3D printing of self-powered piezoelectric sensors enabled gait recognition via deep learning

Abstract

Wearable gait analysis systems leveraging piezoelectric sensing and artificial intelligence hold significant potential for non-invasive monitoring of neurodegenerative disorders. However, challenges remain in current gait analysis due to the high computational costs, the algorithmic complexity, and data privacy constraints. Here, we provide a self-powered gait recognition system based on flexible polyvinylidene fluoride (PVDF) piezoelectric sensors. By combining newly developed electro-assisted three-dimensional (3D) printing, the printed PVDF sensor not only demonstrates a high pressure sensitivity of 75.7 mV kPa⁻¹, but is also endowed with exceptional cyclic stability (>6000 cycles) and robust environmental adaptability. Subsequently, by leveraging a hybrid convolutional neural network-long short-term memory (CNN-LSTM) framework for the extraction of spatial features and temporal dependencies, the gait recognition system achieves a high accuracy of 98.7% of nine physiological postures. In addition, the system also exhibits 93.7% accuracy in identifying Parkinson's gait anomalies, such as stair climbing, falling, and limping, outperforming the conventional machine learning algorithms. This integration of 3D printing and deep learning bridges the gap between wearable devices and artificial intelligence-enhanced diagnostics, offering a scalable platform for personalized disease management.