Measurement Data Analysis With Expert Feedback for Defect Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated algorithms for assessing object functionality in non-destructive testing are imprecise and require extensive manual review, leading to time-consuming and personnel-intensive quality control processes.
Innovation Solution
A method that combines machine learning algorithms with user-reviewed analysis data sets to train the machine learning algorithm, reducing the need for manual intervention by focusing on uncertain results and using simulated data to accelerate the training process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated algorithms are used to assess object functionality, then productivity is improved, but measurement precision deteriorates due to imprecision in defect assessment
Solution Approach 1:
The system implements feedback by submitting uncertain algorithm results to experts for review. Experts correct errors in algorithm assessments, and these corrected data are fed back into the system to retrain and improve the algorithm's accuracy, creating a continuous improvement cycle that resolves the precision-productivity contradiction
Solution Approach 2:
The algorithm performs self-improvement through automated retraining using corrected expert data. The system serves itself by automatically updating its assessment capabilities without requiring continuous manual intervention, thereby maintaining high productivity while improving measurement precision over time
2Measurement precision
If all algorithm results are submitted to expert review, then measurement precision is improved, but loss of time increases due to manual review requirements
Solution Approach 1:
The system applies local quality by selectively reviewing only uncertain algorithm results rather than all results. The algorithm identifies and flags only those cases with low confidence scores for expert review, while automatically accepting clear-cut cases, thereby reducing manual review time while maintaining high precision for critical assessments
Solution Approach 2:
The system uses partial action by having experts review only a subset (the uncertain portion) of algorithm results rather than all results. This partial review approach reduces the time investment required from experts while still sufficiently improving the overall system accuracy through targeted correction of problematic assessments
3Device complexity
If conventional algorithms are used for analysis, then device complexity is reduced, but measurement precision deteriorates due to inability to consider additional defect information
Solution Approach 1:
The system enables the algorithm to learn from expert corrections and automatically improve its assessment capabilities. The algorithm serves itself by retraining on corrected data, gradually acquiring the ability to consider additional defect information and contextual factors without requiring continuous manual intervention or complex rule-setting
Solution Approach 2:
The system implements feedback loops where expert corrections are fed back into the algorithm for retraining. This continuous feedback mechanism enables the algorithm to progressively improve its measurement precision by learning from expert judgments about defect relevance and functionality assessment, resolving the contradiction between simplicity and precision
Data Source
Figure 1
Figure 2~3
Figure 4a
AI summary
A computer-implemented method for analysing measurement data from a measurement of an object, said analysis assessing whether the object corresponds to a target condition, wherein the method (100) has the following steps: Determining (102) measurement data of a plurality of objects; determining (104) analysis data records from the measurement data for the objects, an analysis data record being assigned to one of the objects and having at least one analysis result about the conformity of the assigned object to the target condition; checking (106), by a user, the analysis results of at least some of the analysis data records; adapting (108) an analysis result of a checked analysis data record if the checking by the user results in a different analysis result about the conformity of the assigned object to the target condition; and transmitting (110) at least the adapted analysis data records to the learning algorithm, said learning algorithm modifying itself on the basis of the adapted analysis data records, in order to determine analysis data records from additional measurement data of objects by means of the modified learning algorithm; and the steps are carried out one after another or with an at least partial temporal overlap. Thereby, the invention provides an improved computer-implemented method (100) for analysing measurement data from a measurement of an object, and the method (100) reduces the amount of time spent by the user, has improved error susceptibility and saves costs.