Machine learning techniques have been successfully applied for the intelligent fault diagnosis of rolling bearings in recent years. This study has developed an Improved Multi-Scale Convolutional Neural Network integrated with a Feature Attention mechanism (IMS-FACNN) model to address the poor performance of traditional CNN-based models under unsteady and complex working environments. The proposed IMS-FACNN has a good extrapolation performance because of the novel IMS coarse grained procedure with training interference and the introduced the feature attention mechanism, which improves the model’s generalization ability. The proposed IMS-FACNN model has a better performance than existing methods in all the examined scenarios including diagnosing the bearing fault of a real wind turbine. The results show that the reliability and superiority of the IMS-FACNN model in diagnosing faults of rolling bearings.

近年来，机器学习技术已成功应用于滚动轴承的智能故障诊断。这项研究开发了一种改进的多尺度卷积神经网络，该网络与特征注意机制（IMS-FACNN）模型集成在一起，以解决在不稳定和复杂的工作环境下基于传统CNN的模型的不良性能。提出的IMS-FACNN具有新颖的具有训练干扰的粗粒度过程，并引入了特征关注机制，提高了模型的泛化能力，具有良好的外推性能。所提出的IMS-FACNN模型在包括检查实际风力涡轮机的轴承故障在内的所有检查场景中均具有比现有方法更好的性能。