Multimodal human-in-the-loop artificial intelligence with affective feedback for accelerated high-entropy alloy...

Area of Science:

• Materials Science
• Artificial Intelligence
• Computational Materials Design

Background:

• High-entropy alloys (HEAs) offer advanced properties but are challenging to design due to complexity.
• Conventional design methods face limitations in computational cost and predictive accuracy.
• Existing machine learning (ML) approaches for HEA design suffer from data scarcity and poor interpretability.

Purpose of the Study:

• To develop an integrated human-computer interactive platform for designing high-entropy alloys.
• To incorporate emotional feedback for dynamic adjustment of optimization parameters.
• To improve the efficiency and accuracy of AI-driven material design through human-AI collaboration.

Main Methods:

• Integration of Natural Language Processing (T5 model), Machine Learning (XGBoost), and Multi-Objective Optimization (NSGA-II).
• Implementation of a closed-loop "perception-decision-optimization" workflow with real-time emotion recognition.
• Utilizing SHAP analysis for interpretability and multiscale modeling for validation.

Main Results:

• High accuracy achieved in predicting yield strength and Young's modulus for HEAs.
• Emotion-driven optimization demonstrated effective Pareto front convergence with <1.4% deviation from experimental values.
• SHAP analysis provided insights into the physical mechanisms governing alloy properties.

Conclusions:

• The integrated platform offers a novel paradigm for interpretable and efficient AI-driven material design.
• Incorporating affective feedback enhances human-AI collaboration and aligns design outcomes with expert preferences.
• The approach demonstrates significant potential for accelerating the discovery and development of advanced materials like HEAs.