For our bachelor project, we were given a chance to work with Rebartek AS. They are a startup company that is trying to automate the process of welding rebars with the use of robots. Their goal is to automate the entire process of making rebars from putting them together to sending them out. Doing this removes the safety risks and reduces the amount of manual labor required. Thus, this project is important as we learn new technologies, apply them in actual life as a need for many construction projects, and create an automated process that fits well for the present days.

Keeping this in mind, they gave us the opportunity to create a quality inspection program to automatically check if the welding is good enough or if it would break during transport. To do this, they have supplied us with several datasets that contain hundreds of images of both good and bad weldings. Our job is to manipulate the images to reduce the amount of noise. Then, using machine learning, we achieve a 90% accuracy rate when checking if the welds are good or bad. Our goal is also to have more false negatives than false positives. We would rather have our system called as a good weld bad than calling it as a bad weld good. Thus, if we called it a bad weld good, it could break during transport or after putting it in a building. Instead, we have agreed with Rebartek AS to spend some extra time checking false negatives.