Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Defect Segmentation in OCT scans of ceramic parts for non-destructive inspection using deep learning

Andrés Laveda-Martínez, Natalia P. García-de-la-Puente, Fernando García-Torres, Niels Møller Israelsen, Ole Bang, Dominik Brouczek, Niels Benson, Adrián Colomer, Valery Naranjo

TL;DR

The study tackles automated defect segmentation in OCT scans of ceramic parts for non-destructive testing by training a U-Net with a ResNet34 encoder on manually labeled inclusions. It systematically evaluates multiple loss configurations and employs preprocessing steps to ensure compatible input sizes, achieving a test DSC of 0.979 and high precision/recall (0.983/0.978) with an inference time of 18.98 seconds per volume. The MIR-OCT Scans dataset underpins the evaluation, demonstrating strong segmentation performance over subsurface inclusions and suggesting practical viability for industrial QC in Lithography-based Ceramic Manufacturing. While the results are compelling, the work is limited by a small dataset and a focus on one defect type, indicating clear directions for expanding defect types and dataset scale to bolster generalization and applicability in manufacturing environments.

Abstract

Non-destructive testing (NDT) is essential in ceramic manufacturing to ensure the quality of components without compromising their integrity. In this context, Optical Coherence Tomography (OCT) enables high-resolution internal imaging, revealing defects such as pores, delaminations, or inclusions. This paper presents an automatic defect detection system based on Deep Learning (DL), trained on OCT images with manually segmented annotations. A neural network based on the U-Net architecture is developed, evaluating multiple experimental configurations to enhance its performance. Post-processing techniques enable both quantitative and qualitative evaluation of the predictions. The system shows an accurate behavior of 0.979 Dice Score, outperforming comparable studies. The inference time of 18.98 seconds per volume supports its viability for detecting inclusions, enabling more efficient, reliable, and automated quality control.

Defect Segmentation in OCT scans of ceramic parts for non-destructive inspection using deep learning

TL;DR

The study tackles automated defect segmentation in OCT scans of ceramic parts for non-destructive testing by training a U-Net with a ResNet34 encoder on manually labeled inclusions. It systematically evaluates multiple loss configurations and employs preprocessing steps to ensure compatible input sizes, achieving a test DSC of 0.979 and high precision/recall (0.983/0.978) with an inference time of 18.98 seconds per volume. The MIR-OCT Scans dataset underpins the evaluation, demonstrating strong segmentation performance over subsurface inclusions and suggesting practical viability for industrial QC in Lithography-based Ceramic Manufacturing. While the results are compelling, the work is limited by a small dataset and a focus on one defect type, indicating clear directions for expanding defect types and dataset scale to bolster generalization and applicability in manufacturing environments.

Abstract

Non-destructive testing (NDT) is essential in ceramic manufacturing to ensure the quality of components without compromising their integrity. In this context, Optical Coherence Tomography (OCT) enables high-resolution internal imaging, revealing defects such as pores, delaminations, or inclusions. This paper presents an automatic defect detection system based on Deep Learning (DL), trained on OCT images with manually segmented annotations. A neural network based on the U-Net architecture is developed, evaluating multiple experimental configurations to enhance its performance. Post-processing techniques enable both quantitative and qualitative evaluation of the predictions. The system shows an accurate behavior of 0.979 Dice Score, outperforming comparable studies. The inference time of 18.98 seconds per volume supports its viability for detecting inclusions, enabling more efficient, reliable, and automated quality control.

Paper Structure

This paper contains 11 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Generation of binary masks
  4. Preprocessing
  5. U-Net architecture
  6. Network training and model evaluation
  7. Experimental Settings
  8. Dataset and evaluation metrics
  9. Implementation details
  10. Results
  11. Conclusion

Figures (3)

  • Figure 1: Detail of the original OCT image showing the inclusion defect with bounding box and binary mask.
  • Figure 2: Representative segmentation results (Prediction) compared with manual annotations (Ground Truth). The model consistently achieves accurate defect localization, showing strong concordance with expert labels.
  • Figure 3: Examples of a segmentation outcome (no Prediction) where the model did not capture the annotated defect, illustrating a failure case.