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

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5 dB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。

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 proposed.This method preprocesses the original vibration signals of rolling bearings.Empirical mode decomposition,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 convolutional neural network for training and testing.All models are combined through Stacking to achieve classification of various types of rolling bearing faults.The results show that the diagnostic performance of Stacking model fusion with multi-input is better than that of traditional deep learning models.The diagnostic accuracy 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.