Composite Material Void Detection Using Boundary-Based Training Labels

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for identifying voids in composite materials face accuracy issues due to a lack of teaching data when voids are scarce, particularly in high-quality materials with fewer voids.

Innovation Solution

A method involving the generation of modified teaching data by changing labels assigned to cells outside a member in a composite material if the ratio of voids is below a threshold, followed by machine learning to create a learning model for accurate void identification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning is used to improve void identification accuracy, then identification precision improves, but accuracy decreases when voids are scarce due to lack of teaching data

Engineering Contradiction:

Improvevoid identification accuracyVSAvoidmodel performance on high-quality materials

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The patent applies preliminary action by modifying teaching data before machine learning training. Specifically, it artificially generates void regions in areas outside the member by changing labels of cells located outside the member boundary to void labels, thereby pre-preparing sufficient teaching data for void identification even when original data lacks enough void examples.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent changes the label parameter of cells outside the member from non-void labels to void labels. This parameter transformation creates synthetic void teaching data, enabling the learning model to be trained on sufficient void examples even when the original composite material images contain too few actual voids.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If teaching data is increased to improve model accuracy, then identification precision improves, but data processing complexity increases

Engineering Contradiction:

Improvevoid identification accuracyVSAvoiddata processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by selectively modifying only the cells located outside the member boundary while leaving the internal structure unchanged. This localized label modification approach generates additional teaching data without requiring complex global data processing or transformation of the entire dataset.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS20250252358A1Information processing method, information processing device, and information processing program for generating learning model capable of identifying void in composite material

Publication Date: 2025.08.07 IHI CORP

AI summary

An information processing method, an information processing device, and an information processing program acquire, for a first distribution data, teaching data in which a first label indicating a structural element of a cell is assigned to each cell constituting the first distribution data indicating a physical quantity distribution related to a member, calculate a ratio of cells to which a specific label indicating that the cells are voids is assigned, to the cells constituting the first distribution data based on the teaching data, generate modified teaching data from the teaching data by changing the first label assigned to the cell located outside the member to the specific label, if the ratio is less than the predetermined threshold, and perform machine learning based on the modified teaching data to generate a learning model that estimates the first label based on the first distribution data.