卷积神经网络
方位(导航)
计算机科学
断层(地质)
深度学习
人工智能
人工神经网络
核(代数)
循环神经网络
机器学习
模式识别(心理学)
数学
组合数学
地震学
地质学
作者
J. Shouhu,A. Yongwei,J. Guanwei,X. Weiqing,Weizhe Jia,C. Maolin
标识
DOI:10.1049/icp.2022.1749
摘要
Rolling bearings are the core components of rotating machines. Bearing faults affect economic benefits and safety. Therefore, bearing faults diagnosis is necessary for safe operations. Deep learning neural networks have unique advantages in fault diagnosis. Three traditional fault diagnosis algorithms have been employed with the CWRU FE (fan end) bearing dataset. Calculation accuracy, loss rate, training time of fault diagnosis have been demonstrated. Diagnostic accuracy has been achieved up to 91% with LSTM (long short-time memory network), but training time is too long to meet actual engineering. Meanwhile, above 91.9% and 99.9% diagnostic accuracy have been achieved by CCNN (cyclic convolutional neural network) and WDCNN (the first-layer large convolutional kernel convolutional neural network); The excellent training time is only about 30 s. The results verify that the WDCNN model has more effective fault diagnosis ability for bearing data sets.
科研通智能强力驱动
Strongly Powered by AbleSci AI