A neural-network-based model for the dynamic simulation of the tire/suspension system while traversing road...

Area of Science:

• Automotive Engineering
• Computational Neuroscience
• Machine Learning

Background:

• Tire and suspension dynamics are critical for vehicle performance and ride comfort.
• Accurate simulation of these dynamics is essential for vehicle design and testing.
• Existing models may lack efficiency or accuracy in capturing complex behaviors.

Purpose of the Study:

• To develop and validate a simulation model for tire/suspension dynamics using recurrent neural networks (RNNs).
• To investigate the optimal architecture for RNNs balancing accuracy and computational efficiency.
• To assess the model's performance against experimental data across various operating conditions.

Main Methods:

• Utilized recurrent neural networks (RNNs), a type of artificial neural network with feedback connections.
• Performed parametric analysis to determine the optimal network architecture.
• Trained the neural network model with experimental data from laboratory simulations of road profiles (cleats).

Main Results:

• The RNN model demonstrated good agreement with experimental results under diverse operating conditions.
• The trained neural network effectively predicted the dynamic behavior of elastic bushings and tires.
• The model achieved a favorable tradeoff between accuracy and network size.

Conclusions:

• Recurrent neural networks provide an accurate and computationally efficient tool for simulating tire/suspension dynamics.
• The developed NN model can be integrated into broader vehicle system models.
• Despite being a black-box model, it is valuable for evaluating vehicle ride and NVH performance.