Neural Network Estimator for Aluminum Alloy Damage Tolerance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the damage tolerance and ballistic properties of aluminum alloys are time-consuming and destructive, requiring numerous experimental tests to ensure compliance with specifications, especially in the aerospace and armor industries.
Innovation Solution
A non-destructive method using a neural network estimator that takes measured tensile properties and thickness data from aluminum alloy parts to estimate damage tolerance properties, such as toughness and ballistic resistance, allowing for reduced experimental testing through a confidence interval-based acceptance criterion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If experimental and destructive testing is used to determine damage tolerance properties, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The method performs preliminary non-destructive tensile tests to measure mechanical properties (yield strength, tensile strength, elongation) before final damage tolerance assessment. These preliminary measurements are used to train and apply a neural network estimator, allowing most parts to be evaluated quickly without time-consuming destructive testing. Only parts with estimated properties near acceptance thresholds undergo destructive verification.
Solution Approach 2:
The invention creates a computational model (neural network) that replicates the behavior and properties observed in destructive tests. The neural network is trained on data from destructive tests and then used to predict damage tolerance properties of new parts, replacing the need for repeated physical destructive testing while maintaining measurement precision.
2Reliability
If numerous experimental tests are conducted to ensure compliance with specifications, then reliability is improved, but productivity decreases
Solution Approach 1:
The method performs preliminary non-destructive characterization of mechanical properties through tensile tests, then uses a neural network to predict damage tolerance. This preliminary assessment allows most parts to be cleared quickly without extensive testing, maintaining high inspection throughput while ensuring reliability through the confidence interval approach for parts near acceptance thresholds.
Solution Approach 2:
The invention changes the assessment parameters from direct damage tolerance measurement (requiring destructive tests) to measurable mechanical properties (yield strength, tensile strength, elongation) that can be obtained through non-destructive tensile tests. The neural network transforms these parameter changes while maintaining specification compliance reliability through validated prediction models and confidence intervals.
3Measurement precision
If destructive testing is performed on parts, then measurement precision is improved, but loss of substance increases
Solution Approach 1:
The neural network creates a virtual copy of the destructive testing process, predicting damage tolerance properties from non-destructive tensile test data. This computational copy eliminates the need for physical destructive testing on most parts, preserving part material while maintaining measurement precision through accurate prediction models validated against destructive test data.
Solution Approach 2:
The method performs preliminary non-destructive characterization through tensile tests to gather sufficient data for accurate neural network prediction. This preliminary action provides the necessary information to estimate damage tolerance without destroying the part, reserving destructive testing only for verification cases where the estimated property is close to the acceptance threshold.
Data Source
Figure 1~2A
Figure 2B
Figure 3A
AI summary
The invention concerns a method for checking a damage tolerance property of a part made of aluminium alloy, comprising the following steps: - measuring at least one property representative of a tensile strength of the part; - using the property measured during step a) as input data (x,) of a neural network estimator; - estimating, using the estimator, the property representative of a tensile strength of the part; the method being characterised in that it comprises: - taking account of an acceptance threshold and comparing the property estimated during step c) with the acceptance threshold, taking into account a confidence interval; and - as a function of the comparison: - considering that the part satisfies the check; - or considering that the part does not satisfy the check.