Neural Network Material Model Enhancement: Optimization Through Visualization and Selective Data Removal

Neural network based constitutive models have recently been used to capture soil constitutive response. The neural networks are trained with large data sets. A novel method of enriching these training data sets is presented and tested, showing a limited data set can be expanded to a wider range of problems. A method of selectively reducing data set size is also presented, so that maximum behavior can be gleaned from a minimum amount of data. By using novel visualization techniques to evaluate the learned neural network model responses, it is shown that a large reduction in data set size does not significantly affect the performance of the neural network model.