Force Calibration and Prediction of Soft Stretch Sensor Based on Deep Learning

Abstract

Soft stretch sensors are increasingly used in wearable devices and flexible exoskeleton. This paper presents a novel sensing-actuation integrated unit for elastic tension transmission and force estimation. The unit consists of a capacitive sensor, four elastic bands, which can provide enough stiffness, and two stretchable paper-cut fabric shielding layers, which can greatly shield the external interference. The mechanical and electrical properties of the unit were tested on a universal material testing machine and then a simulation test platform was designed to generate the sine curve with different travels and stretch rates. A great amount of data with a total of 35 cases were collected to train our models. Results demonstrated mean square error (MSE) less than 0.21 N2, normalized root mean square error (NRMSE) less than 1.7% for the selected calibration model, and MSE less than 0.28 N2, NRMSE less than 2.0% for the selected prediction model. Our unit together with its calibration and prediction methods in this paper holds great promise in applications such as lightweight flexible exoskeletons.