Bayesian-driven modulation of recoverable strain in compliant auxetic cementitious composites for optimal energy harvesters

Abstract

Auxetic cementitious cellular composites (ACCCs) offer high deformability that is attractive for mechanical energy harvesting when integrated with flexible piezoelectric materials. However, the intrinsic brittleness of cement-based materials and the complex coupling between auxetic geometry and damage evolution hinder the efficient design of ACCC energy harvesters. This study proposes a novel learning-driven design framework that, for the first time, integrates a physics-based energy harvesting model with Bayesian Optimization (BO) to directly optimize the recoverable hinge-like strain capacity of ACCCs for enhanced electrical output. The optimization maximizes the voltage generated by piezoelectric materials bonded at hinge regions, while using constraints to prevent splitting failure and non-auxetic behavior under compression. The energy harvesting model combines the concrete damage plasticity (CDP) model for pre-compression damage with a secondary elastic model for cyclic loading, enabling prediction of recoverable strain in generalized ACCC geometries. The learning-driven approach proved far more efficient than random generation in identifying optimal ACCC configurations. Experimental validation of the optimized design achieved a peak-to-peak voltage of nearly 15.0 V per cycle, about 2.7 times higher than a reference design. This study provides a learning-driven approach to designing enhanced compliant auxetic cementitious energy harvesters for smart infrastructure applications.