Machine Learning-Based Surrogate Modelling for Efficient Inverse Analysis of Micro-Indentation Response to Determine...

Area of Science:

• Materials Science
• Computational Mechanics
• Data Science

Background:

• Inverse analysis in indentation testing is complex due to nonlinear material response.
• Accurate material parameter identification is crucial for understanding mechanical behavior.

Purpose of the Study:

• To develop a reliable and efficient data-driven method for inverse parameter identification in indentation experiments.
• To integrate artificial neural networks (ANNs) with evolutionary optimization for enhanced inverse analysis.

Main Methods:

• Generated a comprehensive dataset via systematic simulation of indentation processes.
• Trained ANN models to predict indentation responses (displacement-time, force, surface profile) from material parameters.
• Employed evolutionary optimization (genetic algorithm) with ANN-based surrogate models for efficient loss function minimization.

Main Results:

• Trained ANN surrogate models accurately predict indentation responses, replacing computationally intensive simulations.
• The integrated approach enables orders-of-magnitude acceleration of inverse analysis.
• Investigated various mathematical loss functions to ensure robust and unique material parameter determination.

Conclusions:

• The proposed data-driven method successfully performs inverse analysis on simulated indentation data.
• This approach offers a computationally efficient pathway for material characterization.
• Future work will extend the method to experimental indentation data for micro- and nano-indentation testing.