X-Ray Inspection Device for Overlapping Article Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Inspection devices face reduced accuracy when inspecting goods containing plural articles that overlap, as the degree of overlap complicates distinguishing between overlapping regions and contaminant regions in transmission images.

Innovation Solution

An inspection device using machine learning to acquire features from overlapping goods, distinguishing between contaminated and uncontaminated image data by analyzing teaching images of overlapping articles, and incorporating virtual contaminants to enhance training data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning is used to acquire features from teaching images of overlapping articles, then inspection accuracy is improved, but device complexity increases

Engineering Contradiction:
Improveinspection accuracyVSAvoiddevice complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary machine learning training using teaching images that include articles in overlapping states. The learning component acquires features relating to overlapping articles before actual inspection, enabling the inspection component to accurately distinguish overlapping regions from contaminant regions during subsequent inspections without increasing operational complexity

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates teaching images that copy or simulate various overlapping configurations of articles. By training on these copied representations of overlapping states, the system learns to recognize and differentiate overlapping regions from actual contaminants, improving inspection accuracy while maintaining a single inspection device structure

Inventive Principle:
Principle #26Copying

2Measurement precision

If virtual contaminants are incorporated into teaching images to enhance training data, then inspection accuracy is improved, but data processing complexity increases

Engineering Contradiction:
Improveinspection accuracyVSAvoiddata processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

Virtual contaminants act as an intermediary training element. They are incorporated into teaching images to create artificial training samples that help the learning component understand contaminant characteristics. This intermediary approach enables the system to learn contaminant recognition without requiring extensive real contaminant samples, improving accuracy while managing data processing complexity through structured synthetic data generation

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If machine learning features are used to sort inspection images, then reliability of inspection is improved, but processing time increases

Engineering Contradiction:
Improvereliability of inspectionVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The machine learning model performs feature extraction and learns from teaching images in advance before actual inspection operations. This preliminary training phase enables the inspection component to rapidly sort inspection images into contaminated and uncontaminated categories during operation, improving reliability while minimizing processing time during actual inspection through pre-learned feature recognition

Inventive Principle:
Principle #10Preliminary action

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

Enhances inspection accuracy by preventing confusion between overlapping regions and contaminant regions, improving reliability through a combination of machine learning and conventional image processing.

Implementation Method 1

an inspection device that irradiates goods such as foods with radiation such as X-rays

Methodology Applied
Scientific EffectX-ray generation: X-Ray

Implementation Method 2

inspection images obtained from radiation that has passed through the goods

Methodology Applied
Scientific EffectX-ray detection: X-Ray

Data Source

PatentEP3754594B1Inspection device
Publication Date: 2025.10.01 ISHIDA CO LTD
  • EP3754594B1 patent drawingFigure 1
  • EP3754594B1 patent drawingFigure 2
  • EP3754594B1 patent drawingFigure 3

AI summary

It is an object of the invention to provide an inspection device that can inhibit a reduction in the accuracy of inspections of goods that include plural articles that sometimes overlap each other. An X-ray inspection device (10) irradiates, with radiation, goods (G) containing plural articles (A) having a predetermined shape and inspects the goods (G) on the basis of inspection images obtained from radiation that has passed through the goods (G) or radiation that has reflected off the goods (G). The X-ray inspection device (10) includes a storage component (51), a learning component (52c), and an inspection component (52d). The storage component (51) stores, as teaching images, at least the inspection images of the goods (G) that are in a state in which the plural articles (A) overlap each other. The learning component (52c) acquires, by machine learning using the teaching images stored in the storage component (51), features relating to the goods (G) that are in a state in which the plural articles (A) overlap each other. The inspection component (52d) inspects the goods (G) using the features that the learning component (52c) has acquired.