Speaker
Description
The inverse design of shape-morphing structures based on responsive polymeric materials, requires the development of theoretical and computational approaches capable of reproducing the involved physical phenomena. When gel-based morphing elements are considered, shape change capabilities can be easily obtained by harnessing their capability to react to external stimuli, such as humidity, temperature, PH variations, etc. These stimuli, if precisely spatio-temporally controlled, can be used to deform a structure from one configuration to another.
Recent advances in additive manufacturing enabled to easily obtain such systems; however, their design is often affected by multiple sources of uncertainty, including material properties, model inadequacy, and geometric errors.
In this work, we present an uncertainty-aware inverse design framework based on Machine Learning and probabilistic modelling for gel-based shape-morphing structures. The proposed approach integrates Approximate Bayesian Computation (ABC) to explicitly account for model-form and parameter uncertainties within the inverse design process. Neural network surrogates are trained to emulate the forward response of complex shape-morphing systems, including time-dependent deformation processes arising from swelling and diffusion phenomena. This enables the efficient incorporation of transient system dynamics within the Bayesian inference process.
As a representative application, a heterogeneous elastic tube embedding a swelling gel core is investigated, where swelling-induced forces drive shape change. The framework identifies spatial distributions of material properties required to match prescribed target shapes, while explicitly quantifying the impact of uncertainty and noise on the design outcome.
The results demonstrate that incorporating uncertainty significantly improves the robustness and reliability of ML-based inverse design, providing valuable insights for the development of shape-morphing systems with enhanced predictability.