Leveraging machine learning for predicting human body model response in restraint design simulations.

Area of Science:

• Computational mechanics
• Machine learning
• Biomechanical engineering

Background:

• Predicting human body model response is crucial for effective restraint system design.
• Parametric simulations are computationally intensive, necessitating efficient predictive models.

Purpose of the Study:

• To evaluate and compare various machine learning techniques for predicting human body model responses in restraint design simulations.
• To identify the most effective machine learning approach for developing accurate response surface models.

Main Methods:

• Employed Ordinary Least-Squares (OLS), LASSO, Neural Network (NN), Support Vector Regression (SVR), Regression Forest (RF), and an ensemble method.
• Optimized machine learning hyperparameters using grid search and cross-validation to prevent under-fitting and over-fitting.
• Developed response surface models based on 16 independent variables from parametric simulations.

Main Results:

• The ensemble method demonstrated superior performance in predicting simulation responses compared to individual techniques.
• The performance ranking of the methods was: Ensemble > LASSO > SVR > NN > RF > OLS.
• Optimizing metamodel hyperparameters was critical for accurately predicting optimal restraint design parameters.

Conclusions:

• Machine learning, particularly ensemble methods, can effectively predict human body model responses in restraint design.
• Proper hyperparameter tuning is essential for building accurate predictive models and optimizing restraint designs.
• This approach can enhance the efficiency and accuracy of restraint system development simulations.