Infrared Composite Layup Inspection With Neural Defect Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for monitoring composite layups during manufacturing are prone to human error, require significant downtime, and are build-specific, limiting production efficiency and effectiveness.
Innovation Solution
A method using an infrared camera to capture reference images, manually annotate defects, and train a convolutional neural network to generate defect masks for real-time detection and identification in production layups, minimizing equipment setup and pre-programming.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual inspection by QA personnel is used to monitor composite layup quality, then detection accuracy is improved, but production time is significantly reduced due to halting the layup head operation
Solution Approach 1:
The patent replaces the mechanical/visual inspection system with an optical imaging system. Multiple cameras capture images of the composite layup during manufacturing, and image processing algorithms automatically detect defects. This substitution eliminates the need to halt production for inspection while maintaining high detection accuracy through automated image analysis.
Solution Approach 2:
The inspection process is integrated into the continuous layup operation. Cameras continuously capture images during the layup head operation, and defect detection occurs in real-time through automated image processing. This allows the layup head to maintain continuous operation without interruptions, achieving both high productivity and accurate defect detection.
2Measurement precision
If classical machine vision algorithms with multiple cameras are used for in-process monitoring, then defect detection capability is improved, but equipment setup complexity and pre-programming time are significantly increased
Solution Approach 1:
The patent employs a single camera system that performs multiple functions: capturing images during layup, detecting various defect types, and generating quality reports. The image processing algorithms are designed to be build-agnostic, automatically adapting to different composite structures without requiring extensive pre-programming. This universal system reduces equipment complexity while maintaining comprehensive defect detection capability.
Solution Approach 2:
The image processing system automatically calibrates and adapts to each new build configuration without requiring manual pre-programming. The algorithms self-adjust to detect defects specific to the current composite structure being manufactured, eliminating the need for complex setup procedures and extensive pre-programming while maintaining high detection accuracy.
3Measurement precision
If machine vision methods operate below a certain speed threshold to ensure detection accuracy, then defect detection quality is improved, but production rate is negatively impacted
Solution Approach 1:
The system captures images at periodic intervals during the layup process rather than requiring continuous slow operation. The layup head maintains its normal high-speed periodic motion, and images are captured at optimized intervals that ensure sufficient detection accuracy while allowing production to proceed at full speed. This periodic sampling approach eliminates the speed threshold limitation.
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
The method reduces layup head downtime, enhances detection accuracy, and improves production efficiency by automating defect detection across various composite layup geometries.
Implementation Method 1
capturing, using an infrared camera of a reference layup head, a series of reference images
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of detecting defects 508 in a composite layup includes capturing, using an infrared camera 406, reference images 500 of a reference layup 470 being laid up by a reference layup head 404. The method also includes manually reviewing the reference images 500 for defects 508, and generating reference defect masks 504 indicating defects 508 in the reference images 500. The method further includes training, using the reference images 500 and reference defect masks 504, a neural network 600, creating a machine learning model 602 that, given a production image 566 as input, outputs a production defect mask 568 indicating the defect location 510 and the defect type 516 of each defect 508. The method also includes capturing, using an infrared camera 406, production images 566 of a production layup being laid up 560 by the production layup head 562, and applying the model 602 to the production images 566 to automatically generate a production defect masks 568 indicating each defect 508 in the production images 566.