Machine Learning Part Assessment for Functional Tolerance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for assessing the quality of manufactured gas turbine engine components, such as integrally bladed rotors, rely solely on geometric tolerances, leading to the rejection of functionally tolerant parts due to their dimensions being outside nominal ranges.
Innovation Solution
A dual branch assessment method combining physical testing and machine learning is employed, where data from multiple locations on the component is analyzed to determine functional tolerance, using machine learning models to evaluate curvature and profile data alongside physical stress and dynamic testing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual inspection of dimensions is used to assess component quality, then measurement precision is maintained, but productivity is reduced and functionally tolerant parts are incorrectly rejected
Solution Approach 1:
The patent replaces manual mechanical inspection with an automated system combining optical scanning and machine learning algorithms. The optical scanner captures 3D geometric data of the component, and machine learning models assess functional tolerance based on this data, eliminating the need for manual measurement while maintaining accuracy and significantly increasing assessment speed.
Solution Approach 2:
The patent creates a digital copy of the physical component through optical scanning. This digital replica contains precise geometric information that can be analyzed computationally without physically handling or measuring the actual component, enabling rapid automated assessment while preserving measurement precision.
2Manufacturing precision
If geometric tolerances are strictly enforced, then manufacturing precision is ensured, but loss of substance increases due to rejection of functionally tolerant parts
Solution Approach 1:
The patent transitions from enforcing fixed geometric tolerances to using functional tolerance assessment based on operational impact. Machine learning models evaluate whether dimensional variations actually affect component functionality, allowing parts with deviations outside traditional tolerance ranges to be accepted if they remain functionally acceptable, thereby reducing unnecessary rejections.
Solution Approach 2:
Instead of automatically rejecting parts that fall outside geometric tolerances, the system inverts the logic by presuming acceptance and only rejecting parts that are proven to be functionally defective through machine learning analysis. This reversal reduces the loss of functionally acceptable components while maintaining quality standards.
3Reliability
If comprehensive physical testing is performed on all components, then reliability is improved, but loss of time and productivity decrease
Solution Approach 1:
The patent applies partial physical testing rather than comprehensive testing to all components. Machine learning models first screen components and identify those requiring physical testing based on their geometric characteristics and risk profiles. This selective approach maintains high reliability for critical components while significantly reducing the time and resources spent on testing the entire production batch.
Solution Approach 2:
The patent performs preliminary machine learning-based functional tolerance assessment before conducting time-consuming physical tests. This preliminary screening identifies components that are likely to pass or fail, allowing physical testing resources to be focused only on borderline cases or high-risk components, thereby reducing overall testing time while maintaining reliability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of assessing the quality of a manufactured component includes the steps of manufacturing a component (22) and generating data at each of a plurality of locations through the component (22). The generated data is passed to a machine learning branch (35) wherein the generated data is compared to training data at each of the plurality of locations to determine whether the component (22) is of functionally tolerant dimensions at each of the plurality of locations. The manufactured component (22) is accepted (108) should the component be within functionally tolerant dimensions at each of the plurality of locations, and the component is rejected (106) if the component fails to be within functionally tolerant dimensions respectively, at each of the plurality of locations. A system (20) is also disclosed.