Loading...

Simulations of Dynamic Hand-Loading Activities by Using Musculoskeletal Modeling Based on Experimental Versus Full-Body Posture Prediction Neural Network Data

Hosseini, Nesa | 2023

55 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55956 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Arjmand, Navid
  7. Abstract:
  8. Body posture is an essential input of musculoskeletal models that evaluate spinal loads in occupational activities. Posture is either measured in vivo via video-camera motion capture systems or predicted via artificial neural networks (ANNs) [1]. As video-camera measurements are impractical for use in real workstations, we have recently developed an ANN that predicts full-body posture during one- and two-handed static load-handling activities. This ANN, trained based on the posture data of 20 subjects each performing 204 static load-handling activities, uses 3D coordinates of the hand-load, body weight, and body height of the worker to predict 3D coordinates of 41 full-body skin markers. The root-mean-square-error (RMSE) between the ANN predicted and in vivo measured postures during static load-handling activities was ~2.5 cm (averaged for all markers/tasks). The present study aims to: 1) use this ANN to predict full-body posture during twenty-five dynamic lifting tasks performed by seven individuals and 2) predict dynamic spinal loads by the AnyBody Modelling System (AMS) driven by either measured or predicted postures. To predict dynamics postures, hand-load position in dynamic tasks was input into the ANN as function of time. Results indicated that the ANN successfully predicted dynamic postures; the RMSE between the predicted and measured postures for all markers/tasks/subjects (39 markers×25 dynamics tasks×7 subjects) was equal to ~7.4 cm (R2 = 0.98). Moreover, the predicted L5-S1 compression and shear loads by the AMS driven by the predicted or measured postures were in close agreement; normalized RMSEs (averaged for all subjects/tasks) were smaller than 10% (p-value > 0.05) (RMSEL4-L5 Compression = 357.96(N), NRMSE L4-L5 Compression = 9.30%, RMSEL5-S1 Compression = 349.78(N), NRMSE L5-S1 Compression = 9.83%, RMSEL4-L5 Shear = 81.9(N), NRMSE L4-L5 Shear = 8.90%, RMSEL5-S1 Shear = 119.94(N), NRMSE L5-S1 Shear= 9.46%). These results indicate the robustness of the ANN to predict statics and dynamics lifting postures as well as their applicability in predicting spinal loads by musculoskeletal models
  9. Keywords:
  10. Artificial Neural Network ; Posture Prediction ; Anybody Software ; Musculoskeletal Modeling ; Manual Material Handeling ; Spine ; Lumbar Vertebral

 Digital Object List

 Bookmark

No TOC