Simulation of pulsed laser material processing controlled by an extended self-organizing Kohonen feature map

We have simulated the control of laser material processing. The process being controlled was surface polishing. In the simulations we assumed pulsed UV laser beams of 100 ns duration. The controller was an artificial neural network (ANN): an extended self-organizing Kohonen feature map. The controller was trained, i.e. tuned, using examples of laser evaporation of one-dimensional “surfaces” and a Widrow-Hoff type delta-rule for error correction. The strength of the approach lies in the speed of parallel computing and the adaptive properties of the controller in a changing environment, such as drift in laser properties, optical components, etc. The results show an improvement in surface roughness of more than an order of magnitude after an ANN designed laser is shot onto a large surface. Restrictions of the model are discussed.