首页|期刊导航|机械与电子|基于SSA-VMD-CNN-BiLSTM的张力机轴承故障分级诊断

基于SSA-VMD-CNN-BiLSTM的张力机轴承故障分级诊断OA

Multi-level Fault Diagnosis of Tensioner Bearings Based on SSA-VMD-CNN-BiLSTM

中文摘要英文摘要

针对新能源张力机轴承在低速重载、频繁启停和张力波动工况下故障特征微弱、非平稳性强以及多级分类识别困难的问题,构建了一种融合麻雀搜索算法(SSA)优化变分模态分解(VMD)、卷积神经网络(CNN)与双向长短时记忆网络(BiLSTM)的 SSA-VMD-CNN-BiLSTM 轴承故障诊断模型.该模型以本征模态分量(IMF)排列熵之和作为适应度函数,引入麻雀搜索算法对 VMD 模态数 K 和惩罚因子α进行自适应优化,以获得时频聚集性和故障敏感性更优的本征模态分量.随后,将优化后的 IMF 分量输入CNN 提取局部深层特征,并通过 BiLSTM 对故障特征序列的双向时序相关性进行建模,实现新能源张力机发电机主轴轴承正常、内圈故障、外圈故障和滚动体故障等状态的分类识别.实验结果表明,排列熵作为SSA 寻优适应度函数时模型诊断性能较优,所提出的 SSA-VMD-CNN-BiLSTM 模型测试集分类准确率达到 98.33%,Macro-F1 值达到 98.31%,平均运行时间为 11.04 s,均优于 CNN-BiLSTM 和 SSA-VMD-CNN-LSTM 等对比模型.多指标评价、混淆矩阵、统计显著性检验和消融实验结果进一步表明,SSA-VMD、CNN 和 BiLSTM 各模块能够分别增强故障敏感分量表达、局部特征提取和双向时序建模能力.

To address the challenges of the weak fault signatures,strong non-stationarity,and difficul-ty in multi-class fault classification for bearings in new-energy tensioners operating under low-speed heavy-load,frequent start-stop,and tension fluctuation conditions,this paper proposes a hybrid fault di-agnosis model that integrates Sparrow Search Algorithm(SSA)-optimized Variational Mode Decomposi-tion(VMD),Convolutional Neural Network(CNN),and Bidirectional Long Short-Term Memory(BiL-STM),denoted as SSA-VMD-CNN-BiLSTM.In the proposed model,the sum of permutation entro-py of the intrinsic mode functions(IMFs)is adopted as the fitness function,and the SSA is introduced to adaptively optimize the number of decomposition modes K and penalty factor α in VMD,thereby yielding IMFs with superior time-frequency concentration and fault sensitivity.The optimized IMF components are then fed into a CNN to extract local deep features,followed by a BiLSTM network to model the bidirec-tional temporal dependencies of fault feature sequences,enabling the classification of four states of the gen-erator spindle bearing in new-energy tensioners:normal state,inner race fault,outer race fault,and rolling element fault.Experimental results show that using permutation entropy as the SSA fitness function yields the best diagnostic performance.The proposed model achieves a test-set classification accuracy of 98.33%,a Macro-F1 value of 98.31%,and an average running time of 11.04 s,outperforming comparison models,including the CNN-BiLSTM and SSA-VMD-CNN-LSTM.Multi-metricevaluation,confusion matrix analysis,statistical significance tests,and ablation experiments further demonstrate that the SSA-VMD,CNN,and BiLSTM modules respectively enhance fault-sensitive component representation,local feature extraction,and bidirectional temporal modeling capability.

王骅;周俊宏;李文泽;王海燕;吴伟智;杨先

广东电网有限责任公司惠州供电局,广东 惠州 516001广东电网有限责任公司惠州供电局,广东 惠州 516001广东电网有限责任公司惠州供电局,广东 惠州 516001广东电网有限责任公司惠州供电局,广东 惠州 516001广东电网能源发展有限公司,广东 广州 510160广东电网能源发展有限公司,广东 广州 510160

机械制造

张力机轴承故障诊断麻雀搜索算法双向长短期记忆网络

tensionerbearingfault diagnosissparrow search algorithmbidirectional long short-term memory network

《机械与电子》 2026 (8)

26-34,42,10

广东电网有限责任公司科技项目(031300KC23120022(GDKJXM20231447))

评论