Model reduction of structural mechanical response in the time domain

Area of Science:

• Mechanical Engineering
• Computational Mechanics
• Machine Learning

Background:

• Analyzing structural mechanical response states in automotive systems is crucial for safety and performance.
• Traditional methods can be computationally intensive, necessitating efficient model reduction techniques.

Purpose of the Study:

• To introduce a novel model reduction approach for analyzing structural mechanical response states in automotive systems.
• To leverage time-domain signal data from road testing for enhanced model accuracy.

Main Methods:

• Finite element modeling and analysis to identify peak stress areas.
• Road load spectrum signal tests to extract vibration and strain data.
• Machine learning techniques, including response surface and Gaussian process models, for model reduction.
• Hyperparameter tuning and optimization for neural network models.

Main Results:

• Achieved high fidelity index of 98.6% for validation sets and 95.2% for test sets.
• Demonstrated the effectiveness of the reduced-order model in capturing structural mechanical response states.
• Identified optimal hyperparameter configurations for neural network models.

Conclusions:

• The developed reduced-order model is suitable for deployment in monitoring local structural mechanical response states and assessing damage.
• Model reduction techniques, particularly neural networks with optimized hyperparameters, offer a powerful tool for automotive structural analysis.
• Insights into the impact of neural network architecture (cells number, hidden layers) on model performance and generalization were gained.