Triboelectric Nanogenerator Based Smart Electronics via Machine Learning

TLDR

Artificial intelligence drives the need to equip traditional electronics with autonomous analysis, prompting the proposal of triboelectric nanogenerator‑based smart electronics that integrate automatic machine‑learning data analysis. The study aims to develop and evaluate a TENG‑based smart electronic system that automatically analyzes and recognizes voice and handwriting signals using machine‑learning algorithms. The authors fabricate a high‑sensitivity flexible TENG from porous PDMS and copper mesh, collect voice and handwriting signal datasets, and apply ensemble learning and medium‑Gaussian support vector machines for signal recognition, supported by hierarchical clustering to analyze letter relationships. The TENG achieves ≈0.2 V output for voice and handwriting detection, with machine‑learning models attaining 93.3 % accuracy for three‑word pronunciation and 93.5 % accuracy for 26‑letter fingerprint identification.

Abstract

Abstract With the development of artificial intelligence, it is urgent to empower traditional electronics with the ability to “think,” to “analyze,” and to “advise.” Here, a new product concept namely triboelectric nanogenerator (TENG) based smart electronics via the automatic machine learning data analysis algorithm is proposed. In this work, a simple water processing technique is used to fabricate porous polydimethylsiloxane, together with the weaving copper mesh, forming a high sensitivity flexible TENG. The as‐prepared TENG presents high sensitivity for the voice signal and handwriting signal detection with ≈0.2 V amplitude in the common talking and writing condition. Three words' pronunciation are recorded and the ensemble method is used as the machine learning model for the voice signal recognition with a recognition accuracy of 93.3%. To further demonstrate the possibility of applying machine learning algorithm for automatic analysis and recognition, larger database is analyzed. Twenty‐six letters' handwriting signals with total 520 samples are collected and a letter fingerprint library is established for further analysis. Hierarchical clustering and similarity matrix are used to study the intrinsic relationship between letters. “Medium Gaussian support vector machine” is used as machine learning model for the 26‐letter fingerprint identification with recognition accuracy of 93.5%.