Real-time prediction of viscoelastic properties in composite curing using guided waves and multi-task deep learning

Area of Science:

• Materials Science
• Mechanical Engineering
• Signal Processing

Background:

• Composite material curing involves complex chemical reactions and evolving viscoelastic properties.
• Current offline monitoring methods lack real-time performance and efficiency for timely curing state identification.
• Accurate real-time monitoring is crucial for optimizing composite manufacturing and preventing defects.

Purpose of the Study:

• To develop an innovative real-time monitoring method for viscoelastic properties during composite curing.
• To utilize ultrasonic guided waves and deep learning for enhanced detection efficiency.
• To enable accurate prediction of storage and loss moduli for process optimization.

Main Methods:

• Insertion of Piezoelectric Lead-Zirconate-Titanate (PZT) sensors to acquire ultrasonic guided wave signals.
• Application of the random forest algorithm for optimal feature extraction from ultrasonic signals.
• Development of a multi-task learning model (TCN-BiGRU-Attention) for simultaneous prediction of storage and loss moduli.

Main Results:

• The proposed deep learning multi-task learning method effectively predicts the evolution of viscoelastic properties.
• Validation on carbon fiber reinforced plastic specimens demonstrates higher prediction accuracy and robustness.
• The method successfully exploits intrinsic correlations among viscoelastic parameters for improved monitoring.

Conclusions:

• The ultrasonic guided wave and deep learning-based multi-task learning method offers a robust and accurate solution for real-time composite curing monitoring.
• This approach overcomes the limitations of traditional offline methods, enhancing detection efficiency.
• The findings contribute to advanced manufacturing processes for composite materials.