Machine-Learning-Enabled Flexible Piezoelectric Nanogenerator Based on PVDF/KNNBZ Nanofibers for Intelligent Human Motion Recognition

Abstract

The performance of flexible piezoelectric nanogenerators (FPNGs) based on poly(vinylidene fluoride) (PVDF) remains limited due to a poor electroactive β-phase. The incorporation of ceramic fillers as nucleating agents into the PVDF matrix is one of the most effective strategies to enhance the β-phase content. However, conventional piezoelectric ceramics offer limited polarization. In contrast, relaxor ferroelectrics are characterized by dynamic polar nanoregions (PNRs) that readily reorient under external stimuli. This nanoscale responsiveness translates into significantly higher piezoelectric coefficients, markedly amplifying the performance of PVDF based FPNGs. Rather than relying on conventional piezoelectric ceramics, this work introduces Bi3+/Zr4+ co-doped potassium sodium niobate (K0.425Na0.425Bi0.15Nb0.85Zr0.15O3, KNNBZ) as a filler, chosen specifically for its relaxor nature. Simultaneous A- and B-site co-doping destabilizes long-range ferroelectric order and promotes the growth of PNRs. The electrostatic interaction between the negatively charged surface of KNNBZ and PVDF –CH2– dipoles locks chains into the all-trans β-phase conformation, reinforced further by the mechanical stretching and in-situ poling inherent to electrospinning. The optimized device generates an open-circuit voltage of 29 V and a peak power density of 76.3 μW/cm2 and directly powers commercial LEDs. Combined with a machine learning algorithm, the FPNG distinguishes location-specific gait signals with 95.56% accuracy, demonstrating practical utility for intelligent, self-powered wearable health monitoring.