Document Type : Article
Authors
1
M.Sc. Student of Faculty of Mechanical Engineering, Tehran, Iran
2
Ph.D. Student of Faculty of Mechanical Engineering, Tehran, Iran
3
Assistant Prof., Faculty of Mechanical Engineering, Tehran, Iran
4
Department of Mechanical Engineering, Sharif University
10.24200/j40.2024.63396.1697
Abstract
Intelligent detection of rolling element bearing faults is a critical aspect of rotating equipment condition monitoring. Early detection of faults holds significant economic value for industrial units in terms of maintenance and planning. Traditional intelligent fault detection algorithms, which rely on a combination of feature extraction and signal classification, are time-consuming and require a high level of expertise to define appropriate inputs that are the most relevant to the desired output. In comparison to traditional methods, Convolutional Neural Networks (CNNs) can process a large volume of data with high accuracy and automatically extract features from vibration signals. Therefore, in this research, an attempt has been made to use a simple and shallow CNN to determine the health state of rolling element bearings and identify the defective element, which can be inner race, outer race, or rolling element bearing. For this purpose, a CNN model has been employed to investigate three common faults in rolling element bearings. In order to achieve the best performance, various inputs, including time waveform, frequency spectrum, and envelope spectrum, have been utilized and the results are compared to select the best and appropriate input. CNN requires a large amount of data to be trained. So, a laboratory setup has been designed and constructed to collect the required data for training the models and verifying them. After creating artificial faults on the bearings, experiments were conducted under 36 different operating conditions, comprising 9 different speeds, each at 4 different loads, encompassing four healthy states, including healthy, inner race fault, outer race fault, and rolling element fault. The obtained results have illustrated that the fault detector model with the frequency spectrum input is more accurate with an accuracy of 95% than the models receiving the other two inputs.
Keywords: Condition Monitoring, Intelligent Fault Diagnosis, Convolutional Neural Network, Rolling Element Bearing, Variable Operating Conditions, Vibration Analysis.
Keywords
Main Subjects
)