Publication: Pnömatik sistemde yapay sinir ağları ile arıza tespiti
Abstract
PNÖMATiK SiSTEMDE YAPAY SiNiR AGLARI iLE ARIZA TESPiTi Bu çalısmada, pnömatik sistemde bir arıza meydana geldiginde bunun çok çabuk tespit edilmesi ve gerekli önlemlerin alınması için yapay sinir agı gelistirilmistir. Deney amaçlı gelistirmis oldugumuz sise dolum tesisinde bir arıza oldugunda sistemin degisik yerlerine konulan basınç sensörleri ve dogrusal cetvellerden gelen analog degerlere göre hiçbir müdahale olmadan sistemin neresinde arıza oldugu tespit edilebilmektir. Yapay Sinir Agı (YSA) ile deney seti üzerindeki sise yok, B kapak kapama silindiri çalısmıyor, C kapak sıkıstırma silindirine hava gelmiyor, sistemin hava basıncı yetersiz, su yok, sistemin hava basıncı az arızaları bulunmaktadır. Arızalar geri yayılımlı ve karsılastırmalı yapay sinir agı algoritmaları ile tespit edilmektedir. Çalısmada, Geri Yayılım Ag Algoritması ve Karsılastırmalı Ag Algoritmasında egitim, test ve performans sonuçlarına göre en uygun YSA egitim baslangıç degerleri belirlenmistir. Bu degerlere göre; Geri Yayılım Ag Algoritmasında gizli katman sayısı 1, gizli katman nöron sayısı 15, ögrenme oranı 0.3, momentum sabiti 0.5 ve egitim fonksiyonu lm (Levenberg Marquardt) seçilmistir. Agın transfer fonksiyonları tansig ve purelin seçilmistir. Hata miktarı 1e-4 seçilmistir. Karslastırmalı Ag Algoritmasında gizli katman sayısı 15, kohonen ögrenme oranı degeri 0.3, gruplardaki standart sapma 0.01 degeri seçilmistir. Bu en uygun degerlere göre farklı arıza durumlarında YSA nın egitim, test, son agırlık degerleri verilmistir. Arıza tespitinde geleneksel programlama veya PLC yardımı ile arıza tespiti yapmak mümkündür. Yapay sinir agı kullanılmasının sebebi kullanıcıya arızanın VI bulundugu yeri bildirmesi ve farklı sistemler üzerinde bu gelistirilen programın kullanılabilir olmasıdır. Amaç, pnömatik sistemde bir arıza oldugunda sistem üzerinden veri toplama kartı ile alınan verileri kullanarak yapay sinir agı vasıtasıyla gerçek zamanlı olarak arıza tespiti yapmaktır. Program farklı sistemlerdeki arızaları tespit edebilmek için esnek olarak gelistirilmistir. Program Matlab 7.0 da gelistirildi. Matlab’ teki Mablab Neural Network (Yapay Sinir Agı), GUI (Genel Kullanıcı Arayüzü) ve Simulink Toolbox’ ları kullanılmıstır. Elde edilen sonuçlara göre Geri Yayılımlı Ag Algoritmasının en uygun algoritma oldugu ve gelistirilen programın mekatronik sistemlerin bulundugu tesislerde arıza tespitine katkı saglayacaktır.
FAULT DIAGNOSIS ON PNEUMATIC SYSTEM WITH NEURAL NETWORK In this study, an artificial neural network is developed to find an error rapidly on pneumatic system. Also the ANN prevents the system versus the failure. The error on our experimental bottle filling plant can be defined without any interference using analog values taken from pressure sensors and linear potentiometers. The sensors and potentiometers are placed on different places of the plant. Neural network diagnosis faults on plant, where no bottle, cap closing cylinder B is not working, bottle cap closing cylinder C is not working, air pressure is not sufficient, water is not filling and low air pressure faults. The fault is diagnosed by artificial neural network with back propagation and competitive. Beginning value is selected according to back propagation and competitive algorithm’s learning rate, test and performance results. For back propagation algorithm; This value are taken while hidden layer neuron number 15, learning rate 0.3, momentum constant 0.5, learning function lm (Levenberg Marquardt). Transfer function is selected tansig and purelin. Goal is selected 1e-4. For competitive algorithm; hidden layer neuron number 15, kohonen learning rate 0.3, standard irregularity 0.01. It is possible to find an failure by using normal programming or PLC. The reason of using Artificial Neural Network is to give a information where the fault is. However, ANN can be used for different systems. The aim is to find the fault by using ANN simultaneously. In this situation, the error taken place on the pneumatic system is collected by a data acquisition card. The program is flexible. So, the program can be used VIII on different systems to find different errors. Matlab 7.0 is used to develop our software in this work. Matlab Neural Network, GUI(General User Interface) and Simulink are used in Matlab. And the finally, It is seen that the best available program is Back propagation Neural Network Algorithm. It is seen that the Back propagation Neural Network Algorithm is quite available program. It is observed that the algorithm is very capable program for many industrial plants which have mechatronic systems.
FAULT DIAGNOSIS ON PNEUMATIC SYSTEM WITH NEURAL NETWORK In this study, an artificial neural network is developed to find an error rapidly on pneumatic system. Also the ANN prevents the system versus the failure. The error on our experimental bottle filling plant can be defined without any interference using analog values taken from pressure sensors and linear potentiometers. The sensors and potentiometers are placed on different places of the plant. Neural network diagnosis faults on plant, where no bottle, cap closing cylinder B is not working, bottle cap closing cylinder C is not working, air pressure is not sufficient, water is not filling and low air pressure faults. The fault is diagnosed by artificial neural network with back propagation and competitive. Beginning value is selected according to back propagation and competitive algorithm’s learning rate, test and performance results. For back propagation algorithm; This value are taken while hidden layer neuron number 15, learning rate 0.3, momentum constant 0.5, learning function lm (Levenberg Marquardt). Transfer function is selected tansig and purelin. Goal is selected 1e-4. For competitive algorithm; hidden layer neuron number 15, kohonen learning rate 0.3, standard irregularity 0.01. It is possible to find an failure by using normal programming or PLC. The reason of using Artificial Neural Network is to give a information where the fault is. However, ANN can be used for different systems. The aim is to find the fault by using ANN simultaneously. In this situation, the error taken place on the pneumatic system is collected by a data acquisition card. The program is flexible. So, the program can be used VIII on different systems to find different errors. Matlab 7.0 is used to develop our software in this work. Matlab Neural Network, GUI(General User Interface) and Simulink are used in Matlab. And the finally, It is seen that the best available program is Back propagation Neural Network Algorithm. It is seen that the Back propagation Neural Network Algorithm is quite available program. It is observed that the algorithm is very capable program for many industrial plants which have mechatronic systems.