Self‐Powered Bearing Sensing and Real‐Time Fault Diagnosis Enabled by Non‐Invasive Triboelectric Sensors and Edge AI Acceleration
Source: PubMed Central Open Access, NCBI / U.S. National Library of Medicine
ABSTRACT Real‐time perception of bearing operating conditions is essential for ensuring the reliable functioning of rotating machinery, yet conventional monitoring approaches that rely on complex sensor networks and external power supplies are constrained by installation space and environmental interference, hindering the realization of highly integrated, low‐power, and real‐time industrial monitoring. To address this challenge, a non‐invasive single‐electrode triboelectric bearing sensor (NSE‐TBS) is developed, which can be directly attached to the bearing surface. Based on the principle of triboelectric nanogenerators (TENG), the sensor converts mechanical energy into self‐powered condition‐sensing signals. Experimental results demonstrate that the NSE‐TBS enables stable rotational speed tracking, cage skidding detection, and fault feature extraction under various operating conditions. Furthermore, a 1D vision Transformer (1D‐ViT) diagnostic system accelerated by a field‐programmable gate array (FPGA) is implemented. Through optimized dataflow and parallel matrix multiplication engine, the model achieves an inference power consumption of only 4.59W on the FPGA, representing reductions ofandcompared with CPU and GPU implementations, respectively, with a latency as low as 0.108ms and a fault‐classification accuracy of 98%. This study achieves the integration of self‐powered sensors with edge AI acceleration, providing a new pathway for real‐time diagnostics of smart bearings.
Abstract
ABSTRACT Real‐time perception of bearing operating conditions is essential for ensuring the reliable functioning of rotating machinery, yet conventional monitoring approaches that rely on complex sensor networks and external power supplies are constrained by installation space and environmental interference, hindering the realization of highly integrated, low‐power, and real‐time industrial monitoring. To address this challenge, a non‐invasive single‐electrode triboelectric bearing sensor (NSE‐TBS) is developed, which can be directly attached to the bearing surface. Based on the principle of triboelectric nanogenerators (TENG), the sensor converts mechanical energy into self‐powered condition‐sensing signals. Experimental results demonstrate that the NSE‐TBS enables stable rotational speed tracking, cage skidding detection, and fault feature extraction under various operating conditions. Furthermore, a 1D vision Transformer (1D‐ViT) diagnostic system accelerated by a field‐programmable gate array (FPGA) is implemented. Through optimized dataflow and parallel matrix multiplication engine, the model achieves an inference power consumption of only 4.59W on the FPGA, representing reductions ofandcompared with CPU and GPU implementations, respectively, with a latency as low as 0.108ms and a fault‐classification accuracy of 98%. This study achieves the integration of self‐powered sensors with edge AI acceleration, providing a new pathway for real‐time diagnostics of smart bearings. This study achieves the synergistic integration of self‐powered sensing and edge AI acceleration to establish a real‐time fault diagnosis system. The proposed TENG‐based self‐powered bearing sensor (NSE‐TBS) and FPGA‐accelerated edge AI framework fundamentally break through the inherent limitations of conventional monitoring systems, including complex wiring, high power consumption, and insufficient real‐time capability. advs75373-abs-0001 graphical