Development of a training data generator for training artificial neural networks for industrial inspection

Bachelor/study thesis Start: immediately The optical inspection of components in industry is crucial to ensure the high quality of the components produced. One possibility to automate this process is the use of camera systems in combination with machine learning algorithms in the form of artificial neural networks. These usually require a large number of images, especially of defective components, for training. The acquisition of these images is not always possible, or often economically not reasonable. Therefore, this work will investigate whether artificial training data is suitable for training artificial neural networks for inspection.

__________________________________________________________

Your subtasks

• Literature research on artificial neural networks in industrial inspection
• Development of a concept for training data generation based on a dataset of real images of defects
• Implementation of the concept
• Evaluation: analysis of the suitability of the generated training data and comparison with real training data

Your profile

• You are studying mechatronics, electrical engineering, mechanical engineering, computer science engineering or a comparable subject
• You have an interest in AI topics
• Ideally, you have programming skills (Python, etc.)
• You have a high degree of initiative and are inquisitive

If you are interested, please contact: Ole Schmedemann | ole.schmedemann@tuhh.de