Parametric Surface Model for Precision Deviation Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting surface deviations on objects, particularly on flexible components and complex surfaces, are resource-intensive, dependent on lighting and camera setups, and fail to accurately detect small local shape deviations due to high shape tolerance and curvature issues.
Innovation Solution
A method using a parametric surface model for describing surfaces, allowing for a continuous mathematical representation of three-dimensional positions with parameters, enabling precise detection of deviations by comparing measurement data with a target surface model, which can be adapted for different surface areas and curvatures, and using techniques like spline surface descriptions and neural networks for efficient data processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional lighting and camera arrangements are used for surface defect detection, then detection capability is achieved, but setup effort and resource consumption increase significantly
Solution Approach 1:
The patent replaces complex mechanical lighting and camera arrangements with a computational approach using artificial neural networks that process standard images to detect surface defects. The mechanical/optical system is substituted with an information processing system that uses software-based image analysis and pattern recognition algorithms.
Solution Approach 2:
The patent creates a digital reference model (virtual copy) of the defect-free surface using neural network processing of standard images. This virtual reference model is then compared with actual surface images to detect deviations, eliminating the need for complex physical reference standards and measurement setups.
2Measurement precision
If three-dimensional scanning is performed to detect surface deviations, then large-scale shape deviations can be detected, but small local defects remain undetected and data volume increases
Solution Approach 1:
The patent extracts only the essential information needed for defect detection by using neural networks to process standard two-dimensional images rather than acquiring full three-dimensional data. The system extracts surface defect information from conventional images, eliminating the need for resource-intensive 3D scanning while maintaining detection capability.
Solution Approach 2:
The patent uses standard imaging that captures more information than strictly necessary for the intended purpose, then applies neural network filtering to extract only the relevant defect information. This approach uses excessive data capture (standard images contain more information than needed) followed by intelligent processing to obtain the required measurement precision.
3Measurement precision
If artificial neural networks are used to create reference component models from defect-free data, then local form deviations can be determined, but extremely large datasets are generated consuming significant resources
Solution Approach 1:
The patent creates a compressed virtual reference model in the form of a neural network that encapsulates the essential characteristics of defect-free surfaces. Instead of storing and processing massive datasets of reference images, the system uses a compact neural network model that can be efficiently stored and rapidly applied to detect defects in new images.
Solution Approach 2:
The patent transforms the reference data from a large set of detailed surface measurements into a neural network model with optimized parameters. This parameter transformation compresses the reference information into a compact representation that maintains detection accuracy while dramatically reducing storage and computational requirements.
4Reliability
If measurement is performed on entire surfaces to ensure complete coverage, then all areas are detected, but processing time and resource consumption increase
Solution Approach 1:
The patent captures complete surface information using standard imaging but then applies neural network processing to focus computational resources only on detecting actual defects rather than analyzing every pixel uniformly. The system performs partial detailed analysis only where needed (at detected defect locations) rather than exhaustive processing of the entire surface.
Solution Approach 2:
The patent replaces exhaustive mechanical processing of entire surfaces with intelligent software-based defect recognition that can rapidly identify and localize defects. The neural network system substitutes brute-force comprehensive analysis with targeted defect detection, maintaining complete coverage reliability while dramatically improving processing speed.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for detecting deviations of an object surface using a comparison between measured data of the surface and specified reference data. A surface description, at least portions of which are parametric, is generated as a target surface model using the specified reference data, and the comparison is carried out using the target surface model and the measured data. The invention likewise relates to a device for detecting deviations of an object surface using a comparison between measured data of the surface and specified reference data.