Roman Alekseevich Raclav (Bauman Moscow State Technical University) :


Articles:

533.6.011.5:004.622:004.855.5 A method for classifying aircraft surface elements for the numerical-analytical solution of aerodynamic problems

Kotenev V. P. (Bauman Moscow State Technical University/JSC MIC NPO Mashinostroyenia), Raclav R. A. (Bauman Moscow State Technical University), Sapozhnikov D. A. (JSC MIC NPO Mashinostroyenia/МГТУ им.Н.Э.Баумана), Chernyshev I. V. (Bauman Moscow State Technical University/JSC MIC NPO Mashinostroyenia)


doi: 10.18698/2309-3684-2017-3-83104


The study introduces an algorithm for classifying the aircraft surface elements based on a binary decision tree with threshold predicates. According to the initial description of the objects, we developed derived characteristics allowing for the classes to be separated with minimal losses. Moreover, we trained and verified the predicates on synthetic data and described an algorithm of obtaining the data for training. Low values of classification errors and ease of implementation make it possible to apply the algorithm for solving aerodynamic applied problems.


Kotenov V.P. ,Ratslav R.A. ,Sapozhnikov D.A. ,Chernyshev I.V. , A method for classifying aircraft surface elements for the numerical-analytical solution of aerodynamic problems .Маthematical Modeling and Computational Methods, 2017, №3 (15), pp. 83–104



533.6 Modeling a neural network to solve the problem of classifying air frame elements

Bulgakov V. N. (Bauman Moscow State Technical University/JSC MIC NPO Mashinostroyenia), Raclav R. A. (Bauman Moscow State Technical University), Sapozhnikov D. A. (JSC MIC NPO Mashinostroyenia/МГТУ им.Н.Э.Баумана), Chernyshev I. V. (Bauman Moscow State Technical University/JSC MIC NPO Mashinostroyenia)


doi: 10.18698/2309-3684-2018-4-5771


The paper introduces a neural network implemented to classify air frame surface ele-ments. Within the research, we generated a sample containing the surface parameters of classification objects. In order to avoid errors associated with different measurement scales, the criteria were scaled. According to synthetic data, the neural network was trained, and the proposed model was verified. The optimal configuration of the neural network was determined experimentally. As a criterion of optimality, we used the proportion of correct answers from the test and training samples, and carried out calibration and modification of individual model parameters. The classification results of the test sample by the optimal network were summarized in the error matrix. The most significant result was achieved when distinguishing the class of ellipsoids. Separate blocks of the matrix show that the neural network accurately distinguishes the classes of ellipsoids and hyperboloids. The study proposes some ideas for further modification of the algorithm in order to increase the proportion of correct answers when distinguishing the class of paraboloids.


Булгаков В.Н., Рацлав Р.А., Сапожников Д.А., Чернышев И.В. Моделирование нейронной сети для решения задачи классификации элементов корпуса летательного аппарата. Математическое моделирование и численные методы, 2018, № 4, с. 57–71.