Система неразрушающего контроля композиционных материалов на основе нейронных сетей ART-2 и FUZZY-ART

• Дата: 2013
• Автори: Еременко, В.С., Переденко, А.В.
• Формат: Стаття
• Мова: Russian
• Опубліковано: Інститут електрозварювання ім. Є.О. Патона НАН України 2013
• Назва видання: Техническая диагностика и неразрушающий контроль
• Теми: Научно-технический раздел
• Онлайн доступ: http://dspace.nbuv.gov.ua/handle/123456789/102589
• Назва журналу: Digital Library of Periodicals of National Academy of Sciences of Ukraine
• Цитувати: Система неразрушающего контроля композиционных материалов на основе нейронных сетей ART-2 и FUZZY-ART / В.С. Еременко, А.В. Переденко // Техническая диагностика и неразрушающий контроль. — 2013. — № 1. — С. 28-34. — Бібліогр.: 9 назв. — рос.

Solution of the problems of standardless diagnostics of pipes requires application of data processing methods, which are oriented to a wide range of control objects, allows fast and effective diagnostics, are adapted to variation of testing conditions and permit modification of program modules without any significant changes in the main software structure. This paper is devoted to investigation and software realization of modified ART-2 and Fuzzy-ART neural networks to solve the problems of classification of defects in honeycomb panels. Developed neural networks are used in the system of standardless diagnostics of products from composite materials. Structure and operating algorithm of developed neural networks are described. Structure and main modules of the developed software for operation with the described neutral networks are also presented. The advantages of the developed neural network and system as a whole are its architecture flexibility, high performance and reliability of data processing. The paper gives the results of investigation of the developed system based on ART-2 and Fuzzy-ART networks for diagnostics of technical condition of honeycomb panels. The classifier based on the described neural networks can automatically change its settings during training, reaching the highest reliability of control at detection and classification of subsurface defects in honeycomb panels, as well as defects located on the back side of the panel of 2 cm2 area at thickness of composite panel equal to 12.8 mm. Reliability of non-destructive testing with the specified classifier is equal to 90 ? 95%.