Feasibility of using neural networks and genetic algorithms to predict and optimize coated paper and board brightness

The final brightness of coated paper depends on a large number of interacting variables, including top- and bottom-coat formulation and coater settings. This paper reports the use and adjustments of neural networks to predict the brightness of a double-coated paper product based on a historical database. The trained neural network was integrated with a genetic algorithm to achieve a desired brightness at minimum cost. The neural network predicted brightness to within 3% of the true value. The optimum solution proposed by the genetic algorithm was 14% cheaper than the best solution in the database and appeared technically feasible and consistent with conventional understanding of the coating process. The accuracy of both the neural network and genetic algorithm was substantially improved by factorial experiments to maximize performance. For this complex optimization, the capabilities of the genetic algorithm appeared to be limited by computer hardware and software resources. To our knowledge, this represents the first use of an integrated neural network and genetic algorithm in the paper industry.