Graph Neural Network Defect Detection for Additive Melt Pools
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting defects in additively-manufactured parts, such as voids in metal parts, are either time-consuming and expensive (like CT imaging) or destructive (physical inspection), making them unsuitable for high-volume applications.
Innovation Solution
Utilizing graph neural networks to analyze in situ measurements of melt pool characteristics during the additive manufacturing process, converting these measurements into graph data, and predicting defects through supervised learning with labeled datasets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If CT imaging is used for defect detection, then measurement precision is improved, but loss of time and manufacturing cost increase
Solution Approach 1:
The system performs preliminary defect detection during the additive manufacturing process itself by monitoring melt pool characteristics in real-time. This allows defects to be identified before the part is completely manufactured, eliminating the need for time-consuming post-build CT imaging while maintaining detection accuracy through continuous process monitoring
Solution Approach 2:
The patent replaces the mechanical/physical CT imaging system with a computational approach using graph neural networks that analyze optical data (melt pool images) captured during manufacturing. This substitution of detection methodology enables real-time defect identification without the time and cost overhead of traditional CT scanning
2Measurement precision
If physical inspection is used for defect detection, then measurement precision is improved, but the article is destroyed
Solution Approach 1:
The patent replaces destructive physical inspection methods with non-destructive optical monitoring and graph neural network analysis. By substituting mechanical/physical destruction-based detection with computational analysis of melt pool characteristics, the system achieves accurate defect detection while preserving the integrity of the manufactured part
Solution Approach 2:
The system creates a digital replica or model of the manufacturing process through graph representations of melt pool data. This digital copy allows for virtual inspection and defect detection without physically touching or damaging the actual part, enabling repeated analysis without degradation of the inspected object
3Productivity
If in situ monitoring is implemented, then productivity is improved, but device complexity increases
Solution Approach 1:
The graph neural network system serves multiple functions: it processes melt pool images, detects defects in real-time, provides feedback for process control, and generates predictions about final part quality. This multi-functionality consolidates what would otherwise require separate systems into a single integrated platform, managing complexity while enhancing productivity
Solution Approach 2:
The patent introduces graph neural networks as an intermediary layer between the physical manufacturing process and the control system. This intermediary transforms complex optical data into simplified graph representations that are easier to analyze and act upon, reducing the overall system complexity while enabling real-time monitoring and decision-making
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
Enables real-time, non-destructive defect prediction, allowing process halting if defects are likely, thereby saving time and material, and reducing the need for post-build inspections.
Implementation Method 1
storing a graph comprising a plurality of light intensity values measured in situ during an additive manufacturing process
Data Source
AI summary
In situ defect detection of additively-manufactured articles using graph neural networks are provided. One aspect includes a computing device comprising processing circuitry and memory storing instructions that, when executed by the processing circuitry, causes the processing circuitry to store a graph comprising a plurality of light intensity values measured in situ during an additive manufacturing process and to generate an output describing a predicted defect in the graph using a graph neural network, wherein the graph neural network has been trained using labeled training data generated by a process comprising storing a training graph comprising a plurality of training light intensity values measured in situ during a training additive manufacturing process of the training article, determining one or more defect locations of the training article, determining a plurality of training sub-graphs from the training graph, and pairing a training sub-graph with defect information.