Stored Image Reclassification for Component Mounting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In component mounting machines, the large number of stored images makes it time-consuming and labor-intensive to investigate the causes of abnormalities, as existing systems only store images related to feeder and mounting head changes, and image recognition errors lead to misclassification of normal and abnormal images.
Innovation Solution
A stored image reclassification system that uses an inspection device to reclassify images based on their accuracy, measuring component suction orientation and using statistical processing and template matching to identify potentially misclassified images, thereby reducing the need for manual reclassification and narrowing down investigation targets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all images captured during production are stored in the storage device, then the completeness of stored images is improved, but the time and labor required to examine stored images increases
Solution Approach 1:
The patent segments the storage of images by creating different storage periods and criteria. Images are divided into those stored for long periods (abnormal images and images from components with high defect rates) and those stored for short periods or not at all (normal images from components with low defect rates). This segmentation reduces the total number of images requiring long-term storage and examination while maintaining reliability for investigating abnormalities.
Solution Approach 2:
The patent applies local quality by differentiating storage policies based on the specific characteristics of components and images. Instead of uniform storage, images are selectively stored based on component defect rates, image abnormality status, and investigation needs. This allows the system to maintain high reliability for critical cases while reducing overall storage burden.
2Productivity
If only images for feeder and mounting head changes are stored, then the storage burden is reduced, but the ability to investigate abnormalities caused by other devices is impaired
Solution Approach 1:
The patent makes the image storage system universal by expanding its scope beyond just feeder and mounting head changes. The system now stores images related to multiple devices and components on the component mounting machine, including but not limited to feeders and mounting heads. This is achieved by monitoring defect rates across various components and storing images based on comprehensive criteria, enabling the system to investigate abnormalities from any device.
Solution Approach 2:
The patent introduces dynamics by making the storage criteria adaptable and changeable. The system dynamically adjusts which images to store based on real-time defect rate data, component types, and abnormality patterns. This dynamic approach allows the storage system to evolve and cover multiple device types as needed, enhancing versatility while maintaining storage efficiency.
3Productivity
If image recognition system determines images as normal, then processing speed is improved, but misclassification of abnormal images as normal occurs
Solution Approach 1:
The patent implements feedback by using inspection device results to verify and correct image recognition classifications. Images initially classified as normal by the fast image recognition system are cross-checked against inspection device data. If discrepancies are found (e.g., components showing defects in inspection but classified as normal), the system flags these for reclassification. This feedback loop maintains processing speed while improving classification accuracy by catching misclassifications.
Solution Approach 2:
The patent applies partial action by performing thorough verification only on images that show potential misclassification risks, rather than checking every image extensively. The system uses the quick image recognition system for initial classification, then applies additional verification selectively to suspected cases based on inspection device feedback. This approach maintains overall processing speed while improving accuracy for critical cases.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A component picked up by a suction nozzle of a component mounting machine (12) is imaged by a camera, the captured image is processed by an image recognition system (17) to recognize the component, the image is determined to be normal or abnormal based on the recognition result, and in addition to classifying the image as a normal image or an abnormal image and storing the image in a storage device (20), component mounting boards, unloaded from a component mounting machine, are inspected with an inspection device (14). A stored image reclassification computer 22 acquires the inspection result from the inspection device, reclassifies the normal image stored in the storage device, based on the inspection result, as an image whose determination as a normal image is suspect or as an image whose determination as a normal image is not suspect, and then stores the normal image in the storage device.