Application of a neural network in the logical control of an electromechanical system
In some control and automation systems, logical transformations based on logic algebra equations are performed using software or through electrical circuits assembled on logical elements. This article presents a version of the software implementation of logical control as a result of the operation of a neural network. As an example, control is defined as the transformation of a binary set of one dimension into a set of another dimension, according to a given logical function. The conversion is based on the use of an artificial neural network of feedforward signal propagation. When training a neural network, a backpropagation algorithm is used. The training data uses a set of binary numbers as input and output vectors of the network and a logical function that establishes the relationship between them. The neural network and learning algorithm were created in the Python programming language using standard mathematical packages. After training the neural network with a given degree of accuracy, it is switched to the operating state and an input vector was supplied to the network input. The input vector consisted of a given set of binary numbers. In another variant, the coordinates of the input vector from a set of binary numbers randomly changed both in signal magnitude and duration during the operation period. The results of the operation of a neural network in implementing control in the absence and presence of interference in the input vector of the network are presented. The developed program code can be used in standard industrial controllers, both for the proposed transformation and for other logical transformations when changing the structure of the neural network and applying new training data.
Authors: D. V. Chernyshev
Direction: Informatics, Computer Technologies And Control
Keywords: control system, math modeling, Python programming language, neural network, learning algorithm, logical variables
View full article