?
Применение нейронных сетей для прогнозирования прочности древесно-минерального композита
Integrated processing of wood raw materials and waste is an important task that the specialists of the forest complex face. It is im- portant to ensure the receipt of products of deep processing, which are in steady demand in successfully functioning industries, for ex- ample, in construction. Therefore, the technology for obtaining wood-mineral composite building materials made from low-quality wood raw materials is quite promising and relevant. The aim of the work is to develop an intelligent system for studying the effect of changing the ratio of individual components of the initial raw mixture on the strength of wood-mineral composite material (sawdust concrete). Such a system can provide predictive information about the expected strength of material to meet the required performance for thermal insulation or structural material. At the same time, it is possible to solve the issues of ensuring high performance and a competitive price of the final product. The methods of artificial feedforward neural networks, as well as networks of neuro-fuzzy infer- ence, are used in the work. As a training sample for these networks, the results of experimental studies conducted by the authors earlier are used. For the practical implementation of neural networks the MATLAB program is used. The forecast accuracy for the obtained neural networks is 67...78%. For the networks of neuro-fuzzy inference the forecast accuracy turned out to be slightly higher than for the feedforward neural network. On the control sample the average deviation was 12.8% for the feedforward neural network and 10.9% for the networks of neuro-fuzzy inference The resulting neural networks can be successfully adapted to work with other wood-mineral composites. The research materials can be used by manufacturers of wood-mineral composites.