The available (average) power of high-power lasers is steadily increasing. This poses the challenge of providing this power dynamically tailored to the respective laser processing application, be it surface structuring, cutting or 3D printing, in order to ensure efficient and high-quality processing. In dynamic high-power laser beam shaping, a compromise usually has to be made between the applicable amount of (average) laser power and the degrees of freedom for the beam shaping device. In general, the higher the damage threshold is, the fewer are the degrees of freedom for available beam shaping devices[1,2]. One way to overcome this deficit is to first shape the beam with a high resolution and low power output and then amplify the beam to the necessary laser power. The subsequent amplification introduces unwanted changes in the desired beam shape, which needs to be compensated. The current method to compensate the amplification induced changes is to exactly simulate the optical system at hand as well as the amplification process. For this purpose, an Iterative-Fourier- Transformation-Algorithm (IFTA) combined with an additional optimization is used. This method requires prior knowledge of all system and amplification defining parameters, which are non-trivial to determine. Another approach, pursued in this paper, is the use of an artificial neural network (ANN). The ANN is trained through the combinations of different phase masks and the resulting beam shape profiles. This training method should allow the ANN to indirectly map any optical system without any regard to its complexity. Through an appropriate choice of training samples and subsequent training the ANN is able to approximate the mapping function of the optical system including the amplification. The fully trained ANN generates phase masks for the beam shaping process in one step and thus allows highly dynamic beam shaping of arbitrary beam shape profiles.
|