Open AccessStudy conclusion mechanisms in hybridization of neural and logic intelligent systems
Author(s) -
А. М. Пищухин,
Gulnara Akhmedyanova
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/843/1/012021
Subject(s) - artificial neural network , computer science , set (abstract data type) , position (finance) , signal (programming language) , control logic , robot , matching (statistics) , artificial intelligence , task (project management) , function (biology) , division (mathematics) , algorithm , mathematics , engineering , arithmetic , statistics , systems engineering , finance , evolutionary biology , economics , biology , programming language
The paper investigates the integration problems neural networks sections allocated in accordance with the functions performed. As an initial example, we study two trained neural matrices, one of which performs the function of recognizing the position of the manipulator and its speed, and the second generates responses using the corresponding robot drives. In addition to training mechanisms based on the adjustment of transfer weights, the mechanisms based on the supply of excitation / inhibition waves from the edges of the neural network are also affected. At the same time, training is reduced to multiple forced guiding of the robot manipulator along the target path, during which the desired paths of beatings are automatically found using comparators. The matrices are trained separately and then joined using a standard signal generated by a special unit. The division of the neural network into two sections allows building the necessary logic between them. This hybridization significantly expands the range of tasks. As a test problem, the task of searching for a given part from a jointly located set of parts in general position was selected. At the same time, the control system finds a given part, rotates its predetermined model until a matching image is obtained, and thereby understanding how the part is located and how best to capture it occurs.