Multi-Domain Defect Inspection for Real-Time Weld Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing defect inspection methods relying on post-weld visual inspections are inefficient, leading to high rework costs and production delays, and fail to utilize comprehensive sensor data like voltage, current, and gas flow rates effectively.

Innovation Solution

A method that preprocesses multiple input data from non-visual and visual domains, including time-series data transformation and image augmentation, to train a neural network model for defect inspection, integrating diverse sensor data for improved accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If post-weld visual inspection is used for defect detection, then inspection can be performed with simple equipment, but defects can only be identified after welding is completed leading to high rework costs and production delays

Engineering Contradiction:

Improvedefect detection accuracyVSAvoidproduction schedule delay

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent applies preliminary action by performing defect inspection during the welding process itself rather than after completion. Sensors collect voltage, current, and acoustic data in real-time, enabling defect detection before the welding process finishes, thus preventing production schedule delays and reducing rework costs

Inventive Principle:

Principle #10Preliminary action

2Adaptability or versatility

If existing deep learning methods focusing on single modal visual data are used, then the system complexity is low, but comprehensive quality evaluation using visual, thermal, and acoustic data cannot be performed

Engineering Contradiction:

Improvemulti-domain data utilizationVSAvoidneural network model complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent applies segmentation by dividing the multi-domain data processing into separate specialized neural network branches: one branch processes visual images, another processes thermal data, and a third processes acoustic signals. Each branch is optimized for its specific data type, and their outputs are then integrated to achieve comprehensive quality evaluation while managing system complexity

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements multi-functionality by designing a unified neural network architecture that can simultaneously process multiple types of sensor data (visual, thermal, acoustic) through different input branches. This allows the system to perform comprehensive defect detection using diverse data modalities within a single integrated model

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20250384541A1Method for Performing Defect Inspection Using Multiple Domain Data

Publication Date: 2025.12.18 VAZIL CO CO LTD

AI summary

Disclosed is a method for performing defect inspection using a neural network model, which is performed by one or more processors of a computing device. The method may include: obtaining multiple input data having different domains; preprocessing first input data associated with a non-visual domain among the multiple input data; obtaining first training data based on second input data associated with a visual domain among the multiple input data, and the preprocessed first input data; and training a neural network model for performing defect inspection based on the first training data.