Abstract: To address the issues of insufficient information and reliability in single-channel monitoring of gear sensors,noise interference and data distribution differences under varying working conditions,a gearbox fault diagnosis method integrating an enhanced convolutional deep belief network with an adaptive weighted extension network was proposed.The collected multi-channel vibration data were reconstructed using the compressed sensing algorithm.Then,feature extraction was performed using a dilated convolutional deep belief network optimized with a soft pooling layer,and multi-channel features were weighted and fused using an attention mechanism.Next,the gearbox fault classification was completed using a weighted extension neural network optimized with side distance.Validation and comparative analysis were conducted using public datasets.The results showed that the proposed model had higher recognition accuracy compared to convolutional neural network models,deep belief network models and Gaussian convolutional deep belief network models,maintaining good fault diagnosis performance under noise interference and varying working conditions.

Key words: deep learning, convolutional deep belief network, extension neural network, fault diagnosis

摘要： 针对齿轮传感器在单通道状态监测中的信息量和可信度不足、噪声干扰及变工况下数据分布差异等问题,提出一种融合增强卷积深度置信网络与自适应加权可拓网络的齿轮箱故障诊断方法。采用压缩感知算法重构收集到的多通道振动数据;通过引入软池化层优化的膨胀卷积深度置信网络进行特征提取,并采用注意力机制技术加权融合多通道特征;利用侧距优化的加权可拓神经网络完成齿轮故障分类。最后,通过公开数据集进行验证和对比分析表明,该模型相比卷积神经网络模型、深度置信网络模型、高斯卷积深度置信网络模型等具有更高的识别精度,在噪声干扰和变工况条件下具有良好的故障诊断性能。

关键词: 深度学习, 卷积深度置信网络, 可拓神经网络, 故障诊断