Neural Network Feature Recognition for Rotor Blade Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The process of labeling training images for neural networks is time-consuming, costly, and laborious, especially for high-definition images, and is restricted by legal and contractual limitations, such as medical images or images of certain objects like airplane engines, which limits the availability for crowd-sourcing solutions.
Innovation Solution
A system using artificial neural networks with reinforcement learning to generate an action-guidance function that predicts the progression of defect features on rotor blades over time, allowing for automatic identification of defects without relying on labeled training images, thereby reducing the need for extensive image analysis and minimizing the impact of appearance variations due to lighting or shadows.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If crowd-sourcing approaches are used to label training images, then the time and cost involved in labeling is reduced, but not all images are available for public dissemination due to legal and contractual restrictions
Solution Approach 1:
The patent introduces an automated defect detection system that acts as an intermediary between image acquisition and defect identification, eliminating the need for manual labeling while handling restricted images. The system uses machine learning models trained on available data to automatically detect and classify defects in images that cannot be publicly shared, thus resolving the contradiction between reducing labeling time and maintaining image availability constraints.
Solution Approach 2:
The system enables self-service defect detection by automatically analyzing images without requiring human annotators. The automated detection algorithms process images independently, identifying defects through pattern recognition and comparison with known defect characteristics, thereby eliminating dependency on crowd-sourcing while maintaining efficiency.
2Measurement precision
If manual labeling of high definition training images is performed, then accurate training data is obtained, but the process is time-consuming, costly, and laborious
Solution Approach 1:
The patent replaces the mechanical process of manual labeling with an automated computer-based defect detection system. The system uses image processing algorithms and machine learning models to automatically identify and label defects in high-definition images, maintaining labeling accuracy while dramatically reducing the time and labor required compared to manual processes.
Solution Approach 2:
The system changes the parameters of the labeling process by transitioning from human-based subjective judgment to algorithm-based objective analysis. The automated system processes images with consistent, repeatable parameters, eliminating variability in human labeling while maintaining high accuracy through sophisticated image analysis techniques.
3Reliability
If traditional image analysis methods are used, then defect detection is performed, but the system is sensitive to appearance variations due to lighting or shadows
Solution Approach 1:
The patent applies preliminary image processing actions to normalize lighting and shadow effects before defect detection. The system pre-processes images by correcting illumination variations, enhancing contrast, and standardizing appearance characteristics, thereby eliminating the sensitivity to lighting conditions that plagues traditional methods while maintaining reliable defect detection.
Solution Approach 2:
The system dynamically adapts its detection parameters based on image conditions. The machine learning models adjust to varying lighting and shadow conditions in real-time, maintaining consistent defect detection performance across different environmental conditions by learning from diverse training data that includes various lighting scenarios.
Data Source
AI summary
A system includes one or more processors configured to analyze obtained image data representing a rotor blade to detect a candidate feature on the rotor blade and determine changes in the size or position of the candidate feature over time. The one or more processors are configured to identify the candidate feature on the rotor blade as a defect feature responsive to the changes in the candidate feature being the same or similar to a predicted progression of the defect feature over time. The predicted progression of the defect feature is determined according to an action-guidance function generated by an artificial neural network via a machine learning algorithm. Responsive to identifying the candidate feature on the rotor blade as the defect feature, the one or more processors are configured to automatically schedule maintenance for the rotor blade, alert an operator, or stop movement of the rotor blade.