多深度学习模型决策融合的齿轮箱故障诊断分类方法

针对齿轮故障诊断中单一传感器采集信息不完全、容错性不佳及一种神经网络模型具有局限性，传统信号处理技术提取特征困难等问题，提出了多深度学习模型决策融合的齿轮箱故障诊断分类方法，构建了基于卷积神经网络(convolutional neural networks, CNN)和改进堆叠降噪自动编码器(stacked denoising autoencoders, SDAE)的混合网络模型，根据改进的Dempster-Shafer(D-S)证据理论实现决策级融合诊断。以时频信号作为CNN的输入，以频域信号作为SDAE的输入，采用Adam优化算法和dropout、批量归一化技术训练该混合模型。实验结果表明：利用该融合方法对齿轮进行故障诊断相比单个的网络模型CNN和SDAE诊断正确率有所提高，为齿轮故障智能诊断分类提供了新路径。

Gearbox fault diagnosis and classification method based on decision fusion of multi-deep learning models

Aiming at the problems of incomplete information collection by a single sensor, poor fault tolerance, the limitation of a neural network model, and the difficulty of extracting features with traditional signal processing technology in gear fault diagnosis, a gearbox fault diagnosis classification method based on multi-deep learning model decision fusion was proposed. A hybrid network model based on CNN (Convolutional Neural Networks) and improved SDAE (Stacked Denoising Autoencoders) is constructed, and the decision level fusion diagnosis is implemented according to the improved D-S evidence theory. Taking time-frequency signals as the input of CNN and frequent-domain signals as the input of SDAE, the mixed model was trained by using Adam optimization algorithm, dropout and batch normalization techniques. Experimental results show that the accuracy of gear fault diagnosis by using the fusion method is higher than that of single network models CNN and SDAE, which provides a new path for intelligent fault diagnosis and classification of gear.