Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Study on Unsupervised Anomaly Detection and Defect Localization using Generative Model in Ultrasonic Non-Destructive Testing

Yusaku Ando, Miya Nakajima, Takahiro Saitoh, Tsuyoshi Kato

TL;DR

The paper addresses the challenge of automated defect detection in Laser Ultrasonic Visualization Testing (LUVT) when defective data are scarce. It introduces AnoDDPM, a denoising diffusion probabilistic model trained solely on defect-free data, to reconstruct normal LUVT patterns and use the reconstruction difference $|x - \hat{x}|$ as an anomaly map for defect detection and localization. Across comparisons with VAE and f-AnoGAN and against supervised object detectors, AnoDDPM achieves the best performance (AUROC up to $0.909$) and more precise defect localization, demonstrating the strength of unsupervised diffusion-based anomaly detection in NDT. The approach reduces dependence on defect-labeled data and shows promise for scalable inspections in aluminum and potentially extendable to composites like CFRP and concrete.

Abstract

In recent years, the deterioration of artificial materials used in structures has become a serious social issue, increasing the importance of inspections. Non-destructive testing is gaining increased demand due to its capability to inspect for defects and deterioration in structures while preserving their functionality. Among these, Laser Ultrasonic Visualization Testing (LUVT) stands out because it allows the visualization of ultrasonic propagation. This makes it visually straightforward to detect defects, thereby enhancing inspection efficiency. With the increasing number of the deterioration structures, challenges such as a shortage of inspectors and increased workload in non-destructive testing have become more apparent. Efforts to address these challenges include exploring automated inspection using machine learning. However, the lack of anomalous data with defects poses a barrier to improving the accuracy of automated inspection through machine learning. Therefore, in this study, we propose a method for automated LUVT inspection using an anomaly detection approach with a diffusion model that can be trained solely on negative examples (defect-free data). We experimentally confirmed that our proposed method improves defect detection and localization compared to general object detection algorithms used previously.

A Study on Unsupervised Anomaly Detection and Defect Localization using Generative Model in Ultrasonic Non-Destructive Testing

TL;DR

The paper addresses the challenge of automated defect detection in Laser Ultrasonic Visualization Testing (LUVT) when defective data are scarce. It introduces AnoDDPM, a denoising diffusion probabilistic model trained solely on defect-free data, to reconstruct normal LUVT patterns and use the reconstruction difference as an anomaly map for defect detection and localization. Across comparisons with VAE and f-AnoGAN and against supervised object detectors, AnoDDPM achieves the best performance (AUROC up to ) and more precise defect localization, demonstrating the strength of unsupervised diffusion-based anomaly detection in NDT. The approach reduces dependence on defect-labeled data and shows promise for scalable inspections in aluminum and potentially extendable to composites like CFRP and concrete.

Abstract

In recent years, the deterioration of artificial materials used in structures has become a serious social issue, increasing the importance of inspections. Non-destructive testing is gaining increased demand due to its capability to inspect for defects and deterioration in structures while preserving their functionality. Among these, Laser Ultrasonic Visualization Testing (LUVT) stands out because it allows the visualization of ultrasonic propagation. This makes it visually straightforward to detect defects, thereby enhancing inspection efficiency. With the increasing number of the deterioration structures, challenges such as a shortage of inspectors and increased workload in non-destructive testing have become more apparent. Efforts to address these challenges include exploring automated inspection using machine learning. However, the lack of anomalous data with defects poses a barrier to improving the accuracy of automated inspection through machine learning. Therefore, in this study, we propose a method for automated LUVT inspection using an anomaly detection approach with a diffusion model that can be trained solely on negative examples (defect-free data). We experimentally confirmed that our proposed method improves defect detection and localization compared to general object detection algorithms used previously.
Paper Structure (26 sections, 8 equations, 8 figures, 4 tables)

This paper contains 26 sections, 8 equations, 8 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. History of Machine Learning Adoption in Ultrasonic Non-Destructive Testing
  4. Anomaly Detection
  5. LUVT Dataset
  6. What is LUVT?
  7. Creating Dataset
  8. Proposal Method
  9. Generative Models
  10. VAE (Variational Autoencoder)
  11. GAN (Generative Adversarial Networks)
  12. Diffusion Model
  13. Defect Localization Method
  14. Exploration of Generative Models and Performance Evaluation for Defect Detection
  15. Overview of the Comparative Experiments on Generative Models
  16. ...and 11 more sections

Figures (8)

  • Figure 1: LUVT device.
  • Figure 2: Examples of LUVT images. (a) is an example of an image without defect, while (b) is an example of an image with a defect at the bottom. In the image without defects, it is evident that waves propagate from the top of the image. In the image with a defect, anormalities in the waves occur because the incident wave encounters the defect.
  • Figure 3: Appearance of ultrasonic before and after reaching a defect.
  • Figure 4: Flow of the proposed method.
  • Figure 5: ROC curves for generative models.
  • ...and 3 more figures