Fpga Implementation of 1d Convolutional Neural Network for Early Detection of Bearing Faults in Induction Motors
Loading...
Files
Date
2022
Authors
Dal, Barış
Journal Title
Journal ISSN
Volume Title
Publisher
İzmir Ekonomi Üniversitesi
Open Access Color
OpenAIRE Downloads
OpenAIRE Views
Abstract
Asenkron Motorlar, stabilite, düşük maliyet ve kolay bakım sağladıkları için çeşitli endüstriyel uygulamaların temel parçasını oluşturmaktadır. Asenkron motorların arızalanması, üretim zincirinin yavaşlamasına neden olarak ciddi bir para kaybına neden olabilir veya çevreye ve insan sağlığına zararlı olabilir. Asenkron motorlar, dönen parçalar arasında sürtünmeyi azaltıp, hız ve performansı artıran rulmanlar içerir ve bu rulmanlardaki arızalar en çok karşılaşılan motor arızalarıdır. Bu arızaların erken tespiti, sonradan oluşabilecek büyük problemlerle uğraşmak yerine, motorun çok daha ucuza tamir edilmesine veya değiştirilmesine olanak sağlar. Rulman hatalarının tespiti için literatürde farklı yaklaşımlar bulunmaktadır, ancak 1B-Evrişimsel Sinir Ağı (ESA) kullanarak sorunu çözmek için Alanda Programlanabilir Kapı Dizisi (FPGA)/Uygulamaya Özgül Tümdevre (ASIC) tasarımını kuran detaylı bir çalışma bulunmamaktadır. Bu yüksek lisans tezinde, asenkron motorlarda rulman hatalarının erken teşhisi için 1B-Evrişimsel Sinir Ağının FPGA üzerinde uygulaması sunulmuştur. Önerilen model, Case Western Reserve Üniversitesi (CWRU) tarafından sağlanan 4 farklı sınıfa ait (sağlıklı, bilye hatası, iç bilezik arızası ve dış bilezik arızası) titreşim sinyalleri veri kümesini kullanır. Önerilen mimarinin parametreleri, eğitilmiş modelden 32 bitlik kayan nokta sayıları olarak çıkarılır. Daha sonra, FPGA üzerinde bu sayıların depolanması için sayıların sabit nokta (8 bit) temsilleri belirlenir. Sonraki adımda, önerilen modeldeki her katmanın matematiksel eşleniği Verilog Donanım Tanımlama Dili (HDL) kullanılarak geliştirilmiş ve Vivado 19.1 yazılımı kullanılarak Xilinx-ZYBO Z7-10 FPGA kartında gerçeklenmiştir.
Induction motors are the core part of various industrial applications because they provide stability, low cost, and easy maintenance. The breakdown of the induction motors may lead to a slow down the production chain resulting in a serious money loss, or it may be harmful to the environment and people's health. Induction motors include roller bearings between rotating parts which reduce the friction and increase the speed and performance and the faults on these bearings are the most confronted motor failure. The early detection of these failures facilitates repairing or replacement of the motor with considerably less amount of money rather than dealing with tremendous problems that may occur later. The literature presents different approaches to detect the bearing faults, but there has not been a detailed work that establishes Field Programmable Gate Array (FPGA) /Application Specific Integrated Circuit (ASIC) design to solve the problem using 1D Convolutional Neural Network (CNN). In this thesis, FPGA implementation of 1D CNN for early detection of bearing faults in induction motors is introduced. The proposed model employs a benchmark Case Western Reserve University (CWRU) dataset which provides vibration signals of 4 classes (healthy, ball bearing, inner-race, and outer-race). Parameters of the proposed architecture are extracted from the trained model as 32-bit floating-point numbers. Next, the fixed-point (8-bit) representations of the parameters are determined for mapping to FPGA. Then, the mathematical model of each layer in the proposed model is developed utilizing Verilog and implemented on Xilinx-ZYBO Z7-10 FPGA board using Vivado 19.1 software.
Induction motors are the core part of various industrial applications because they provide stability, low cost, and easy maintenance. The breakdown of the induction motors may lead to a slow down the production chain resulting in a serious money loss, or it may be harmful to the environment and people's health. Induction motors include roller bearings between rotating parts which reduce the friction and increase the speed and performance and the faults on these bearings are the most confronted motor failure. The early detection of these failures facilitates repairing or replacement of the motor with considerably less amount of money rather than dealing with tremendous problems that may occur later. The literature presents different approaches to detect the bearing faults, but there has not been a detailed work that establishes Field Programmable Gate Array (FPGA) /Application Specific Integrated Circuit (ASIC) design to solve the problem using 1D Convolutional Neural Network (CNN). In this thesis, FPGA implementation of 1D CNN for early detection of bearing faults in induction motors is introduced. The proposed model employs a benchmark Case Western Reserve University (CWRU) dataset which provides vibration signals of 4 classes (healthy, ball bearing, inner-race, and outer-race). Parameters of the proposed architecture are extracted from the trained model as 32-bit floating-point numbers. Next, the fixed-point (8-bit) representations of the parameters are determined for mapping to FPGA. Then, the mathematical model of each layer in the proposed model is developed utilizing Verilog and implemented on Xilinx-ZYBO Z7-10 FPGA board using Vivado 19.1 software.
Description
Keywords
Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering, Asenkron motorlar, Induction motors, Evrişimli sinir ağları, Convolutional neural networks, FPGA, FPGA, HDL, HDL, Rulmanlar, Bearings, Sabit noktalar, Fixed points
Turkish CoHE Thesis Center URL
Fields of Science
Citation
WoS Q
N/A
Scopus Q
N/A
Source
Volume
Issue
Start Page
1
End Page
73
Collections
Sustainable Development Goals
9
INDUSTRY, INNOVATION AND INFRASTRUCTURE

