Defect Image Separation in Thermogram Sequences Using Weighted Naive Bayesian Classifier
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for separating defect images from thermogram sequences, such as Fuzzy C-means, do not deeply analyze the physical meanings of transient thermal responses, leading to reduced accuracy in defect separation.
Innovation Solution
A method using a weighted Naive Bayesian classifier and dynamic multi-objective optimization to classify transient thermal responses based on extracted features like energy, temperature change rates, and average temperatures, selecting representative responses for enhanced defect image separation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If Fuzzy C-means is used to classify TTRs, then the classification process is simple, but the accuracy of defect separation deteriorates because physical meanings are not deeply analyzed
Solution Approach 1:
The patent transforms the classification approach by changing from simple distance-based clustering to a parameter-driven classification system. Five physical parameters (energy, average temperature, maximum temperature, temperature change rate during heating, temperature change rate during cooling) are extracted and discretized to create a comprehensive feature space that captures the physical meanings of TTRs, thereby improving classification accuracy without excessive complexity
Solution Approach 2:
The patent introduces a Weighted Naive Bayesian Classifier as an intermediary between raw TTR data and defect separation results. This classifier acts as a mediator that systematically processes the five extracted parameters through probability calculations and weighted decision rules, providing a rational bridge between physical parameter analysis and accurate defect identification
2Measurement precision
If more physical parameters are extracted and analyzed, then the accuracy of defect separation improves, but the complexity of the classification system increases
Solution Approach 1:
The patent segments the complex classification task into five distinct physical parameter extraction steps, each focusing on a specific aspect of TTR behavior. By dividing the analysis into separate parameter extraction modules (energy, average temperature, maximum temperature, heating rate, cooling rate), the system manages complexity through modular organization while comprehensively capturing physical meanings
Solution Approach 2:
The patent implements dynamic multi-objective optimization to adaptively determine the weights of different physical parameters based on the specific characteristics of the thermogram sequence. This dynamic adjustment allows the system to optimize the balance between using multiple parameters for accuracy while adapting the system complexity to the actual problem requirements rather than using fixed complex structures
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
This approach improves the rationality of clustering and accuracy of defect separation by deeply analyzing physical meanings in transient thermal responses and dynamically optimizing representative responses, resulting in a more precise defect image extraction.
Implementation Method 1
Infrared thermal image detection is one kind of NDT, which obtains the structure information of material through the control of heating and the measurement of surface temperature variation
Implementation Method 2
the distribution of Joule heat can be affected by the location of the defect(s) of the material under test. The high Joule heat leads to high temperature area and the low Joule heat leads to low temperature area
Data Source
AI summary
The present invention provides a method for separating out a defect image from a thermogram sequence based on weighted naive Bayesian classifier and dynamic multi-objective optimization, we find that different kinds of TTRs have big differences in some physical quantities. The present invention extracts these features (physical quantities) and classifies the selected TTRs into K categories based on their feature vectors through a weighted naive Bayesian classifier, which deeply digs the physical meanings contained in each TTR, makes the classification of TTRs more rational, and improves the accuracy of defect image's separation. Meanwhile, the multi-objective function does not only fully consider the similarities between the RTTR and other TTRs in the same category, but also considers the dissimilarities between the RTTR and the TTRs in other categories, thus the RTTR selected is more representative, which guarantees the accuracy of describing the defect outline. And the initial TTR population corresponding to the approximate solution for multi-objective optimization is chosen according to the previous TTR populations, which makes the multi-objective optimization dynamic and reduces its time consumption.