ISSN 1991-2927
 

ACP № 4 (58) 2019

Author: "Denis Igorevich Stenyushkin"

Vadim Viktorinovich Shishkin, Ulyanovsk State Technical University, Candidate of Engineering, Associate Professor; graduated from the Faculty of Radio-Engineering at Ulyanovsk Polytechnical Institute; Professor at the Measuring-Computing Complexes Department of Ulyanovsk State Technical University, Dean of the Faculty of Information Systems and Technologies at UlSTU, an author of articles in the field of automated design of industrial products and intellectual data analysis. [e-mail: shvv@ulstu.ru]V. Shishkin,

Denis Igorevich Stenyushkin, Ulyanovsk State Technical University, graduated from the Faculty of Information Systems and Technologies at Ulyanovsk State Technical University; Post-Graduate Student at the Measuring-Computing Complexes Department of Ulyanovsk State Technical University, an author of articles in the field of automated design of industrial products and intellectual data analysis. [e-mail: denisstenyushkin@yandex.ru]D. Stenyushkin,

Michael Genrikhovich Bron, R&D in ScanMaster Systems (IRT) Ltd. (Israel), graduated from the Faculty of Radio-Engineering at Ulyanovsk State Technical University; Vice President of R&D in ScanMaster Systems (IRT) Ltd. (Israel); an author of articles in the field of ultrasonic inspection in industry. [e-mail: misha@scanmaster-irt.com]M. Bron

Mathematical Models and Methods for Real-time Analysis of Railway Rails Ultrasonic Defectograms 38_8.pdf

The article describes a system of mathematical models and methods devoted to a real-time analysis of the ultrasonic defectograms of railway rails during a testing process. The system includes models and methods for a preliminary ultrasonic data processing including a data reading, a range adjusting and a data combining for separate channels and also for a defect search and classification. The preliminary data processing is based on the ultrasonic data reading with a signal queue and their further algebraic modifications aiming to make them suitable for a further processing. The defect search and classification is based on an artificial neural network of Simplified Fuzzy ARTMAP architecture that is modified in order to deal with the input array elements with a wide value range. A decision making method based on a decision tree is introduced in order to solve the occurring conflicts. The introduced models and methods can be effectively implemented basing on modern parallel computing approaches. The tests showed out that the defect recognition rate is not less than 85%.

Defectogram analysis, rails defects detection, neural networks.

2014_ 4

Sections: Information systems

Subjects: Information systems, Mathematical modeling.


Vadim Viktorinovich Shishkin, Ulyanovsk State Technical University, Candidate of Engineering, Associate Professor; graduated from the Faculty of Radio-Engineering at Ulyanovsk State Polytechnical Institute; Professor at the Measuring-Computing Complexes Department of Ulyanovsk State Technical University, Dean of the Faculty of Information Systems and Technologies at UlSTU, an author of articles in the field of automated design of industrial products and intellectual data analysis. [e-mail: shvv@ulstu.ru]V. Shishkin,

Denis Igorevich Stenyushkin, Ulyanovsk State Technical University, graduated from the Faculty of Information Systems and Technologies at Ulyanovsk State Technical University; Post-Graduate Student at the Measuring-Computing Complexes Department of Ulyanovsk State Technical University, an author of articles in the field of automated design of industrial products and intellectual data analysis. [e-mail: denisstenyushkin@yandex.ru]D. Stenyushkin,

Alexander Vyacheslavovich Mikheev, Ulyanovsk State Technical University, graduated from the Faculty of Information Systems and Technologies at Ulyanovsk State Technical University; Post-Graduate Student at the Measuring-Computing Complexes Department of Ulyanovsk State Technical University, an author of articles in the field of measuring-computing system design and intellectual data analysis. [e-mail: sr.alex.anderson@gmail.com]A. Mikheev

Automated Design of Artificial Neural Network Based Classifiers 37_14.pdf

The article describes a system of models and methods for automated design of neural network-based classifiers. The system bases on the closed design cycle with feedback in form of analysis of gathered experience. The system’s goal is decrease of design process’ time costs. Parameters space is utilized as a formal system for experience representation. Parameters space is a space of points that represent sets of classifiers’ parameters. The space is constructed from parameters, which values are being specified through whole classifier’s life cycle from specification up to operation. This formalism allows to formulate the experience and analysis operations performed over it in terms of sets and operations over them. This results into representation improvement and automation costs decrease. Experience analysis bases on comparison of parameters space points’ projections on dedicated directions. It allows to detect classifier design solutions that are convenient to reuse. These solutions include classifier architectures, architectural parameters and neuron input weights. Using the detected parameters values as initial values in a new classifier’s training and setting up process lead to time costs decrease of 15%.

Artificial neural network-based classifiers, classifiers design, parameters space, closed design cycle.

2014_ 3

Sections: Computer-aided engineering

Subjects: Computer-aided engineering, Artificial intelligence.


© FRPC JSC 'RPA 'Mars', 2009-2018 The web-site runs on Joomla!