AI Defect Detection Using Coreset Patch Matching for NDI

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Non-destructive inspection techniques generate vast amounts of complex data that are challenging to evaluate for detecting anomalies, particularly in components like aircraft wing panels, as faults are difficult to visually detect.

Innovation Solution

A two-tiered denoising process is implemented for anomaly detection models, involving image-level and patch-level noise elimination, followed by feature matching techniques using a coreset of selected patch features to establish a normal baseline, enhancing the model's efficiency and precision in identifying anomalies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional anomaly detection methods are used on NDI data, then the inspection process can be performed, but the accuracy and reliability of defect detection is insufficient due to the complexity and volume of data

Engineering Contradiction:
Improvedefect detection accuracyVSAvoiddata complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex NDI data into smaller, manageable patches that can be processed individually. By dividing the large-scale inspection data into discrete patches, the system can apply machine learning models to each patch separately, improving detection accuracy while managing computational complexity through localized analysis rather than processing the entire dataset at once.

Inventive Principle:
Principle #1Segmentation

2Reliability

If comprehensive NDI data collection is performed to ensure thorough inspection, then more potential defects can be detected, but the time and computational resources required increase significantly

Engineering Contradiction:
Improveinspection reliabilityVSAvoidinspection time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by pre-processing and organizing NDI data into standardized patches before actual anomaly detection. This includes data normalization, patch extraction, and creating a structured format that can be quickly processed by machine learning models, thereby reducing the time required during the actual inspection while maintaining comprehensive coverage and reliability.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If detailed feature analysis is performed on all NDI data to improve defect detection, then detection precision increases, but computational complexity and processing time increase

Engineering Contradiction:
Improveanomaly detection precisionVSAvoidcomputational power
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent applies local quality by focusing detailed feature analysis only on specific patches that show signs of potential anomalies, rather than performing exhaustive analysis on all data. The machine learning models are trained to identify regions of interest within patches and concentrate computational resources on those areas, achieving high detection precision while reducing overall computational power requirements.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4592948A1Ai based defect detection for non-destructive inspection
Publication Date: 2025.07.30 THE BOEING CO
  • EP4592948A1 patent drawingFigure 1
  • EP4592948A1 patent drawingFigure 2
  • EP4592948A1 patent drawingFigure 3A

AI summary

The present disclosure provides methods and techniques for anomaly detecting using feature matching models. A plurality of normal images are received. A plurality of patch features are generated by processing each of the plurality of normal images. A coreset comprising one or more coreset samples is generated, where the one or more coreset samples are selected from the plurality of patch features. A test image is received. One or more test patch features are generated by processing the test image. An anomaly score is generated by comparing the one or more test patch features with the one or more coreset samples.