Prediction of ground reaction forces using the artificial neural network from capacitive self-sensing values of composite ankle springs for exo-robots

Abstract

Exoskeleton robots, which are wearable robots that assist human movement, have recently been developed to augment the performance of the human body. In this study, the ankle spring of lower-limb exoskeleton robots was created from a foam core sandwich-structured composite (FCSC) to support the running movement by storing and releasing elastic energy and to predict kinetic data using the self-sensed capacitance data from the structure. Through finite element analysis, the deformation behaviors of the FCSC ankle spring were analyzed and compared with those of a typical fiber-reinforced plastic composite ankle spring. Using a universal test machine, the capacitive-based self-sensing capability of the FCSC ankle spring was examined under several loading rates and cyclic loading. Furthermore, running tests with the lower-limb exoskeleton robot equipped with the FCSC ankle spring were performed on a treadmill. Horizontal and vertical ground reaction forces (GRFs) and capacitance values were obtained from the force plate embedded in the treadmill and FCSC ankle spring, respectively, at various running speeds. These data were used to train an artificial neural network (ANN) model as input and output data. Consequently, the trained ANN model predicted the kinetic data using the capacitance values of the FCSC ankle spring without additional sensors.