Component Inspection Using Adaptive Anomaly Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous visual inspection systems for components are limited by predefined libraries and labeled training data, making them non-robust to environmental changes and defect appearance variations, and require significant modifications for different tasks, thus restricting their versatility and effectiveness in detecting anomalies.
Innovation Solution
A method and system that record sensor readings using multiple sensors, determine appearance metrics, and update nominal metrics based on differences exceeding thresholds, allowing for adaptive and robust anomaly detection by varying sensing parameters such as imaging resolution, position, and lighting conditions, enabling the identification and detailed inspection of anomalous regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If predefined libraries and labeled training data are used for autonomous visual inspection, then the system can detect known defects, but it becomes non-robust to environmental changes and defect appearance variations
Solution Approach 1:
The system changes the parameter of appearance metric calculation from fixed predefined values to dynamically updated nominal appearance metrics that adapt to environmental conditions and defect variations, allowing the inspection system to maintain reliability across varying environments
Solution Approach 2:
The system performs self-updating of the nominal appearance metric based on sensor readings from the component being inspected, enabling it to adapt to environmental changes and defect variations without requiring external retraining or manual intervention
2Reliability
If multiple sensing parameters are used to inspect anomalous regions, then detection confidence improves, but inspection time and system complexity increase
Solution Approach 1:
The system applies different sensing parameters locally only to regions identified as anomalous, rather than uniformly across the entire component, thereby improving detection confidence for suspicious areas while minimizing additional inspection time
Solution Approach 2:
The inspection process is made dynamic by adaptively selecting which regions require additional sensing parameters based on real-time anomaly detection, allowing the system to concentrate resources where needed and reduce overall inspection time
Data Source
AI summary
A method of inspecting a component for anomalous regions includes recording a plurality of sensor readings using one or more sensors, each reading corresponding to a different region of the component; determining an appearance metric for each reading; and determining a nominal appearance metric based on individual values of the appearance metric for a first subset of the readings. The method includes, for a particular sensor reading outside the first subset: determining a difference between the nominal appearance metric and the appearance metric of the particular sensor reading; updating the nominal appearance metric based on the particular sensor reading; and, based on the difference exceeding a threshold: determining that the particular sensor reading is anomalous and corresponds to an anomalous region of the component, and recording additional sensor readings of the anomalous region using one or more sensing parameters that differ from those used to record the anomalous sensor reading.