Machine learning coupled highly sensitive and robust polyvinylidene fluoride thin-film sensor for wearable motion recognition

Abstract

Emerging applications in the field of health monitoring and exoskeleton robotics have led to an urgent demand for high-performance body motion recognition. However, the available motion recognition systems face challenges due to shortcomings including high cost, complex structure, low accuracy, and poor reliability. This work demonstrates a flexible, sensitive, and robust polyvinylidene fluoride (PVDF) composite thin-film sensor with enhanced piezoelectric polarization effect for highly efficient human motion recognition. The thin-film sensor presents a high sensitivity of 27.06 KPa−1 and piezoelectric voltage of 8.7 V, together with a broad detection range of 0.01–3 MPa and low attenuation of 5.6% after 30 000 loading cycles, which are superior to those of many of the reported piezoelectric sensors and commercial SDT1-028K sensor. Employing first-principles calculations, we show that doping of Cu in aluminum zinc oxide (AZO) facilitates the transfer of piezoelectrically excited charges and enhances the electron-transferring capacity of the Cu-AZO/PVDF hybrid structure, leading to stronger polarization effect and enhanced piezoelectric properties. A machine learning coupled multi-sensor network is engaged with inputting piezoelectric signals from insoles and knees, exhibiting excellent overall classification rate of 95.54% for six human motions.