Development of bio-inspired flexible artificial skin and sensory information processing using a Kohonen artificial neural network

Abstract

Artificial skin is a synthetic membrane structure that mimics the flexibility and sensory functions of biological skin. Similar to receptors in biological skin sending signals to neurons in the brain, artificial skin needs sensors capable of converting information into electrical signals and transmitting them. Artificial skins are of increasing interest for prosthetics, soft robotics, virtual reality, wearable devices, and emerging medical applications, such as potentially reducing the number of amputations due to foot ulcers found in 25% of diabetic individuals. In this work, we successfully demonstrated a selforganizing map based on Kohonen artificial neural network to mimic how a brain distinguishes and pinpoints stimuli on skin. An artificial skin was developed from low-methoxyl pectin, a natural substance found in plants, which resulted in flexible films whose electrical conductivity increased with temperature. After optimization, with 10V bias, a nearly 10,000% increase in current was obtained as temperature increased from room temperature to T≈78 °C. Differential voltage measurement data was used to train the Kohonen artificial neural network. Various learning rates, sigma values, and different iterations were investigated, and results in a representative two-dimensional map were successfully obtained, reflecting the topology of the pectin artificial skin with a distinct hot spot. The results demonstrate the ability of this combined method of a few electrodes and Kohonen maps to mimic how a brain distinguishes and pinpoints stimuli on the skin.