WaveGraphNet: Physics-Consistent Guided-Wave Damage Localization through Coupled Inverse-Forward Graph Learning

2026-05-22 · Source: cs.LG updates on arXiv.org · Field: Science & Research — Engineering & Applied Sciences, Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

WaveGraphNet is a novel coupled inverse–forward graph learning framework designed for guided-wave damage localization in Carbon Fiber Reinforced Polymer (CFRP) plates. It addresses the challenge of poor generalization to unseen structural regions when training data is limited. The system models the sparse transducer network (e.g., 12 transducers on a 500x500 mm² plate) as a graph, using an inverse branch to predict defect coordinates from spectral descriptors of differential guided-wave responses. A forward branch acts as a physics-consistent regularizer during training, predicting path-wise energy-deviation patterns. Evaluated on the OGW-1 SHM Plate benchmark with spatial hold-out splits, WaveGraphNet achieved the lowest unseen-zone Mean Absolute Error (MAE) (e.g., 0.220±0.027 in Split A, 0.262±0.016 in Split B) and a 0.0% false-positive rate, demonstrating improved robustness in extrapolation compared to non-graph and graph baselines.

Key takeaway

For Machine Learning Engineers developing SHM systems for composite structures, this research highlights that integrating physics-guided forward consistency into graph neural networks significantly improves damage localization in previously unseen regions. You should consider adopting coupled inverse-forward graph learning frameworks to enhance model robustness and reduce false positives (0.0% FPR) when spatial training data is inherently sparse. This approach is crucial for bridging the gap between laboratory performance and reliable field deployment.

Key insights

Coupling inverse graph localization with a physics-consistent forward model improves damage localization extrapolation in unseen regions.

Principles

• Explicitly modeling sensing layouts as graphs improves generalization.
• Forward consistency regularization reduces ambiguity in inverse problems.
• Spatial hold-out evaluation reveals true generalization capability.

Method

WaveGraphNet uses a three-stage training: inverse-branch pretraining, forward-branch pretraining, then joint optimization with localization, forward-consistency, and physics-correction losses. Inference uses only the inverse branch.

In practice

• Represent transducer networks as graphs for SHM tasks.
• Incorporate physics-guided regularization for sparse inverse problems.
• Use spatial hold-out splits for robust model evaluation.

Topics

• Structural Health Monitoring
• Guided Waves
• Graph Neural Networks
• Damage Localization
• Inverse Problems
• CFRP Plates
• Physics-informed AI

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• A Multi-Fidelity Convolutional Autoencoder-Transfer Learning Framework for Guided-Wave-Based Damage Diagnosis Using Large Simulated and Limited Experimental Datasets — A multi-fidelity CAE-transfer learning framework enables accurate GWSHM damage diagnosis using large simulated and limited experimental data.
• Mesh Graph Neural Network Framework for Accelerating Finite Element Simulation for Arbitrary Geometries — Mesh Graph Networks effectively generalize finite element analysis predictions across arbitrary structural geometries by encoding relative features.
• Extended pseudo-spectral physics-informed neural networks for phase-field models — ESPINN accurately infers phase-field model parameters and free-energy structure from limited transient data, even with noise.
• Physics-Guided Spatiotemporal Learning for Coastal Wave Peak Period Estimation from Video — Physics-guided deep spatiotemporal learning from video can accurately and consistently estimate coastal wave peak periods, overcoming traditional monitoring limitations.
• MoSA: Motion-constrained Stress Adaptation for Mitigating Real-to-Sim Gap in Continuum Dynamics via Learning Residual Anisotropy — MoSA learns residual anisotropy and heterogeneity using physics priors to close the real-to-sim gap in continuum dynamics.
• Multi-Modal Spatio-Temporal Graph Neural Network with Mixture of Experts for Soil Organic Carbon Prediction — SpTGNN unifies multi-modal data, heterogeneous graphs, and uncertainty quantification for accurate soil organic carbon prediction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.