In order to increase the flexibility, quality, and output of production machines, this thesis investigates an architecture that creates software agents with cognitive capabilities. These agents can learn from human experts how to weld to a penetration depth of 1.1 mm, or how to maintain a minimum kerf width during cutting. Using dimensionality reduction, classification, and reinforcement learning, they are able to provide themselves with feedback. Within an industrial environment they can monitor lack of fusion in zinc-coated steel lap weld experiments. Furthermore, in the closed-loop real-time control of laser power they maintained the processing goal to within 15% deviation while the speed altered at 650%.
«
In order to increase the flexibility, quality, and output of production machines, this thesis investigates an architecture that creates software agents with cognitive capabilities. These agents can learn from human experts how to weld to a penetration depth of 1.1 mm, or how to maintain a minimum kerf width during cutting. Using dimensionality reduction, classification, and reinforcement learning, they are able to provide themselves with feedback. Within an industrial environment they can monito...
»