Glass optimization using neural network

The possibility of using neural network to handle discrete variables (glass materials) hi lens design is investigated. First, a two-dimensional neuron array is established, in which the minimum of the network energy function corresponds to a design result with controlled chromatic aberrations, acceptable monochromatic aberrations and with a proper combination of selected real glasses. The values of connection matrix and the bias currents are then calculated by means of ray tracing. They are applied to update the neuron asynchronously and randomly, until the valid solutions are achieved. 21 recommended Chinese optical glasses are selected to form a small catalog for the neural network model to reduce the number of the neurons and increase the convergence rate of optimization. A test program is developed using the Macro-PLUS language in CODE V and a double Gauss camera lens is successfully optimized with the model.