A Systematic Design Approach for Magneto-Active Elastomers with Bayesian Network Classifiers Considering Manufacturing Variability

Abstract Smart materials have demonstrated great potential for applications requiring flexibility, shape morphing, and reconfigurability. Among these, magneto-active elastomers (MAEs) are especially promising, achieving relatively large actuation deflections and forces under non-contact magnetic fields with rapid response times. Optimizing the structure and material properties of such advanced materials typically requires a systematic design strategy. Most approaches employ forward design methods, which are time intensive as they evaluate multiple design variables and the resultant performances to find optimal solutions. In this work, an inverse design strategy is proposed for MAE to efficiently identify the feasible design space that meets predefined performance requirements, such as actuated displacement, force thresholds, and actuation energy. Using a supervised Bayesian network classifier (BNC) adapted to training data and user-defined performance thresholds, this approach identifies a feasible design space capable of achieving customized performance goals. The BNC model is further enhanced by incorporating the manufacturing variability of additive manufacturing (AM) processes, specifically thickness variation. The design strategy is demonstrated with an MAE unimorph, using training data from an analytical model based on classic beam theory with free deflection and blocked force as performance metrics. The BNC mapping of the feasible design space demonstrates a resultant lower bending stiffness for displacement targets and a higher bending stiffness for the actuation force threshold. The feasible design space is further reduced to satisfy performance targets and manufacturing reliability. This approach serves as a powerful tool for identifying high performing designs that account for manufacturing variability for advanced material.