Högskolan i Skövde

Quality inspections are essential in almost all manufacturing processes and can be undertaken using an automatic vision system. The main challenge when using AI-based vision systems is gathering enough training data for the AI to perform well. This study investigates the use of simulationgenerated data to address this challenge. A real-world industrial case study evaluates the feasibility of simulationgenerated images for training an AI to be used in an industrial vision system for quality inspection. Results from the study show that the approach has potential, but that simulationgenerated images cannot be used solely. However, real-world images must be mixed into the training data set to achieve satisfactory results.