Abstract:Rayleigh waves are successfully employed for the characterization of surface defects through the utilization of both time and frequency domain analysis. However, a significant challenge arises when attempting to cover a broad range of the defect depth to Rayleigh wavelength ratio with a single method. This paper presents an enhanced methodology that integrates optimized machine learning (ML) algorithms to expand the coverage of defect depth assessment. An experimentally validated numerical model is firstly established to simulate the interaction of Rayleigh waves with surface defects and utilized to generate a significant quantity of labeled Rayleigh wave signals. The features of received Rayleigh wave patterns are effectively extracted and serve as indicators of defect parameters. The ML framework encompasses the extraction and selection of optimal features, the construction of model architecture, and the tuning of hyperparameters. Three dimensionality optimization techniques—principal component analysis, correlation coefficient analysis, and expert knowledge integration—are compared. Additionally, six machine learning models (decision tree, support vector machine, k-nearest neighbors, random forest, gradient-boosting machine, and artificial neural network) are optimized using particle swarm optimization and subjected to comparison. The results reveal that: (1) principal component analysis yields the highest performance, suggesting the potential for feature dimensionality reduction without relying on expert knowledge; (2) among the models examined, the random forest model and artificial neural network with hyperparameter tuning demonstrate superior performance in predicting defect depth; and (3) the optimized machine learning methods consistently outperform empirical approaches in accurately predicting defect depths across a wide range. Furthermore, the proposed method shows promise for extension to determine additional parameters associated with surface defects, including their width and angle of inclination.

Deep learning based ultrasonic reconstruction of rough surface morphology

Ultrasonic Array Tomography-Oriented Subsurface Crack Recognition and Cross-Section Image Reconstruction of Reinforced Concrete Structure Using Deep Neural Networks

Wavefront Reconstruction Based on Deep Transfer Learning for Microscopy.

Deep learning for full-field ultrasonic characterization

Ultrasonic adaptive plane wave high-resolution imaging based on convolutional neural network

Deep learning for high-resolution seismic imaging

Machine learning enhanced characterization of surface defects using ultrasonic Rayleigh waves

Laser ultrasonic inspection of surface and subsurface defects in materials based on deep learning

Direct Reconstruction of Ultrasound Elastography Using an End-to-End Deep Neural Network.

DeepUCT: Complex cascaded deep learning network for improved ultrasound tomography

Deep learning for light scattering computation: Reconstructing light scattering fields from 1-D randomly rough surfaces as an example

A multiscale residual U-net architecture for super-resolution ultrasonic phased array imaging from full matrix capture data

Curvature and Distance Amplitude Correction for Ultrasonic Testing of a Component with Curved Surface Using Deep Learning Method

Toward online layer-wise surface morphology measurement in additive manufacturing using a deep learning-based approach

Ultrasonic Rough Crack Characterization Using Time-of-Flight Diffraction With Self-Attention Neural Network

Surface profile recovery from electromagnetic field with physics--informed neural networks

Characterizing Near-surface Velocity Structures Via Deep Learning

Training Deep Networks with Only Synthetic Data: Deep-learning-based Near-Offset Reconstruction for (Closed-Loop) Surface-Related Multiple Estimation on Shallow-Water Field Data

Deep learning for enhancing wavefield image quality in fast non-contact inspections

Reconstructing undersampled photoacoustic microscopy images using deep learning

Neural-IMLS: Self-supervised Implicit Moving Least-Squares Network for Surface Reconstruction