Uncertain dynamics characteristic forecasting in composite plates with multi-defects of electric aircraft via physics-augmented meta-learning

•A physics-enhanced Meta-learning Framework is proposed to predict dynamic characteristics of composite plate with defects.•Attention meta-learning mechanism is used for cross task meta knowledge weighted extraction.•The effectiveness of the PMF has been verified through comparison with physics-data...

Full description

Saved in:

Bibliographic Details
Published in:	Aerospace science and technology Vol. 164; p. 110363
Main Authors:	Xu, Duo, Zang, Jian, Song, Xu-Yuan, Zhang, Zhen, Zhang, Ye-Wei, Chen, Li-Qun
Format:	Journal Article
Language:	English
Published:	Elsevier Masson SAS 01.09.2025
Subjects:	Composite plates with defects Deep Learning Dynamics Analysis Few-shot Physics-enhanced meta-learning framework (PMF) Composite plates with defects Physics-enhanced meta-learning framework (PMF) Deep Learning Few-shot Dynamics Analysis
ISSN:	1270-9638
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	•A physics-enhanced Meta-learning Framework is proposed to predict dynamic characteristics of composite plate with defects.•Attention meta-learning mechanism is used for cross task meta knowledge weighted extraction.•The effectiveness of the PMF has been verified through comparison with physics-data driven models.•Physics mechanisms transfers prior solutions as new input features to the training of model.•Extremum stiffness equations are embedded as physical constraints into loss function of the model. The dynamics characteristic forecasting of composite plates with defects faces fundamental barriers stemming from multiscale material heterogeneity, stochastic manufacturing defects, and the gap between idealized theoretical models and real-world engineering scenarios. To address these limitations, this study introduces a physics-enhanced meta-learning framework (PMF) that synergizes cross-task meta-knowledge with attention-weighted transfer and stiffness-bounded physical constraints. A similarity-driven attention mechanism enables adaptive transfer of cross-task meta-knowledge to defective plate structures, derived through meta-learning on theoretical-experimental hybrid datasets. Furthermore, the extremal stiffness equations (max/zero stiffness to intact/perforated plate) are incorporated into the framework as physical boundary constraints. This integration allows adaptive prediction across the entire stiffness spectrum, thereby generalizing dynamics modeling for arbitrary defect-induced stiffness distributions. Experimental validation demonstrates that this approach achieves higher-fidelity accuracy in predicting nonlinear dynamics responses and defect-driven modal under few-shot conditions compared to physics-data driven models and purely data-driven models, while reducing computational costs. The PMF addresses the issues of inaccurate prediction and high costs of traditional models under complex conditions such as experimental uncertainties, limited datasets, and significant sample variations, especially in solving defect-induced modal transitions and nonlinear dynamics responses. The PMF bridges the gap between idealized models and real-world engineering conditions, providing a new intelligent modelling tool with both physical interpretability and data-driven adaptability for engineering fields such as the structural health monitoring of aviation composite materials.
ISSN:	1270-9638
DOI:	10.1016/j.ast.2025.110363