The initial velocity and structural characteristics of any vehicle are the main factors affecting the vehicle response in case of frontal Impact. Finite Element (FE) simulations are essential tools for crashworthiness analysis, however, the FE models are getting bigger which increases the simulation time and cost. An advanced recurrent Artificial Neural Network (ANN) is used to store the nonlinear dynamic characteristics of the vehicle structure. Therefore, hundreds of impact scenarios can be performed quickly with much less cost by using the trained networks. The equation of motion of the dynamic system was used to define the inputs and outputs of the ANN. The back-propagation learning rule was used to adjust the connecting weights and biases of the developed Network. To include the dynamics of the system, the delayed acceleration was fed back as an input to the ANN together with the velocity and displacement.
A Finite Element (FE) model for a simple box beam with rigid mass attached to it was developed to represent a general crushable object. The simulation results were performed by impacting this model into a rigid wall with different initial velocities. The displacement, velocity and acceleration curves obtained from the simulation — for the C.G. of the moving mass — were used to train the ANN. After a successful training phase, the ANN was tested by predicting a new acceleration curve. The points of the acceleration curve were predicted sequentially since only one point of the curve is predicted through one cycle of the NN operation. The predicted acceleration curve showed a good correlation with the actual curve obtained from the simulation. During the recall phase, the predicted acceleration of a new state was integrated twice to obtain the velocity and displacement by using a second order integration scheme. Then, the displacement, velocity and acceleration of this new state were fed to the ANN to predict the next state acceleration, and so forth.
The results indicated that the recurrent ANN can accurately capture the frontal crash characteristics of any impacting structure, and predict the crash performance of the same structure for any other crash scenario within the training limits. The current paper considered only the front impact, however, an offset and oblique impact scenarios will be included in further research.