Abstract
Fiber orientation plays an important role in the mechanical behavior of short fiber-reinforced composites, yet its internal distribution is difficult to characterize nondestructively. This study proposes a deep learning-based framework to reconstruct fiber orientation fields from boundary ultrasonic wave responses. A two-dimensional anisotropic wave propagation model with free-boundary conditions was developed to simulate wave signals in composite plates, where the local anisotropy was governed by sectional fiber orientation. Based on this model, a synthetic dataset was generated using a sectional representation of the plate. The simulated sensor signals were used as inputs to a multilayer perceptron, while the section-wise principal-direction angles were encoded using an angle aware representation to account for periodicity. The trained model directly predicted the orientation field from the simulated boundary wave responses. Reconstruction accuracy was evaluated using wrap-aware angular error metrics. Parametric studies were conducted to examine the effects of dataset size, excitation frequency, anisotropy strength, sensor ablation, and wave transmission factor on prediction performance. The results show that the proposed framework can effectively reconstruct the main features of the section-wise principal-direction field and remains reasonably robust under varying sensing and material conditions. The proposed approach provides a promising data-driven strategy for nondestructive characterization of internal fiber orientation in short fiber-reinforced composites.
| Original language | English |
|---|---|
| Article number | 114936 |
| Journal | Mechanical Systems and Signal Processing |
| Volume | 260 |
| DOIs | |
| State | Published - Oct 1 2026 |
Keywords
- Deeplearning
- Elastic waves
- Inverse modeling
- Short fiber-reinforced composites
Fingerprint
Dive into the research topics of 'Deep learning based reconstruction of fiber orientation in short fiber reinforced composites from ultrasonic wave responses'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver