Peer-Reviewed Article
Automated traction analysis for worn shoes: combining novel imaging technology, convolutional neural networks, and hydrodynamic modelling to predict friction performance
Aristizábal Pla G, Beschorner KE; Automated traction analysis for worn shoes: combining novel imaging technology, convolutional neural networks, and hydrodynamic modelling to predict friction performance; Footwear Science, vol. 18(1), pp. 15-25, 2026.
Abstract
Slips and falls are a major cause of workplace injuries, often exacerbated by worn shoe outsoles. Tread wear leads to increased under-shoe fluid pressures and increased slip risk. Frustrated total internal reflection (FTIR) technology offers a cost-effective approach for assessing the slip risk for worn shoes. However, current FTIR-based image analysis methods are not fully automated, and existing fluid pressure models have not been applied to the complex geometries that form on actual worn shoes. This study aimed to advance the use of FTIR for assessing shoe slip risk by: (1) assessing the accuracy of a U-Net network (a convolution neural network) in automatically identifying contact regions from FTIR images, and (2) investigating the association between model-predicted fluid pressures of the identified worn regions and slip outcome, coefficient of friction (COF) and experimentally measured peak fluid pressures. Fifty-five participants wearing worn slip-resistant shoes completed walking trials, including unexpected exposure to a slippery surface. Experimental peak fluid pressures were recorded using an array of fluid pressure sensors embedded in the floor and slip outcome was determined from heel kinematics. The COF was measured with a mechanical slip testing device. The shoes from the participants were scanned with a FTIR device. The U-Net was able to predict contact regions that overlapped with the ground truth contact regions by 80%. Higher predicted peak fluid pressures were associated with higher experimentally measured peak fluid pressures (R2 = 0.22), lower COF values (R2 = 0.50), and increased slip risk (p = 0.007). In conclusion, this study demonstrates the feasibility of using FTIR technology combined with convolutional neural networks and hydrodynamic modelling to automate the evaluation of slip risk due to the shoe’s worn condition.
People
