ARTIFICIAL NEURAL NETWORKS FOR MEASURING THE MOISTURE OF BULK MATERIALS

One of the urgent tasks of industrial production is improving the quality of verification of received, manufactured, and stored products, reducing energy consumption in the production of a final product. The solution to this problem is impossible without the creation of devices that control the quality indicators of materials within the limits of the permissible error. One of the most common indicators of the quality of bulk materials is moisture. In the national economy and industrial production, it is required to determine the moisture content of more than 500 different substances and materials. The paper shows the possibility of using neural networks (NN) to control the moisture content of bulk materials and gives explanations about the essence of input, output, and hidden layers, weight coefficients, etc. the number of layers of the neural network and its other parameters. A neural network with two input inputs and ten output signals is considered. Frequencies characterizing changes in capacities (dielectric constants), depending on changes in the moisture content of the test material (for example, grain) and the temperature of bulk materials, are taken as input signals. Neural network modeling is based on Kolmogorov\u27s theorem, which makes it possible to represent arbitrary continuous sets of functions in a one-dimensional boundary (limited) numerical form on a unit segment [0, 1]. Based on the research carried out, the main parameters of neuro-models have been established, which determine the quality of training a neural network