Unsupervised Deep Learning for Advanced Forming Limit Analysis in Sheet Metal: A Tensile Test-Based Approach

Area of Science:

• Materials Science and Engineering
• Manufacturing Processes
• Computational Mechanics

Background:

• Accurate sheet metal formability assessment is crucial for optimizing manufacturing processes.
• Traditional Nakajima tests for forming limit curves (FLC) have limitations due to friction and complex strain paths.
• Existing methods may underestimate the formability of lightweight materials and high-strength steels.

Purpose of the Study:

• To develop and evaluate a machine learning approach for determining the forming limit curve (FLC) using uniaxial tensile tests.
• To adapt a convolutional neural network (CNN) for FLC prediction in uniaxial tensile tests.
• To assess the accuracy and transferability of a stereo camera-based FLC determination method.

Main Methods:

• Utilized a weakly supervised convolutional neural network (CNN) originally designed for Nakajima tests, adapted for uniaxial tensile tests.
• Developed a stereo camera-based system for real-time strain measurement and analysis.
• Trained and validated the models using materials like AA6016, DX54D, and DP800, employing cross-validation and iterative data composition.

Main Results:

• The CNN stereo camera-based approach successfully predicted major strains for various materials and thicknesses, showing close agreement with ISO standards.
• Specific predictions for DX54D (0.8mm and 2.0mm) and AA6016 (1.0mm) closely matched ISO values.
• A divergence was noted for DP800 (1.0mm), but overall, the method demonstrated quantitative alignment with the cross-section method.

Conclusions:

• Machine learning, particularly CNNs combined with stereo camera measurements, offers a viable alternative for accurate FLC determination in uniaxial tensile tests.
• This approach overcomes limitations of traditional methods and shows potential for underestimating formability in certain materials.
• The developed method provides a quantitative and reliable means for assessing the formability of lightweight and high-strength steels.