Welding Image Defect Detection Using Weld Pool Validation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing welding defect detection technologies require complex databases to relate features extracted from images to defects, making them inconvenient and prone to reliability issues due to changes in welding objects, parameters, or imaging conditions, and may generate erroneous quality data from blown-out highlights or shadows.

Innovation Solution

A processing device uses a first model to directly detect weld pools and defects from welding images, determining the appropriateness of defect detection results based on weld pool detection, thereby eliminating the need for a feature-defect relationship database and enhancing user convenience while increasing reliability by validating defect existence through cumulative pixel sums and consecutive determinations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a database relating features to defects is used for defect detection, then defect detection can be performed, but the system becomes complex and requires frequent updates when welding conditions change

Engineering Contradiction:
Improvedefect detection reliabilityVSAvoiddatabase management complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts and removes the complex feature-defect relationship database from the system. Instead of using a database that requires manual management and updates, the invention directly detects defects from welding images using image processing algorithms, eliminating the intermediary database structure that causes complexity while maintaining defect detection capability

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent creates a universal defect detection system that can handle various welding conditions (different objects, parameters, and imaging conditions) without requiring database updates. The image processing approach is universally applicable across different welding scenarios, making the system adaptable without the complexity of maintaining condition-specific databases

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Ease of operation

If feature extraction from welding images is used, then defect detection is possible, but the system generates erroneous results from blown-out highlights or shadows

Engineering Contradiction:
Improvedefect detection convenienceVSAvoiddefect detection accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent applies beforehand cushioning by implementing preprocessing steps that compensate for potential image quality issues before defect detection. The system addresses the problem of blown-out highlights and shadows in advance through image processing techniques that normalize or correct these artifacts, ensuring accurate defect detection even under varying imaging conditions

Inventive Principle:
Principle #11Beforehand cushioning (Prior cushioning)

Solution Approach 2:

The patent introduces an intermediary image processing layer between the welding image and defect detection. This intermediary processing stage handles the transformation and normalization of images, filtering out artifacts like blown-out highlights and shadows before the actual defect detection occurs, thereby improving measurement precision while maintaining ease of operation

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4023383A1Processing device, welding system, processing method, and storage medium
Publication Date: 2022.07.06 KK TOSHIBA
  • EP4023383A1 patent drawingFigure 1
  • EP4023383A1 patent drawingFigure 2A~2B
  • EP4023383A1 patent drawingFigure 3A~3B

AI summary

The present application relates to a processing device and corresponding method and storage medium with corresponding program. A processing device (10) acquires a first detection result and a second detection result by inputting a first image to a first model. The first model detects a welding element and a defect according to an input of a welding image. The first image is imaged when welding. The first detection result relates to the welding element. The second detection result relates to the defect. The processing device (10) determines an appropriateness of the second detection result by using the first detection result.