Advanced beam shaping for laser materials processing based on diffractive neural networks.

We propose a method based on neural network training algorithms for the design of diffractive neural networks - with the aim to perform advanced laser beam shaping in the NIR/VIS spectrum for laser materials processing. The method enables the efficient design of systems including multiple cascaded diffractive optical elements (DOEs) and allows the simultaneous optimization for complex (intensity and phase) target field distributions in multiple target planes. The multi-target boundary condition in the optimization method offers great potential for advanced laser beam shaping.

Rigorous approach for unifying wave optics and state of the art rendering techniques.

With the capabilities of diffractive optics there is a rising demand for determining the light interaction of diffractive elements with arbitrary illumination and scenery. Since the structured surfaces' scale lies within the visible wavelengths and below, the light's interaction cannot be simulated with state of the art geometric optic rendering approaches. This paper presents a new model for the inclusion of wave-optical effects into Monte Carlo path rendering concepts. The derived method allows the coupling of a rigorous full-field approach with the concept of backward ray propagation through complex scenes. Therefore, the rendering of arbitrarily structured periodic optical components is now possible for complex sceneries for design, verification and testing purposes. The method's performance is demonstrated by comparing rendering results of complex sceneries including CDs with corresponding photographs.

Freeform optics design for extended sources in paraxial approximation exploiting the expectation maximization algorithm.

Freeform optics generating specific irradiance distributions have been used in various applications for some time now. While most freeform optics design algorithms assume point sources or perfectly collimated light, the search for algorithms for non-idealized light sources with finite spatial as well as angular extent is still ongoing. In this work, such an approach is presented where the resulting irradiance distribution of a freeform optical surface is calculated as a superposition of pinhole images generated by points on the optical surface. To compute the required arrangement of the pinhole images for a prescribed irradiance pattern, the expectation maximization algorithm from statistics is applied. The result is then combined with a ray-targeting approach for finding the shape of the corresponding freeform optical surface. At its current state, the approach is applicable to Gaussian input irradiances, single-sided freeform optics and for the paraxial case. An example freeform optical surface for laser material processing is shown and discussed demonstrating the performance and the limitations of the presented approach.