Mobil Manipülatörler için Derin Öğrenme Tabanlı Arıza Teşhis Sistemi

Özetçe - Mobil manipülatörler, çeşitli endüstrilerde önemli bir rol oynamakta olup, üretim ve lojistikten sağlık hizmetleri ve felaket müdahalesine kadar birçok göreve katkı sağlamaktadır. Bu nedenle dünya genelinde çeşitli endüstriler büyüdükçe mobil manipülatörlerin de büyümesi gerekmekte ve bu da güvenilir çalışmalarının büyük bir öneme sahip olmasını gerektirmektedir. Bu makale, mobil manipülatörler için derin öğrenme tekniklerini kullanarak arıza teşhis sisteminin geliştirilmesine yönelik yeni bir yaklaşım sunmaktadır. Geleneksel arıza teşhis yöntemleri, karmaşık kurallara dayalı bir sistem ve uzman bilgisi gerektirir ki bu da sürdürme ve geliştirme açısından zorlayıcı olabilir. Bu çalışmada robot mobil manipülatörün her bir eklemi modellenerek, modellenen eklemlere belirli hareketler yaptırılmıştır ve veriler kayda alınmıştır. Sonrasında elde edilen verilere belli oranlarda eklem arızası verilmiştir ve eklem arızaları her bir eklem için aynı zamanda verilmiştir. Arıza verilen eklemlerde derin öğrenme yöntemi kullanılarak, gerçeğe en yakın sonuç elde edilmeye çalışılmıştır, önerilen sistemde, özellikle evrişimsel sinir ağları gibi derin öğrenme algoritmalarından yararlanarak, eklemlerine dayalı olarak arızaları otomatik olarak öğrenniştir ve Kuka Youbot'un dinamikleri arıza teşhisi için kullanılmıştır. Arıza, eklemlerinden elde edilen verilerden tespit edilmiştir ve bu veriler, Yapay Sinir Ağı (YSA) kullanılarak bir karar yapısı aracılığıyla işlenerek karar verilmiştir.   Abstract –  Mobile manipulators play a significant role in various industries, contributing to a wide range of tasks from production and logistics to health care and disaster response. As various industries grow globally, there's a corresponding need for the growth of mobile manipulators, emphasizing the importance of their reliable operation. This paper presents a novel approach to developing a fault diagnosis system for mobile manipulators using deep learning techniques. Traditional fault diagnosis methods rely on a complex rule-based system and expert knowledge, which can be challenging to maintain and develop. In this study, each joint of the robotic mobile manipulator was modeled, subjected to specific movements, and the data were recorded. Subsequently, joint failures were introduced to the collected data at certain ratios, and these failures were simultaneously applied for each joint. Using deep learning techniques on the joints with induced faults, efforts were made to obtain results closest to reality. The proposed system, deep learning algorithms like convolutional neural networks, automatically learns faults based on their joints. The dynamics of the Kuka Youbot were used for fault diagnosis. Faults were detected from the data obtained from its joints, and this data was processed through a decision structure using Artificial Neural Networks (ANN) to make a determination.

Mobile manipulators play a significant role in various industries, contributing to a wide range of tasks from production and logistics to health care and disaster response. As various industries grow globally, there's a corresponding need for the growth of mobile manipulators, emphasizing the importance of their reliable operation. This paper presents a novel approach to developing a fault diagnosis system for mobile manipulators using deep learning techniques. Traditional fault diagnosis methods rely on a complex rule-based system and expert knowledge, which can be challenging to maintain and develop. In this study, each joint of the robotic mobile manipulator was modeled, subjected to specific movements, and the data were recorded. Subsequently, joint failures were introduced to the collected data at certain ratios, and these failures were simultaneously applied for each joint. Using deep learning techniques on the joints with induced faults, efforts were made to obtain results closest to reality. The proposed system, deep learning algorithms like convolutional neural networks, automatically learns faults based on their joints. The dynamics of the Kuka Youbot were used for fault diagnosis. Faults were detected from the data obtained from its joints, and this data was processed through a decision structure using Artificial Neural Networks (ANN) to make a determination.