Development of a CT Image Analysis Model for Cast Iron Products Based on Artificial Intelligence Methods

Area of Science:

• Materials Science
• Computer Science
• Manufacturing Engineering

Background:

• Automated defect detection in ductile iron castings is challenging due to visual similarities between voids and nodular graphite.
• Accurate qualitative assessment of casting process parameters is crucial for producing high-quality ductile iron parts.

Purpose of the Study:

• To evaluate the effectiveness of digital image classifiers for analyzing tomographic images of ductile iron castings.
• To develop a decision-making support system for the qualitative assessment of casting processes.

Main Methods:

• Development and testing of three distinct convolutional neural network (CNN) models with varying architectures.
• Analysis of two-dimensional tomographic images using these CNN models for image classification.
• Focus on differentiating between void images (potential defects) and nodular graphite images.

Main Results:

• The study demonstrates the feasibility of using CNN-based image classifiers for ductile iron casting analysis.
• Performance metrics indicate the potential for accurate classification, despite the inherent challenges.
• The developed models form a component of a larger decision-making system for quality control.

Conclusions:

• Digital image classification using CNNs offers a viable approach to enhance the quality assessment of ductile iron castings.
• Further development can lead to more robust automated systems for defect detection and process optimization.
• This research contributes to improving the efficiency and reliability of ductile iron casting production.