(General session) Mining and underground construction machineries / Rock Engineering / Resource based economy and social system / Mining technologies

[2K0201-11-02] Bearing health diagnosis system of excavator motor using deep learning

○Dorjsuren Yandagsuren^1,3, Tatsuki Kurauchi¹, Hisatoshi Toriya¹, Tsuyoshi Adachi ¹, Youhei Kawamura² (1. Akita University, 2. Hokkaido University, 3. Mongolian University of Science and Technology)

司会：児玉淳一 (北海道大学)，安達毅 (秋田大学)

Keywords:Bearing diagnostic, Excavator electric motor, Vibration analysis , Signal processing, Deep learning

Excavators are very important machines to mining technological processes, particularly in open-pit mines. Many types of excavators are being used in mines, such as draglines, shovels and hydraulic excavators. Reliability is required for all of them, and motors strongly influence to reliability. Bearings are important parts of the motor, and their failure takes 51% of total motor failures. Therefore, we considered about motor bearing’s health of the electric rope shovel, which is used in the Baganuur mine in Mongolia. The model of a shovel is EKG-5a, and it is applied for coal excavating. In the past, motor bearing health condition was estimated by skilled operators and vibration analysis. However, this diagnostic is not good enough to provide reliability of this machine cause of human’s wrong prediction. Therefore, this research has proposed a deep learning-based diagnostic system for motor bearing health condition monitoring. The convolutional neural network (CNN) was suggested as deep learning to provide our purpose of the study. CNN is quite a trend technology of artificial intelligence in the last two decades that is usually used for the classification of images and waveform signals. Our research consists of three main parts such as data collection (measurement), signal analysis and CNN training. First, accelerometers were mounted on the operating motor to measure the normal and failure vibration of the bearing, respectively. Second, collected data were used to determine normal and failure vibration features of bearing based on various kinds of analysis such as spectral and time-frequency. Third, collected data were used for training CNN model as well, which is used to classify input data. Finally, the trained CNN is used to early diagnose motor bearing health deterioration.