Abstract
Aluminum alloys used in aerospace can be monitored for cracks using acoustic emission (AE) technology, ensuring structural safety. However, data-driven deep learning models for AE source planar localization of defects in aerospace aluminum alloy plates suffer from poor prediction ability for unknown data and require a large amount of data. Additionally, physical models are heavily influenced by noise. To address these challenges, this paper proposes a multimodal fusion convolutional physical information neural network (MFC-PINN) model that considers numerical-type and image features, improving the model’s comprehensive performance and generalization ability. The model also uses physical equations as constraints, reducing the dependence on large amounts of data and improving the prediction ability for unknown data. MFC-PINN performs well on datasets in complete, incomplete, and noisy conditions, especially in predicting unknown data in high-noise conditions. The study ensures the reproducibility of experiments and the validity of modules through sub-sampling cross-validation and ablation study.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call