Design, Development, and Testing of Machine Learning Models to Estimate Properties of Friction Stir Welded Joints

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Friction stir welding (FSW) is a critical joining process for various materials.
• Accurate prediction of mechanical properties like ultimate tensile strength (UTS) and hardness is essential for quality control.
• Supervised machine learning offers a promising approach for modeling complex material behavior.

Purpose of the Study:

• To estimate the ultimate tensile strength (UTS) and hardness of friction stir welded joints.
• To compare the performance of six different supervised machine learning models for this estimation task.
• To identify the most accurate model for predicting mechanical properties in FSW.

Main Methods:

• Utilized six supervised machine learning models: linear regression, support vector regression, decision tree regression, random forest regression, K-nearest neighbour, and artificial neural network (ANN).
• Trained and tested models using a dataset of 200 entries with input parameters including tool traverse speed, tool rotational speed, pin diameter, shoulder diameter, tool offset, and tool tilt.
• Employed ANN due to its superior performance in preliminary assessments.

Main Results:

• Artificial neural networks (ANN) demonstrated the highest accuracy among the evaluated models.
• The ANN model successfully estimated the UTS and hardness of friction stir welded joints.
• Extracted key relationships between input parameters (pin diameter, tool offset, tool rotation speed) and output properties (UTS, hardness).

Conclusions:

• The artificial neural network (ANN) approach is highly effective for accurately predicting the UTS and hardness of friction stir welded joints.
• ANN model predictions show strong consistency with experimental results, validating its generalization capability.
• The study successfully established a data-driven method for optimizing FSW process parameters to achieve desired mechanical properties.