Shape optimization of spline-based punches for enhanced dry adhesion via a narrow neural network surrogate

The adhesion between a rigid substrate and spline-defined elastic punches is investigated with focus on the asymptotic normal stress near the contact edge. Interfacial stress distributions are obtained from automated finite element simulations in Abaqus using Python scripting. Based on these results, a surrogate-based optimization framework is developed in which a narrow feedforward neural network predicts the characteristics of the asymptotic normal stress solution from the punch geometry. The asymptotic normal stress solution allows for the semi-analytical determination of the energy release rate for edge-initiated interfacial defects, which is a primary governing factor of adhesion strength. The optimization aims to minimize the asymptotic normal stress. The results show that the proposed approach accurately captures the highly nonlinear geometry-stress relationship and identifies spline-based punch geometries with improved adhesion performance compared to a straight cylindrical punch. This framework can be used to design new geometries with improved adhesion characteristics. The data and code will be provided in the Supplemental Material for reproducibility and further exploration.