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

The curing of composite materials is a complex process involving dynamic chemical reactions, and the evolution of their viscoelastic properties can be accurately characterized by the storage modulus and loss modulus. However, existing monitoring methods predominantly rely on offline detection, making it difficult to identify the curing state in time for early termination. This leads to insufficient real-time performance and limited detection efficiency. In this study, an innovative monitoring method based on ultrasonic guided waves and deep learning-based multi-task learning is proposed for accurate monitoring of viscoelastic properties during the curing of composite materials. During the experiment, Piezoelectric Lead-Zirconate-Titanate (PZT) sensors are inserted into the composite material to acquire ultrasonic guided wave signals throughout the curing process. The random forest algorithm is employed to extract features from the ultrasonic guided waves and construct an optimal feature subset. Subsequently, a multi-task learning model integrating a temporal convolutional network (TCN), a Bidirectional Gated Recurrent Unit (BiGRU), and an attention mechanism is developed to simultaneously predict the evolution of the storage modulus and loss modulus of composite materials. The deep learning-based multi-task learning method enables effective learning by exploiting the intrinsic correlations among the viscoelastic parameters. The effectiveness of the proposed method is validated on carbon fiber reinforced plastic specimens with unidirectional and quasi-anisotropic stacking. The experimental results demonstrate that the proposed method has higher prediction accuracy and stronger robustness compared to traditional methods.