Design Of Optical Free-form Surface By Weighted Superposition And Machine Learning Algorithm

Optical freeform surfaces are widely used in non-imaging optical systems,such as LED lighting,laser shaping,and so on.The etendue of the light source has a great influence on designing optical freeform surfaces,and it keeps non-zero in practical non-imaging optical systems.Therefore,it is of great significance to design freeform lens with such light sources of non-zero etendue.In this thesis,how to design freeform lens for extended LED source or laser to control light distribution is studied.The weighted superposition and machine learning algorithm in designing optical freeform surface have been mainly studied.The main content of the thesis is divided into three parts.1.An algorithm of superposition of two-dimensional weighted surfaces is proposed to design freeform lens to control light distribution for extended LED light sources.5 sampling points are taken on the light source surface with equal separation,which are used as 5 point light sources.Then the initial freeform surface is designed for any point light source.The finial freeform surface is generated by the overlap of the five weighed freeform surfaces.The relationship between the uniformity of irradiance distribution and the weight factors are constructed.Particle swarm optimization algorithm was used to seek the best weight factors.The optimal freeform surface is generated by the overlap of the five weighed freeform surfaces with the best weight factor.The light distribution of the extended LED light source is adjusted by using the freeform lens.The simulation result shows that the optimal freeform lens enables the uniformity of the irradiance of the target plane to reach 75%,which is 15%higher than that generated by the initial lens.2.The algorithms of superposition of weighted surfaces and weighted grids are employed to design freeform lens for laser shaping.The algorithms of superposition of weighted surfaces is proposed to design freeform lens to control light distribution for the divergent beam with circular aperture.The influence of weight factor on shaping effect is studied.By randomly generating 100 groups of weight factors in a certain range to design 100 groups of freeform lens.The best freeform lens is selected from the 100 groups of freeform lens.The best freeform lens enable the uniformity of the irradiance on target plane to reach83.2%,which is 10.2% higher than that of the initial lens.The algorithm of superposition of weighted grids on target plane is employed to designed freeform lens to transform the circular laser beam with Gaussian irradiance distribution to uniform irradiance distribution on rectangular target plane.The influences of the grids pattern on design of the freeform lens are studied.Three different fundamental grids were weighted and superimposed to form the final grid,with a weight coefficient of 0.30,0.35 and 0.35.With the superimposed grids,the freeform lens was designed to produce a uniform irradiance distribution on a rectangular target plane,with a uniformity of 78%.The assembly tolerances of freeform lens are studied and the influence of assembly tolerances on uniformity are analyzed.Also the range of assembly tolerances of the freeform lens are given.3.The machine learning and particle swarm optimization algorithm are employed together to design freeform lens for laser beam shaping.The program is programmed to automatically collect system parameters and irradiation uniformities as the training data,in which the weight factors,the size of the target plane and the distance from the lens to the target plane are the input data,and the irradiance uniformities of the target plane are the output data.The collected data are trained to generate a general neural network,which constructed the mapping relationship between uniformity and system parameters.Particle swarm optimization algorithm was used to find the optimal weight factors.With the optimal weight factors,the shaping lens is designed.This method does not need no ray-tracing in the process of optimization and overcomes the shortcoming of traditional optimization method that need tracing rays repeatedly in the iterative process.In order to verify the effectiveness of this method,two sets of shaping lenses are designed to produce uniform irradiance distribution on circular and rectangular target planes respectively,and the uniformity can reach about 90% and 85.64%.