A tool named active lap could fit the aspheric optical surface to grit and polish controlled by CMAC neural networks is introduced. The solution to control the surface of active lap by CMAC neural networks is depicted based on the analyses of active lap control system. The structure and principle of CMAC neural networks model are introduced. A method to test and reconstruct the surface of active lap is put forward. We have the pulse voltage and Zernike polynomial coefficients of surface as input samples and output samples to train the CMAC neural networks until the output meet demand, then the deform of lap’s surface could be controlled by the trained CMAC neural networks. The original data from the micro displacement sensor matrix are interpolated and fitted to reconstruct the surface of active lap, the coefficient of Zernike polynomial fitted form reconstruct data by Gram-Schimdt method is looked as input samples to train the CMAC neural networks. The experiment of three units active lap was made to test the method mentioned above, the feasibility for multi units active lap controlling is discussed base on the simulation of 9 units and 18 units active lap, as well as the analysis of experiment and simulation results.
[1]
Jiang Wen-han.
Investigation of deformation experiment for active polishing lap
,
2004
.
[2]
A. Sideris,et al.
Learning convergence in the cerebellar model articulation controller
,
1992,
IEEE Trans. Neural Networks.
[3]
R H Nagel,et al.
Practical design and performance of the stressed-lap polishing tool.
,
1994,
Applied optics.
[4]
W. T. Miller,et al.
CMAC: an associative neural network alternative to backpropagation
,
1990,
Proc. IEEE.
[5]
James H. Burge,et al.
Stressed-lap polishing of 3.5-m f/1.5 and 1.8-m f/1.0 mirrors
,
1992,
Optics & Photonics.