Testing technology for tensile properties of metal materials based on deep learning model

Area of Science:

• Materials Science and Engineering
• Computational Materials Science
• Artificial Intelligence in Materials Testing

Background:

• Current tensile properties testing methods for metallic materials lack a universally suitable research approach.
• Accurate characterization of metallic material properties is crucial for engineering applications.
• The application of advanced computational models in materials testing is an emerging area.

Purpose of the Study:

• To explore the application of the neural Turing machine (NTM) model for metallic material tensile properties testing.
• To improve the NTM model for faster and more explicit results in tensile property analysis.
• To evaluate the performance of enhanced NTM models (H-NTM, AH-NTM) in terms of training time and accuracy.

Main Methods:

• Initial application of the standard Neural Turing Machine (NTM) model to analyze tensile properties.
• Development and implementation of improved NTM architectures (H-NTM, AH-NTM) for enhanced performance.
• Comparative analysis of training times and accuracy across NTM, H-NTM, and AH-NTM models.
• Data filtering based on indentation intervals to reduce errors in tensile property determination.

Main Results:

• Improved NTM models (H-NTM, AH-NTM) demonstrated reduced training times compared to the standard NTM.
• Specific improvements in training time were observed for replication (6.0%), addition (8.8%), and multiplication (7.3%).
• Removing data from an indentation interval of 0.5-0.7 mm significantly reduced errors.
• An indentation interval of 0.8-1.5 mm yielded stress values closer to experimental tensile test results (yield strength 219.9 MPa, tensile strength 258.8 MPa).

Conclusions:

• The Neural Turing Machine model shows significant application value in the exploration of metal material tensile properties testing technology.
• Enhanced NTM models offer faster and more accurate analysis of material tensile properties.
• Data preprocessing, specifically handling indentation intervals, is critical for improving the reliability of NTM-based material testing.