由于舰船电机轴承信号类型较多，且大多数为无用信号，降低了故障信号分量之间的相关性，因此研究改进MSEA-CNN的船舶电机轴承故障诊断方法。利用自适应线性神经网络，设计移除非轴承故障分量滤波器，在轴承振动信号内剔除非故障信号分量，提取轴承故障信号分量。通过生成式对抗神经网络，得到各故障信号分量的样本标签，实现故障信号分量分类。融合样本标签与故障信号分量，获取故障诊断训练集。利用注意力与多尺度卷积神经网络，建立故障诊断模型。实验证明：该方法可有效提取故障信号分量；该方法提取故障信号分量特征间的相关系数较低，说明该方法具备较优的特征提取效果。
Because there are many types of ship motor bearing signals, and most of them are useless signals, which reduces the correlation between fault signal components, the fault diagnosis method of Ship Motor Bearing Based on msea-cnn is studied and improved. Using adaptive linear neural network, a filter for removing non bearing fault components is designed, and the fault signal components are removed from the bearing vibration signal to extract the bearing fault signal components. Through the generative countermeasure neural network, the sample labels of each fault signal component are obtained to realize the classification of fault signal components. The sample label and fault signal component are fused to obtain the fault diagnosis training set. A fault diagnosis model is established by using attention and multi-scale convolution neural network. Experiments show that this method can effectively extract fault signal components. The correlation coefficient between fault signal component features extracted by this method is low, which shows that this method has better feature extraction effect.
2022,44(14): 119-122 收稿日期：2022-01-07
DOI：10.3404/j.issn.1672-7649.2022.14.025
分类号：TM307.1
作者简介：李忠(1970-)，男，硕士，高级工程师，研究方向为交通运输工程及航标管理
参考文献：
[1] 张仲懿, 姚玉南, 梁言. 基于贝叶斯网络的船舶发电机运行过热故障分析[J]. 武汉理工大学学报(交通科学与工程版), 2019, 43(2): 311–315
[2] 杨晓珍, 周继续, 邓举明. 基于改进贝叶斯分类的电机轴承故障诊断系统研究[J]. 机床与液压, 2020, 48(20): 172–175
[3] 姜磊, 朱明斯, 李飞龙. 基于AMD和小波阈值去噪的电机轴承故障诊断[J]. 机械设计与制造, 2021(9): 47–50
[4] 张习习, 顾幸生. 基于集成学习概率神经网络的电机轴承故障诊断[J]. 华东理工大学学报(自然科学版), 2020, 46(1): 68–76
