Deep learning-based fault diagnosis system for mobile manipulators

Mobile manipulators play a crucial role in various industrial fields, ranging from production to logistics, healthcare services to disaster response, demonstrating their effectiveness across a broad spectrum. As industries expand globally, the growth of these manipulators and their reliable operation are becoming increasingly significant. This study introduces a novel methodology based on deep learning techniques for fault diagnosis in mobile manipulators. Contrasting traditional methods that require complex rules and expert knowledge, this research has conducted modeling on each joint of the manipulator and collected data through the application of movements to these joints. Faults were added to the gathered data, and deep learning methods, particularly convolutional neural networks, were utilized for fault identification. Various degradation rates on nine different joints were meticulously examined. To facilitate movement of the mobile robot manipulator from one area to another, five separate tests using distinct joint movements were performed. Five different scenarios for each joint were created, applying varying rates of faults in each scenario, culminating in a total of 255 distinct scenarios evaluated. These datasets were processed using Neural Network (NN) and Long Short-Term Memory (LSTM) algorithms available in Matlab's Toolbox, assessing the effectiveness of these methods in fault detection. The study compared normal operational data with faulty conditions. The dynamics of the Kuka Youbot were employed in this fault diagnosis study, with data obtained from the joints processed through an Artificial Neural Network (ANN) for decision-making, while CoppeliaSim was used during the simulation phase. This approach offers a more efficient alternative to traditional methods, enhancing sustainability and development.