摘要
由于电机结构及其运行环境复杂,导致各类故障与故障特征存在较强的非线性关系,单一信号信息含量有限,无法满足诊断需求。针对此问题,以电流、磁场信号为监测信号,提出基于注意力机制改进的支持向量机-自适应提升算法(SVM-AdaBoost,简称SAB)的故障诊断方法。首先,通过希尔伯特变换和快速傅里叶变换提取信号频域特征;其次,通过SAB分类器,对多源样本分别进行训练,获取各子分类器预测结果;最后,基于注意力机制调整权重矩阵参数,对电流、电磁信号进行信息融合,改进SAB分类器以提高故障诊断的准确率。研究结果表明:不同信号对各类故障的敏感程度不同;所提方法可以实现对转子断条故障、定子短路故障、轴承故障的诊断分类,与传统方法对比,该方法明显提高了故障诊断的鲁棒性和准确性。
Due to the complex motor structure and its operating environment,there is a strong nonlinear rela-tionship between different faults and their fault features,the motor diagnosis driven by the single signal cannot satisfy the requirements.Considering this dilemma,a motor fault diagnosis method driven by multi-sensor infor-mation based on attention mechanism and SVM-AdaBoost(SAB)is proposed in this study.First,the corre-sponding frequency features of current and magnet signals are extracted by Hilbert transform and Fourier trans-form.Then,the SAB classifier is utilized to train multi-source signals and obtain sub-classifier results respective-ly.Finally,based on attention machanism,an attention weight matrix is adjusted to fuse the information and cal-culate the final diagnosis results.The proposed method is verified by realizing the diagnosis of broken rotor bar fault,stator short circuit fault and bearing fault.The proposed method reveals that the sensitivity of different sig-nals to various faults is different.Compared with the traditional methods,the proposed method examine its supe-riority in terms of the robustness,generalization ability and fault diagnosis accuracy.
作者
令波
杨咏
穆旻皓
张文俊
LING Bo;YANG Yong;MU Minhao;ZHANG Wenjun(Wuhan Second Ship Design and Research Institute Wuhan,430074,China;No.704 Research Institute of CSSC Shanghai,200000,China)
出处
《振动．测试与诊断》
EI
CSCD
北大核心
2023年第3期603-608,626,共7页
Journal of Vibration,Measurement & Diagnosis
