Machine Learning-Based Force and Activity Recognition in Total Knee Replacement Using Hybrid...

Smart orthopedic implants require sensing systems that can monitor joint loading without increasing dependence on batteries or complex implanted electronics. Here, we report an implant-integrated hybrid triboelectric-piezoelectric nanogenerator (TPNG) built from six multilayer Cu/PVDF transducers distributed across the medial and lateral compartments of a tibial-tray-inspired total knee replacement (TKR) package. Coupling triboelectric contact-separation to the piezoelectric response of the stack increases the peak-to-peak voltage, while the transducers retain 92.4% of their peak-to-peak output after approximately 350,000 gait cycles. The present study evaluates the sensors under predominantly centered axial loading, which consists of 90% of loading going through the knee joint. Using a single full-wave rectifier across all six transducers charges a 0.22 μF capacitor to 68.1 V, corresponding to 509.9 μJ of stored energy, and drives a 168-LED array, demonstrating practical energy delivery from physiological joint loading. In addition, the machine learning algorithms are integrated for force and activity recognition. Using five independent 120-cycle recordings per activity with recording-level data separation, the CNN-BiLSTM classifier achieved 99.7% accuracy on the held-out recording across eight activities. A separate CNN-BiLSTM model further provided proof-of-concept reconstruction of an unseen patient's axial force profile from PVDF voltage alone. Together, these results establish a multifunctional PVDF-based sensing architecture that combines energy harvesting, fatigue-resistant transduction, and learned biomechanical interpretation from a single device, providing a pathway toward battery-free smart joint implants.