of evaporation-driven multi-scale 3D printing.

The freeform generation of active electronics can impart advanced optical, computational, or sensing capabilities to an otherwise passive construct by overcoming the geometrical and mechanical dichotomies between conventional electronics manufacturing technologies and a broad range of three-dimensional (3D) systems. Previous work has demonstrated the capability to entirely 3D print active electronics such as photodetectors and light-emitting diodes by leveraging an evaporation-driven multi-scale 3D printing approach. However, the evaporative patterning process is highly sensitive to print parameters such as concentration and ink composition. The assembly process is governed by the multiphase interactions between solutes, solvents, and the microenvironment. The process is susceptible to environmental perturbations and instability, which can cause unexpected deviation from targeted print patterns. The ability to print consistently is particularly important for the printing of active electronics, which require the integration of multiple functional layers. Here we demonstrate a synergistic integration of a microfluidics-driven multi-scale 3D printer with a machine learning algorithm that can precisely tune colloidal ink composition and classify complex internal features. Specifically, the microfluidic-driven 3D printer can rapidly modulate ink composition, such as concentration and solvent-to-cosolvent ratio, to explore multi-dimensional parameter space. The integration of the printer with an image-processing algorithm and a support vector machine-guided classification model enables automated, in-situ pattern classification. We envision that such integration will provide valuable insights in understanding the complex evaporative-driven assembly process and ultimately enable an autonomous optimisation of printing parameters that can robustly adapt to unexpected perturbations.