12th World Congress on Industrial Process Tomography
Guided Wave Based Surface Temperature Mapping for Sliding Bearing using Two-Stream Convolutional Neural Network
Xingyu Fan1*, Kexin Huang1, Runlin Chen1, Rui Li1, Bo Sun1, Tong Zhao1
1School of mechanical and precision instrument engineering, Xi’an University of Technology, Xi’an, China
*Email: 105266@xaut.edu.cn
ABSTRACT
As a core component of mechanical transmission systems, the temperature of sliding bearing liners directly influences lubrication characteristics, wear behaviour, and operational safety. Despite the widespread use of noncontact temperature measurement technologies, extracting temperature-sensitive features from guided wave signals under complex boundary conditions and multi-modal interference remains challenging. Traditional methods based on time-of-flight (TOF) often suffer from low accuracy in temperature estimation. To overcome this limitation, we propose a surface temperature mapping approach that leverages ultrasonic guided wave signals and a Two-Stream Convolutional Neural Network (Two-Stream CNN). An array of ultrasonic guided wave sensors is deployed on the sliding bearing liner surface to acquire raw waveform signals and construct corresponding envelope features, which are fed into a dual-stream architecture for parallel high/low-frequency processing: the high-frequency stream captures transient guided wave fluctuations, while the low-frequency stream extracts long-term envelope attenuation characteristics. Unlike conventional parallel CNNs using only raw signals, our approach jointly fuses waveform features, envelope attenuation, and their spatial distributions, thereby significantly enhancing network sensitivity to subtle temperature changes. Validation experiments show that the Two-Stream CNN accurately reconstructs the liner surface temperature mapping and localizes heat sources via deep feature fusion, demonstrating superior performance in confined-space online temperature monitoring.
Keywords: Ultrasonic guided waves, Sliding bearing, Two-Stream Convolutional Neural Network, Temperature Mapping.
Industrial Application: Ultrasonic.
Sign-in to access the full text
Copyright © International Society for Industrial Process Tomography, 2025. All rights reserved.