350 rub
Journal Neurocomputers №11 for 2011 г.
Article in number:
Intelligent information system with a neural network for the condition forecasting of mechanical details connection type «shaft - bushing»
Authors:
E. V. Teslenko, V. V. Andreev
Abstract:
Connections type «shaft-bushing» have received the widest application in mechanical engineering. Necessary connections qualities are provided by various size combinations of interfaced surfaces which are called landings. There are three methods of landing setting: empirical, similarity and calculation. In Computer-aided design (CAD) the complete design solutions with the indication of such tolerance zone and such element landings, which have demonstrated their value in practice, are picked out. Common faults of the listed methods - complexity of similarity signs revealing; probable use of erroneous recommendations, absence of the authentic data about influence of the form deviation and interfaced surfaces quality on working capacity of the connection. Well-known design procedures for connections «shaft- bushing» are rough-and-ready methods because they don't take into consideration all the factors influencing the connection. In the research we suggest the technique and the program for the condition forecasting of connection type «shaft-bushing» by the use of neural network algorithms. The input data for neural network work is the set which is made by the factors influencing formation of tightness in connections. Dimensional factors arrive from 3D-model of the detail. The predicted tightness limits received as a result of surface connection are the output data (result of system work). The task of the neural network is the selection of weight coefficient for tightness forecasting. The feed-forward artificial neural network (ANN) has been generated to solve the problem. For ANN-creation and -training we used software Neural Network Wizard. Training was made on the basis of the training sample containing 400 lines of considered variants connections examples. Efficiency indicators were defined from the point of error size in tightness limits forecasting as a result of the whole range of factors action. The RMS-error on control sample was the criterion of training. In the course of modeling we used three- and four-layer feed-forward (without feedback) with nonlinear function of activation ANN. We subjected to training neural networks with number of neurons in the latent layer from 20 to 29. The program forecasting the condition of connection type «shaft-bushing» is included into intelligent information system of the automated technological support of CAD-designing. Objectively existing restriction of training sample volume is offered to compensate improving a network with the use of accumulated data. As a result of constant gathering of experimental data about durability of connections with tightness and their arrival on the input of a neural network a network gets trained in wider range of the data. It allows forecasting tightness more precisely. In the developed information system we can make a proved, more exact, carried out in an automatic mode landing choice. It will provide reliability of connections work and will improve mechanical engineering products quality.
Pages: 26-31
References
  1. ГОСТ 25347-82 ОНВ. Единая система допусков и посадок. Поля допусков и рекомендуемые посадки.
  2. Дунаев П.В., Леликов О.П., Варламова Л.П. Допуски и посадки. Обоснование выбора. М.: Высш. шк. 1984.
  3. Гречищев И.С., Ильяшенко А.А. Соединения с натягом: Расчеты, проектирование, изготовление. М.: Машиностроение. 1981.
  4. Басов К.A. ANSYS: Справочник пользователя. М.: ДМК-Пресс. 2005.
  5. АлямовскийA.A. SolidWorks/COSMOSWorks. Инженерный анализ методом конечных элементов. М.: ДМК-Пресс. 2004.
  6. ELCUT. Моделирование двумерных полей методом конечных элементов. Версия 5.8. СПб.: ПК «ТОР». 2010.
  7. Андреев В.В., Тесленко Е.В. Автоматическое формирование массива конструктивно-технологических признаков деталей интеллектуальной информационной системой // Вестник БГТУ им. В.Г. Шухова. 2010. № 3. С. 170-172.
  8. Каллан Р. Основные концепции нейронных сетей: Пер. с англ.  М.: ИД «Вильямс». 2001.