基于双重注意力机制与迁移学习的风力发电机行星齿轮箱故障诊断

Fault Diagnosis of Planetary Gearbox of Wind Turbine Based on Double Attention Mechanism and Transfer Learning

针对风力发电机行星齿轮箱故障数据稀缺、难以提取、故障识别准确率较低的问题,提出一种双重注意力机制与迁移学习相结合的故障诊断方法。首先,将行星齿轮箱原始振动数据进行归一化后输入卷积神经网络中提取特征;然后将特征图分别输入到位置注意力机制和通道注意力机制中提取高级特征;最后进行特征融合、输出诊断结果。在变工况迁移时,将源域模型通过参数迁移到目标域工况后进行微调并输出预测类别。试验结果表明所提方法迁移后的故障识别准确率在98%以上,相比于支持向量机(support vector machine,SVM)、极限梯度提升(extreme gradient boosting,XGBoost)等其他模型有大幅度提高。

In order to solve the problem that the fault data of wind turbine planetary gearbox is scarce and difficult to extract,which leads to the low accuracy of the final fault identification,a fault diagnosis method combining dual attention mechanism and transfer learning is proposed.Firstly,the original vibration data of the planetary gearbox are normalized and input into the convolutional neural network to extract the features.Then the feature maps are input into the location attention mechanism and channel attention mechanism respectively to extract advanced features.Finally,feature fusion is performed to output diagnostic results.In the case of variable working condition migration,the source domain model is fine-tuned and the prediction categories are output after the parameter migration to the target domain condition.The experimental results show that the fault identification accuracy of the proposed method after migration is above 98%,which is significantly improved compared with other models such as support vector machine(SVM)and extreme gradient boosting(XGBoost).