多输入Stacking模型融合滚动轴承故障诊断
Fault Diagnosis of Rolling Bearings Based on Stacking Model Fusion with Multi-Input
白健 1郝润芳 1程永强 2闫文恒 1徐博仁 1郭立旺1
作者信息
- 1. 太原理工大学电子信息与光学工程学院,晋中 030600
- 2. 太原理工大学电子信息与光学工程学院,晋中 030600;山西省能源互联网研究院,太原 030006
- 折叠
摘要
针对现有单输入模型抗噪声能力不强、泛化能力不足的问题,提出了一种基于多输入Stacking模型融合的滚动轴承故障诊断方法.该方法对滚动轴承的原始振动信号进行预处理,分别对原始信号进行经验模态分解、变分模态分解和多分辨率分析,将3 种预处理后的信号输入到改进的卷积神经网络和改进的双输入卷积神经网络中进行训练及测试;各模型通过Stacking方法进行融合,以实现滚动轴承各种类型故障的诊断.结果表明,多输入Stacking模型融合方法的诊断性能优于传统的深度学习模型,在信噪比为5dB的条件下达到了98.9%的诊断准确率.该模型的故障诊断性能稳定,具有很好的鲁棒性和泛化能力.
Abstract
The existing single-input model is weak in anti-noise and generalization.To solve the above problems,a rolling bearing fault diagnosis method based on Stacking model fusion with multi-input is pro-posed.This method preprocesses the original vibration signals of rolling bearings.Empirical mode decompo-sition,variational mode decomposition and multi-resolution analysis are applied to the signal.Input the three preprocessed signals into the improved convolutional neural network and the improved dual-input convolu-tional neural network for training and testing.All models are combined through Stacking to achieve classifi-cation of various types of rolling bearing faults.The results show that the diagnostic performance of Stac-king model fusion with multi-input is better than that of traditional deep learning models.The diagnostic ac-curacy of 98.9%was achieved under the condition of a signal-to-noise ratio of 5 dB.The fault diagnosis performance of this model is stable,with good robustness and generalization ability.
关键词
故障诊断/模型融合/深度学习/滚动轴承Key words
fault diagnosis/model fusion/deep learning/rolling bearing引用本文复制引用
基金项目
山西省重点研发计划(20220701192135)
出版年
2024
NETL
NSTL
万方数据