A NOVEL PATTERN RECOGNITION METHOD FOR SELF-POWERED TENG SENSOR EMBEDDED TO THE ROBOTIC HAND

Abstract

This paper details the design and validation of a robotic hand resembling human anatomy, featuring triboelectric nanogenerator (TENG) sensors for enhanced shape and material identification. In contrast to conventional piezoelectric sensors—often susceptible to temperature shifts and high production expenses—TENGs offer a self-sustaining power source, simplified circuitry, economical fabrication, and robust durability. Leveraging these advantages, this work introduces a novel machine learning framework that converts sequential TENG sensor data into two-dimensional representations, subsequently analyzed by a 2D convolutional neural network (CNN). Comparative studies with a standard 1D CNN approach reveal marked improvements in performance: the 2D CNN model achieves classification accuracies of 98% for shape recognition and 99% for material discrimination, surpassing the respective 94% and 98% attained via 1D CNN. Integral to this methodology are TENG sensor fabrication, noise suppression measures, a custom robotic hand design, and associated control electronics. Real-time tests confirm the proposed system’s resilience and adaptability in unstructured environments, highlighting the promising role of integrating TENG sensors with advanced neural network architectures for autonomous, dexterous manipulation across a range of industrial applications.