Inverse Neural Network Approach for Optimizing Chemical Composition in Shielded Metal Arc Weld Metals

Area of Science:

• Materials Science
• Computational Materials Science
• Welding Engineering

Background:

• Optimizing chemical compositions for specific mechanical properties in shielded metal arc weld metals is crucial for material performance.
• Existing methods may not efficiently explore the complex relationship between composition and properties.
• A data-driven approach can accelerate the discovery of novel weld metal alloys.

Purpose of the Study:

• To develop a hybrid machine learning framework for optimizing shielded metal arc weld metal compositions.
• To achieve targeted mechanical properties, including yield strength, ultimate tensile strength, and Charpy impact transition temperatures.
• To enable inverse design for accelerated welding alloy development.

Main Methods:

• A neural network model was trained on an experimental database of over 950 shielded metal arc weld metals.
• Bayesian optimization was used to enhance the predictive accuracy of the neural network model.
• A genetic algorithm was integrated with the neural network for inverse design, exploring composition space to meet property targets.

Main Results:

• The neural network model demonstrated high predictive accuracy for key mechanical properties.
• The hybrid framework successfully identified multiple feasible chemical compositions matching desired mechanical property targets.
• The study validates the efficacy of neural network-assisted inverse design in welding alloy development.

Conclusions:

• The hybrid machine learning framework offers a powerful tool for optimizing weld metal compositions.
• This approach accelerates the discovery and design of advanced welding alloys with tailored properties.
• The findings highlight the potential of integrating machine learning with materials science for innovation.