Assessing Feed-Forward Backpropagation Artificial Neural Networks for Strain-Rate-Sensitive Mechanical Modeling

Area of Science:

• Computational mechanics
• Materials science
• Artificial intelligence

Background:

• Designing metal and alloy products requires accurate constitutive models for computational mechanics.
• Testing material properties under extreme conditions like explosive impacts is often challenging and constrained.
• Accurate material models are crucial for optimizing manufacturing processes and product design.

Purpose of the Study:

• To assess the efficiency and accuracy of various artificial neural network (ANN) architectures for identifying Johnson-Cook material model parameters.
• To develop an ANN-based strategy for material parameter identification, especially when experimental data is limited.
• To determine the optimal ANN configuration for predicting material behavior under diverse strain rates.

Main Methods:

• Investigated four different ANN architectures for parameter identification.
• Implemented an ANN-based computational tool for material parameter estimation.
• Evaluated ANN performance based on predictive capability using effective flow stress-strain data.
• Tested the approach on Ti64 alloy and three virtual materials.

Main Results:

• The study identified a specific perceptron-based ANN architecture (66 inputs, 30 hidden neurons) as highly accurate.
• The ANN-based strategy provided adequate results across a range of strain rates relevant to manufacturing.
• The chosen ANN configuration demonstrated high predictive accuracy for material behavior.
• The method is effective even with a reduced amount of experimental data.

Conclusions:

• Artificial neural networks offer an efficient and accurate method for identifying material model parameters, particularly the Johnson-Cook model.
• The developed ANN-based approach is suitable for general manufacturing and product design applications.
• The findings are especially valuable for scenarios with constrained high-impact material testing, enabling robust material characterization.