基于混合特征和堆栈稀疏自编码器的齿轮箱故障诊断

针对复杂工况下齿轮箱多故障信号诊断准确率低的问题,提出了一种基于混合特征和堆栈稀疏自编码器的齿轮箱故障诊断方法。从微观信号特征角度提取奇异值特征和小波分解后的样本熵特征;从宏观角度提取故障信号时域特征,将3种特征进行融合,并输入到由稀疏自编码和Softmax堆栈得到的深度神经网络中进行特征优化和分类识别。实验结果表明:在2种不同工况下,对6种齿轮箱故障数据进行诊断均表现出较高分类识别精度,且所构建的分类模型综合性能上均高于文中其他对比模型,因此本文作者所提出的方法能有效地进行齿轮箱故障诊断。

Fault Diagnosis of Gearbox Based on Hybrid Features and Stacked Sparse Auto-encoder

Aimed at the problem of low accuracy of multi-fault signal diagnosis of gearbox under complex working conditions, a gearbox fault diagnosis method based on hybrid features and stacked sparse auto-encoder was proposed. The singular value feature and the sample entropy feature decomposed by wavelet were extracted from the perspective of microscopic signal feature. The time-domain features of the fault signal were extracted from the macro perspective, and the three features were fused and input into the deep neural network obtained by sparse auto-encoder and Softmax stack for feature optimization and classification recognition. The experimental results show that the diagnosis of six gear box fault data shows high classification and recognition accuracy under two different working conditions, the comprehensive performance of the constructed classification model is higher than that of other comparison models, so the proposed method can effectively diagnose the gearbox fault.