Assembly Unit Defect Detection Using Feature Constellations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting defects in assembly units during optical inspection are inefficient and require significant manual intervention, leading to potential defects being overlooked or propagated through the assembly process.
Innovation Solution
A computer system employs machine learning and computer vision techniques to automatically detect defects by analyzing feature constellations, defining nominal feature ranges, and flagging anomalous features, reducing reliance on manual verification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual verification methods are used for defect detection, then operational simplicity is maintained, but productivity is reduced and measurement precision is compromised
Solution Approach 1:
The patent replaces manual mechanical inspection with an automated optical inspection system that uses machine learning models to detect defects. The system captures images of assembly units, extracts feature constellations, and automatically compares them against reference models to identify anomalies, eliminating the need for manual verification while improving both accuracy and throughput.
Solution Approach 2:
The inspection system performs self-verification by automatically comparing detected feature constellations against stored reference constellations. The machine learning model independently identifies defects without requiring external manual intervention, enabling the system to serve itself in the defect detection process.
2Productivity
If automated detection systems are implemented, then productivity is improved, but device complexity increases
Solution Approach 1:
The automated inspection system is divided into distinct functional modules: image capture module, feature extraction module, comparison module, and defect identification module. Each module performs a specific task in the inspection workflow, making the overall complex system manageable through functional segmentation and independent optimization of each component.
Solution Approach 2:
The optical inspection system is designed to inspect multiple types of assembly units across different assembly stages using the same core technology platform. The machine learning models can be trained on various feature constellations and applied universally to detect defects in different product types, reducing the need for separate specialized systems.
3Loss of time
If manual verification is used, then ease of operation is maintained, but loss of time increases
Solution Approach 1:
Reference feature constellations are pre-captured and stored in the system during a setup phase before actual production inspection begins. During production, the system only needs to compare incoming images against these pre-established references, eliminating the need for real-time manual verification and significantly reducing inspection time while maintaining operational simplicity through automated comparison.
Data Source
AI summary
A method includes: accessing an initial image depicting a verified assembly unit; and detecting an initial constellation of features in the initial image. The method further includes: accessing a first image depicting an unverified assembly unit; detecting a first constellation of features in the first image; characterizing differences between corresponding features in the initial constellation of features and the first constellation of features; identifying a first dimension of a first feature of interest exhibiting a first difference exceeding a threshold difference; receiving manual verification the first feature of interest from the first constellation of features, the first dimension offset from a target dimension of the first feature of interest from the initial constellation of features; and defining a first nominal feature range for the first feature of interest, the range bounded by the first dimension and the target dimension.