2/22/24. GANs and Steel Welding Analysis
Frank L -
For the subtask that I’ve been given on identifying cracks in steel welding images, I was given a few updates from Hasnaa, the PhD student that I’ve been working with on the GAN project. My next few steps are going to be turning the data samples into workable images. So I’ll be transforming excel files full of numbers into colored pictures with a python program. Hasnaa walked me through some of the steps and the program is actually going to be a lot simpler than I thought, so it should be done pretty soon.
As promised, here’s a basic explanation on how Generative Adversarial Networks (GANs) work. A GAN is a type of machine learning framework that has two main parts to it: generator and discriminator. The generator is trained based on a dataset of images, and then tries to create its own images that replicate the same style, features, etc. The discriminator is trained based on the same dataset, and it works to determine whether or not the images given as an input are authentic images or fake ones. These two neural networks compete against each other, constantly trying one up the other. The generator wants to get so good at its job that it creates images “realistic” enough to fool the discriminator, meanwhile the discriminator tries to become the most accurate judge possible. Through repeated trials, both parts of the AI model learn and improve.
The research project that I’ve been aiding over the past few months applies a GAN model to analyze and generate resistance heat welding images. Resistance heat welding is just a special manufacturing process by the way. Some of the images look like this:
Comments:
All viewpoints are welcome but profane, threatening, disrespectful, or harassing comments will not be tolerated and are subject to moderation up to, and including, full deletion.