Fusion of Flexible Sensors and Computer Vision for Biomechanical Modeling of Simulated Wind Turbine Maintenance Operations

Abstract

Accurate biomechanical modeling is essential for assessing musculoskeletal loading during occupational tasks, but traditional laboratory motion capture is often impractical for field applications. This study developed and evaluated a sensor fusion framework combining flexible wearable sensors with single-view mobile video keypoints for biomechanical modeling of simulated wind turbine maintenance operations. Eighteen healthy participants performed three tasks: symmetric lifting, bolt-torque tensioning, and ladder climbing. Motion was captured simultaneously using 13 BioStamp flexible sensors, a marker-based motion capture system, and mobile video processed with pose detection algorithms. A convolutional neural network was trained using leave-one-subject-out cross-validation to predict whole-body joint angles referenced to OpenSim inverse kinematics. The fusion model achieved an average RMSE of 9.3°, improving estimation accuracy by 5% compared with the sensor-only model and 21% compared with the video-only model. Findings suggest that combining flexible sensors with video-derived keypoints offers a scalable approach for field-based occupational biomechanics assessment.