An advanced recurrent Artificial Neural Network (ANN) was developed and used to model and analyze the frontal impact of two different vehicles. The main factors affecting the vehicle response — in case of frontal impacts — are the impact velocity and the structural characteristics of the vehicle. Each vehicle was represented by a corresponding ANN, and the mutual interaction between the two vehicles during the impact was considered in the ANN input. An adaptive learning technique was used to correct both the connections’ weights and the steepness factor of the neurons’ continuous activation functions to reduce the computation time during the supervised learning phase.
Nonlinear finite element (FE) simulations were performed for two different vehicle models in frontal impact scenarios using different impact velocities. The results obtained by the FE simulations were used in training and testing the proposed ANN model. The passenger compartment acceleration of the vehicle during the first 100 ms of the impact event was the main criteria considered in the crash analysis. The performance of the proposed ANN configuration was very good in both training and testing phases. The proposed technique explained the ability of the adapted ANNs to store the nonlinear structural characteristics of the two vehicles after a successful training phase. These two networks were then used together to predict a new different impact scenario. The predicted velocity and acceleration curves for the new impact scenario correlated well with the corresponding curves obtained from the FE simulation.
The current technique was applied only for frontal impact scenario, however, further research will be considered to extend the methodology to be applicable for both offset and oblique impact scenarios.