A data-driven model for pressure distribution measurements by a four-electrode polymer sensor

Abstract

Machine learning techniques have significantly enhanced signal handling and prediction accuracy in electronic skins by facilitating the extraction of useful information hidden in the sensory outputs. We present a polymer sensor with four irregularly shaped electrodes enabling energy-efficient sensing and improved data interpretation. We first compute the resistance change for the sensing element under pressure. The finite element method is used to solve the three-dimensional nonlinear elasticity. The electric potential distribution is simulated using an arbitrary Lagrangian-Eulerian formulation. We then build reduced-order models for detecting pressure distribution for different pressure cases. The inverse models built using deep neural networks showed good prediction accuracy and resolution. The irregular arrangement of the electrodes resulted in low correlation coefficients between the input resistances, and therefore, efficient predictions of the four-electrode sensor. It is demonstrated that the present four-electrode sensor could replace at least four sensors in an array. For arbitrary pressure distributions over a 2 × 2 surface resolution, a model is trained with a mean accuracy of 22 Pa in the range of 1–20 kPa. Additionally, for a single square-shaped pressure with arbitrary magnitude and surface area, position prediction accuracies of 99% and 96% are obtained at 4 × 4 and 8 × 8 resolutions, respectively. Moreover, the models showed low sensitivity to the uncertainty in the measured signals.