Rethinking Phased Array Automated Defect Classification: From Image Recognition to Geometry-Based Algorithms
Tracks
NDT Methods
| Thursday, October 24, 2024 |
| 10:30 AM - 11:00 AM |
| 209/210 - Technical Session |
Details
The AI hype is nearing its mainstream peak and has made its way into non-destructive testing (NDT). While many have attempted to leverage machine learning for defect classification in phased array ultrasonic testing (PAUT), numerous challenges persist. Unfortunately, much of the fundamental principles of ultrasonic testing have been overlooked in favor of image-based machine learning training.
In this presentation, we will showcase a newly developed AI-based demonstration app that operates without the need for any ultrasonic training data. Utilizing the GPT-4.o model, this app uploads images and text prompts, and then GPT-4 classifies defects into one of two categories: planar flaws (such as lack of fusion, lack of penetration, and cracks) and non-planar flaws (such as inclusions). The app employs a clever trick to enhance angle dependency, demonstrating the advantages of geometry and trigonometry-based algorithmic AI systems over conventional image recognition models.
We will also discuss the ideal requirements for defect classification and envision what a sophisticated defect classification system should entail.
In this presentation, we will showcase a newly developed AI-based demonstration app that operates without the need for any ultrasonic training data. Utilizing the GPT-4.o model, this app uploads images and text prompts, and then GPT-4 classifies defects into one of two categories: planar flaws (such as lack of fusion, lack of penetration, and cracks) and non-planar flaws (such as inclusions). The app employs a clever trick to enhance angle dependency, demonstrating the advantages of geometry and trigonometry-based algorithmic AI systems over conventional image recognition models.
We will also discuss the ideal requirements for defect classification and envision what a sophisticated defect classification system should entail.
Speaker
Mr Jesse Groom
PAUT II
Rethinking Phased Array Automated Defect Classification: From Image Recognition to Geometry-Based Algorithms
Presentation Description
The AI hype is nearing its mainstream peak and has made its way into non-destructive testing (NDT). While many have attempted to leverage machine learning for defect classification in phased array ultrasonic testing (PAUT), numerous challenges persist. Unfortunately, much of the fundamental principles of ultrasonic testing have been overlooked in favor of image-based machine learning training.
In this presentation, we will showcase a newly developed AI-based demonstration app that operates without the need for any ultrasonic training data. Utilizing the GPT-4.o model, this app uploads images and text prompts, and then GPT-4 classifies defects into one of two categories: planar flaws (such as lack of fusion, lack of penetration, and cracks) and non-planar flaws (such as inclusions). The app employs a clever trick to enhance angle dependency, demonstrating the advantages of geometry and trigonometry-based algorithmic AI systems over conventional image recognition models.
We will also discuss the ideal requirements for defect classification and envision what a sophisticated defect classification system should entail.
In this presentation, we will showcase a newly developed AI-based demonstration app that operates without the need for any ultrasonic training data. Utilizing the GPT-4.o model, this app uploads images and text prompts, and then GPT-4 classifies defects into one of two categories: planar flaws (such as lack of fusion, lack of penetration, and cracks) and non-planar flaws (such as inclusions). The app employs a clever trick to enhance angle dependency, demonstrating the advantages of geometry and trigonometry-based algorithmic AI systems over conventional image recognition models.
We will also discuss the ideal requirements for defect classification and envision what a sophisticated defect classification system should entail.
Biography
Jesse Groom is a seasoned professional in the field of non-destructive testing (NDT). With extensive experience in various NDT techniques, Jesse has developed a deep understanding of phased array ultrasonic testing (PAUT) and its applications. Known for his interesting and unique projects, Jesse's innovative approach and commitment to advancing NDT practices highlight his dedication to the field.
