Borescope Pose Estimation via Virtual CAD Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing borescope technologies face challenges in accurately determining the pose and location of defects within inaccessible regions of objects, such as aircraft engines, due to limited depth sensor data and the need for large amounts of labeled training data for pose estimation.
Innovation Solution
An object pose prediction system that includes a training system and an imaging system. The training system uses machine learning models to learn different poses associated with a target object by receiving multiple training image sets, including 2D and 3D images. The imaging system processes real-time 2D images using the trained model to predict the pose of the test object and generate a 3D digital representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional borescope inspection methods are used, then the inspector can view images from inaccessible regions, but the exact borescope tip location and pose cannot be determined
Solution Approach 1:
The patent introduces a CAD model as an intermediary representation that bridges the gap between 2D borescope images and 3D spatial information. The CAD model serves as a virtual reference framework that enables pose estimation without requiring direct 3D sensing in the inspection region.
Solution Approach 2:
The patent creates a virtual copy of the physical object using a CAD model. This digital replica is then used for pose estimation by comparing rendered views of the CAD model with actual borescope images, eliminating the need for direct 3D measurement in the inaccessible region.
2Measurement precision
If deep learning models are trained with large amounts of labeled training data to improve pose estimation accuracy, then measurement precision improves, but the complexity and data requirements increase
Solution Approach 1:
The patent performs preliminary pose estimation using a simplified model or initial algorithm to generate approximate pose parameters. These preliminary results are then used to guide more accurate refinement, reducing the need for extensive labeled training data while maintaining high precision.
Solution Approach 2:
The patent replaces traditional mechanical approaches to pose estimation (which would require complex sensor systems and large training datasets) with a computational approach using CAD model rendering and image comparison, significantly reducing data requirements while maintaining accuracy.
3Measurement precision
If 3D sensors are added to capture depth information, then pose estimation accuracy improves, but the device complexity and cost increase
Solution Approach 1:
Instead of physically adding 3D sensors to the borescope, the patent creates a virtual 3D model (CAD copy) of the inspected object. Depth information is extracted by comparing the 2D borescope images with rendered views of the CAD model, eliminating the need for physical depth sensors in the inspection probe.
Solution Approach 2:
The patent substitutes physical 3D sensing hardware with a computational approach using CAD model rendering and 2D image analysis. This replaces complex mechanical/optical depth sensing systems with software-based depth inference, reducing device complexity while maintaining measurement precision.
Data Source
AI summary
An object pose prediction system includes a training system and an imaging system. The training system is configured to repeatedly receive a plurality of training image sets and to train a machine learning model to learn a plurality of different poses associated with the test target object in response to repeatedly receiving the plurality of training image sets. The imaging system is configured to receive a 2D test image of a test object, process the 2D test image using the trained machine learning model to predict a pose of the test object, and output a 3D test image including a rendering of the 2D test image having the predicted pose.