Hybrid Data-Driven Deep Learning Framework for Material Mechanical Properties Prediction with the Focus on Dual-Phase...

Area of Science:

• Computational Material Science
• Mechanical Engineering
• Artificial Intelligence

Background:

• Understanding material mechanical behavior traditionally requires extensive and costly experiments.
• Advances in computational material science and machine learning offer potential to reduce experimental needs.
• Predicting material properties under various conditions is crucial for engineering applications.

Purpose of the Study:

• To develop a computational method for predicting the mechanical behavior of dual-phase steels.
• To create a reliable dataset of microstructures and mechanical properties using simulations.
• To leverage deep learning for accurate property prediction, minimizing experimental costs.

Main Methods:

• A data pipeline combining phase field simulations and finite element analysis was established.
• Experimentally validated simulations generated a dataset for dual-phase steels under varied heat treatments.
• A hybrid deep learning model, integrating ResNet50 and VGG16 with hyperparameter optimization, was developed.

Main Results:

• The hybrid deep learning model accurately predicted yield stress, ultimate stress, and fracture strain for dual-phase steels.
• Prediction errors for mechanical properties were consistently below 1%.
• The model demonstrated effectiveness in forecasting material behavior under novel treatment conditions.

Conclusions:

• The developed machine learning approach significantly reduces the reliance on experimental methods for material characterization.
• The hybrid deep learning model offers a highly accurate and efficient tool for predicting dual-phase steel mechanical properties.
• This work paves the way for accelerated material design and development through computational predictions.