Deep Neural Network-based Reliability Assessment for Adaptive Gripper Fingers Utilizing a Mechanical Neural Network Lattice Structure

Abstract

In this study, a reliability assessment was performed, considering physical uncertainties on a conceptual design of adaptive gripper fingers comprising a mechanical neural network (MNN) lattice structure to achieve a target displacement. The feasible design domain was efficiently derived using topology optimization by applying various gripper operating conditions, considering environmental factors. Accordingly, a finite element model was designed based on an MNN lattice structure. Additionally, a deep neural network model was constructed based on a data set derived using the optimal Latin hypercube design method. The reliability assessment was performed by using Monte Carlo simulations and applying the structural safety factor and the stroke, which is the maximum distance between the gripper fingers, as reliability conditions. Thus, the reliability of the adaptive gripper fingers design was analyzed for each case, and the applicability of an actual work site was evaluated by considering physical uncertainties.