A machine learning workflow for 4D printing: understand and predict morphing behaviors of printed active structures

Abstract

Understanding and predicting morphing response of printed active structures remain a challenge in 4D printing. To tackle it, in this paper, we present a consolidated data-driven approach enabled by an ensemble of machine learning (ML) algorithms. First, three ML algorithms were employed to quantitatively correlate a geometrical feature (thickness) with the final morphing shapes indicated by curvatures and curving angles. Among them, the gradient boosting algorithm achieved correlation factors (R 2) of 0.96 and 0.94 when predicting the curvatures and curving angles by using the data collected from 150 experiments. The random forest model enabled to rank the importance of fabrication parameters in determining the shape morphing behaviors. To forecast the dynamic response of printed structures, three time series forecast algorithms were implemented based on the time-dependent image data during morphing processes of the printed active structures. Among them, the exponential smoothing method achieved an average mean absolute percentage error of 0.0139. This work offers a proof-of-concept on how the ensemble ML algorithms can be employed to delineate and predict morphing mechanism of printed active structures, thus providing a new paradigm for advancing the state-of-the-art research in 4D printing.

The Design of 4D‐Printed Hygromorphs: State‐of‐the‐Art and Future Challenges

“…For 4D-printed structures, neural network algorithms map the influence between the input parameters highlighted in the previous sections (i.e., material distribution, geometry, and fiber orientation) and the actuation of the structure (bend angle, curvature, position, and displacement). The neural network algorithm could be applied to experimental data [304,346] or to numerical models. [139,322,323] However, if model-based optimization is to be implemented, the model efficiency must be pre-emptively assessed.…”

A brief review of mechanical designs for additive manufactured soft materials

“…Sossou et al [139] developed a computational design tool that integrated the design, modeling and inverse optimization of complex 4D shape morphing structures. Zhang et al [141] used a convolutional neural network to quickly predict the large deformation of digital active materials. In the work of Su et al [142] , machine learning was adopted to establish the correlation between manufacturing parameters and experimentally obtained curvatures of shape morphing SU-8 strips.…”

“…Additive manufacturing has broken the shackles of traditional manufacturing technology and brought new bottom-up manufacturing concepts, thereby offering researchers numerous opportunities to design soft . (B) Convolutional neural network for 4D shape-morphing beams [141] . (C) 4D shape morphing wing designed by evolutionary algorithm [139] .…”

Mechanomaterials: A Rational Deployment of Forces and Geometries in Programming Functional Materials

“…[299] Data-driven machine learning (ML) may serve as an alternative solution for this problem. ML in general materials discovery has been recruited for nonlinear and time-dependent materials mechanics, [300] model-free identification of solid mechanics, [301] prediction of grain size, [302] coarse-grained dislocation microstructures [303] and bandgaps, [304] optimization of lattice structures, [305] complex design and process optimization of additive manufacturing, [306,307] shape memory alloy development, [308] nano-bio interactions, [295] smart mechanical sensors, [309] etc.…”