Loading...

Development of Artificial Neural Network to Estimate Spinal Loads Using a Coupled Musculoskeletal Finite Element Model

Mohammadi, Hassan | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58184 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Arjmand, Navid
  7. Abstract:
  8. Low back pain is highly prevalent, with great economic costs, and one of the most common causes of chronic disability among adults. Risk factors for lumbar injuries include physical factors, social demographic characteristics, individual habits, and psychosocial factors. From a biomechanics point of view, the role of the spine is to create upper body motion and withstand external loads; thus, to investigate physical factors, the loads generated in different segments of the spine during high-risk activities must be compared with their acceptable limits. Several biomechanical models have been developed to estimate the forces generated in the components of the spine. These models are classified into equilibrium-based models and stability-based models, depending on the method used to calculate muscle forces, where equilibrium-based models calculate muscle forces using optimization or EMG signals. The coupled musculoskeletal finite element model, based on equilibrium and using optimization of the sum of cubes of muscle stresses, calculates the muscle forces of the spine and incorporates the inactive parts (discs, facet joints, and ligaments) of the lumbar spine with great detail. This coupled model, in comparison to beam joints, spherical joints, and hybrid model, has shown lower errors in estimating the in vivo data of L4-L5 intradiscal pressures. Considering four inputs: sagittal trunk flexion angle, mass of the load carried in hands, load horizontal distance to shoulder joint, and lumbopelvic ratio, 500 symmetric static lifting activities were simulated with the coupled model. To ensure the convergence of the coupled model in simulating standing posture or heavy flexed lifting activities, the learning rate for penalty of muscle forces was defined. The results from the coupled model simulation generated an input-output dataset, and using this dataset, a neural network predictor for model outputs such as intradiscal pressure, maximum annulus stress, disc compression and shear forces, facet joints force, ligament forces, and muscle forces was developed. The appropriate selection of hidden layer sizes, activation functions, and weights updating method resulted in a satisfactory performance of the MLP with three hidden layers in predicting the test data outputs (coefficient of determination 0.98
  9. Keywords:
  10. Artificial Neural Network ; Spinal Loads ; Static Lifting ; Coupled Musculoskeletal Finite Element Modeling ; Symmetric Static Lifting ; Low Back Pain

 Digital Object List

 Bookmark

...see more