UDC

519.6

Authors

Lyudmila V. Kremleva*, Oksana I. Bederdinova*, Andrey N. Eliseev*
*Northern (Arctic) Federal University named after M.V. Lomonosov (Arkhangelsk, Russian Federation)

Abstract

The paper presents the approach description to the analysis of the design technology information
with the use of an artificial neural network (ANN) and a classic back-propagation algorithm. On its basis
the software for creating, training and functioning of a fully connected neural network of random topology
is developed. We have analyzed the process data of tests of contour form milling tools obtained
by the approximation method of experimental dependencies of the regression formulas. The results of
numerical experiments using the ANN are described. During the first experiment we have used a fully
connected ANN for the combination of “process material – feed direction”, including 3 neurons. In the
second experiment the ANN training including 6 neurons has been provided. We have assessed the accuracy
of the obtained data obtained by the ANN method in comparison with the classical approaches
of processing and the use of experimental data. The forecast of the output parameters, such as the
vibration level and the quality of the resulting surface using the ANN has a higher accuracy compared
with the assessment of phenomenological models. The method based on the neural networks allows us
to choose the cutting conditions for a given combination of “process material – feed direction” to provide
the required parameters of the technological operations. The ANN has no restrictions on the number of
the analyzed factors, can process the numeric, text or Boolean data types and reflects the subjective
evaluations of the research object by the designer, which is impossible with the classical experimental
approach using the regression models. Therefore, the ANN with the accumulated and analyzed knowledge
can generate the values of the quantitative characteristics of the designed technological operations,
taking into account the specific features of the production. This led us to the conclusion about the
prospects of further research of the ANN use in the analysis and storage of production data and for the
acquiring of new knowledge.

Keywords

artificial neural network, learning algorithm for artificial neural network, design of process operations, regression model, neuron, synaptic connection, numerical experiment

References

1. Norenkov I.P. Osnovy avtomatizirovannogo proektirovaniya [Fundamentals of Computer-Aided Design]. Moscow, 2009. 430 p. 
2. Korotkiy S. Neyronnye seti: osnovnye polozheniya [Neural Networks: Conceptual Issues]. Saint Petersburg, 2002. 10 p. Available at: http://www.rema44.ru/resurs/study/ai/materials/N1.DOC (accessed 25.02.2016). 
3. Korotkiy S. Neyronnye seti: algoritm obratnogo rasprostraneniya [Neural Networks: Back-Propagation Algorithm]. Saint Petersburg, 2000. 15 p. Available at: http://www.rema44.ru/resurs/study/ai/materials/N2.DOC (accessed 25.02.2016). 
4. Kremleva L.V. Povyshenie effektivnosti protsessa mekhanicheskoy obrabotki slozhnoprofil’nykh poverkhnostey liteynykh modeley iz drevesno-kompozitnykh materialov: dis. … d-ra tekhn. nauk [Improving the Efficiency of the Machining Process of Complex Surfaces of Wood-Composites Foundry Patterns: Dr. Eng. Sci. Diss.]. Arkhangelsk, 2010. 306 p.