Loading...

In Vivo-Based Artificial Neural Networks to Predict 3d Human Body Posture and Lumbosacral Joint Moment During Lifting Activities

Aghazadeh Shabestari, Farzad | 2019

437 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 51808 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Arjmand, Navid
  7. Abstract:
  8. Manual material handling (MMH) is associated with mechanical back injuries. To manage such injuries, musculoskeletal models are employed to estimate spinal loads during MMH. These models require, as input, 3D-position of the hand-load and body posture whose measurements involve skillful and time-consuming motion analysis investigations.In order to facilitate the procedure of posture measurements and load estimations, three coupled artificial-neural-networks (ANNs) were developed. To predict whole body posture, ANN1 was trained based on our novel measurements on 15 individuals. Each individual performed several static-tasks by holding 0, 5, and 10 kg weight at 9 different anterior-left positions and at 5 heights (0, 30, 60, 90, and 120 cm/floor). Posture was measured via Simi Reality Motion Systems that recorded 3D-position of 15 skin-markers on the C7, T12, L5, pelvis, left/right shoulders, elbows, wrists, knees and ankles. The ANN1 predicted posture by identifying the relationship between 5 inputs (hand-load magnitude, its 3D-position and body height) and 45 outputs (3D-position of markers); total of 69885 input/output datasets: 1553 tests and 45 marker-positions. The ANN2 predicted posture by segments euler angles through 5 inputs identical to the former ANN and 42 outputs (euler angles of 14 links); total of 65226 input/output datasets: 1553 tests and 42 euler angles. Moreover, for each subject/task, the 3D L5-S1 external moments were evaluated using the measured posture, hand-load position/magnitude, and anthropometric data. The ANN3 identified the relationship between the foregoing inputs (plus body weight) and L5-S1 moments; total of 3106 datasets: 1553 tests 2 moments. Finally, predicted posture and hand-load position/magnitude were input into a previously-developed/validated ANN4 that predicted lumbar disc loads and trunk muscle forces.Trained ANNs could predict 3D body posture with positions (R= 0.987/RMSE= 7 cm) , with euler angles (R= 0.92/RMSE= 29.9 deg) and L5-S1 moments (R= 0.987/RMSE= 13.8 Nm) from novel sets of inputs that were not included in the training processes.For any given MMH task, the ANNs were capable of predicting body posture, L5-S1 moments, and lumbar loads via easy-to-measure inputs
  9. Keywords:
  10. Artificial Neural Network ; Simi Motion Tracking ; Marker ; Lumbosacral (L5-S1) Moment ; Posture Prediction

 Digital Object List

 Bookmark

No TOC