Fault identification of piezoelectric sensor based on 1DCNN-BiLSTM-cross-attention
-
摘要: 针对宏纤维复合材料(macro fiber composite, MFC)压电传感器的故障辨识问题，提出一种基于一维卷积神经网络(one-dimensional convolutional neural network, 1DCNN)、双向长短期记忆(bidirectional long short-term memory, BiLSTM)网络融合交叉注意力机制(cross-attention)的故障辨识方法。实验模拟不同类型的传感器故障来获取故障数据，分别采用快速傅里叶变换和经粒子群算法优化参数的变分模态分解处理数据，获得时频域的混合数据，并用1DCNN-BiLSTM-cross-attention (1DCBCA)模型对数据进行故障辨识，准确率为99.4%。同其他对比模型相比，所提模型辨识准确率更优，能很好地适用于MFC压电传感器的故障辨识。Abstract: To address the fault identification problem of macro fiber composite (MFC) piezoelectric sensors, a fault identification method based on a one-dimensional convolutional neural network (1DCNN) and a bidirectional long short-term memory (BiLSTM) network integrated with a cross-attention mechanism was proposed. Different types of sensor faults were experimentally simulated to obtain fault data. The data were processed by fast Fourier transform and by variational mode decomposition with parameters optimized by a particle swarm optimization algorithm, respectively, and the resulting features were combined to obtain mixed time-frequency domain data. Subsequently, fault identification was performed on the data using the 1DCNN-BiLSTM-cross-attention (1DCBCA) model, and an accuracy of 99.4% was obtained. Compared with other models, the proposed model yields superior identification accuracy and is highly applicable to the fault identification of MFC piezoelectric sensors.
-
图 5 梁振动位移控制前后变化
Figure 5. Variation of beam vibration displacement before and after control
表 1 1DCNN网络结构
Table 1. 1DCNN network structure
网络层 网络结构 输入尺寸 输出尺寸 输入 — [32, 6, 1024 ]— 第1层 Conv1d + ReLU + Pooling [32, 6, 1024 ][32, 32, 512] 第2层 Conv1d + ReLU + Pooling [32, 32, 512] [32, 64, 256] 第3层 Conv1d + ReLU + Pooling [32, 64, 256] [32, 128, 128] 
下载: 导出CSV
表 2 BiLSTM网络结构
Table 2. BiLSTM network structure
网络层 操作 输入维度 输出维度 输入 划分堆叠，调换维度 [32, 6, 1024 ][32, 128, 48] BiLSTM第1层 32隐藏单元BiLSTM [32, 128, 48] [32, 128, 64] BiLSTM第2层 64隐藏单元BiLSTM [32, 128, 64] [32, 128, 128]
下载: 导出CSV
表 3 交叉注意力结构
Table 3. Cross-attention structure
网络层 输入数据来源 输入维度 Q 1DCNN输出 [32, 128, 128] K BiLSTM输出 [32, 128, 128] V BiLSTM输出 [32, 128, 128]
下载: 导出CSV
表 4 PSO-VMD参数结果
Table 4. Result of PSO- VMD parameters
参数 k a 9次数据 5；5；5；5；5；5；5；5；5 2108 ；2104 ；2120 ；2094 ；2105 ；2107 ；2103 ；2109 ；2108 平均值 5 2106
下载: 导出CSV
表 5 混淆矩阵结构
Table 5. Confusion matrix structure
混淆矩阵 预测正类 预测反类 真实正类 TP（真正例） FN（假反例） 真实反类 FP（假正例） TN（真反例）
下载: 导出CSV
表 6 各模型评价指标
Table 6. Evaluation metrics of different models
模型 准确率/% 精确率/% 召回率/% F1分数/% 1DCBCA(PSO-VMD) 98.1 98.2 98.1 98.1 1DCNN 83.6 85.2 83.6 83.6 BiLSTM 89.8 90.7 89.8 89.9 1DCNN-BiLSTM 96.5 96.6 96.5 96.5 1DCBCA(原始数据) 97.4 97.4 97.4 97.4 CWT-CNN 95.3 95.4 95.3 95.3 BiGRU 86.2 80.6 80.8 80.9 1DCBCA 99.4 99.5 99.5 99.5
下载: 导出CSV
-
[1] 唐令权. 复合材料板Lamb波结构响应多域特征融合损伤评估[D] . 南京: 南京邮电大学, 2023. [2] 童浩, 阮先虎, 林峰, 等. 基于可解释性深度学习的传感器异常数据诊断[J] . 传感器与微系统, 2024, 43(12): 41 − 44. [3] Seba A M, Gemeda K A, Ramulu P J. Prediction and classification of IoT sensor faults using hybrid deep learning model[J] . Discover Applied Sciences, 2024, 6(1): 9. doi: 10.1007/s42452-024-05633-7 [4] 孙渊, 曹俊杰, 唐矫燕, 等. 基于改进YOLOv8曲轴表面缺陷检测算法[J] . 组合机床与自动化加工技术, 2024(10): 77 − 81. [5] He B Y, Zhu C L, Li Z X, et al. A Bayesian cnn-based fusion framework of sensor fault diagnosis[J] . Measurement Science and Technology, 2024, 35(4): 046110. doi: 10.1088/1361-6501/ad1a86 [6] 赵小强, 张亚洲. 利用改进卷积神经网络的滚动轴承变工况故障诊断方法[J] . 西安交通大学学报, 2021, 55(12): 108 − 118. [7] Gao B W, Xu J, Zhang Z R, et al. Marine diesel engine piston ring fault diagnosis based on LSTM and improved beluga whale optimization[J] . Alexandria Engineering Journal, 2024, 109: 213 − 228. doi: 10.1016/j.aej.2024.08.075 [8] Guo M H, Xu T X, Liu J J, et al. Attention mechanisms in computer vision: A survey[J] . Computational Visual Media, 2022, 8(3): 331 − 368. doi: 10.1007/s41095-022-0271-y [9] Hao H T, Wang K. Fault detection and diagnosis of pneumatic control valve based on a hybrid deep learning model[J] . Instrumentation, 2023, 10(4): 12 − 26. [10] 邓丽, 邬群勇, 杨水荣. 融合SSAE深度特征学习和LSTM网络的PM2.5小时浓度预测[J] . 环境科学学报, 2020, 40(9): 3422 − 3434. [11] Yu X H, Yu S C, L X Q, et al. On normalized least mean square based interference cancellation algorithm for integrated sensing and communication systems[J] . ZTE Communications, 2024, 22(3): 21 − 28. [12] 朱少民, 夏虹, 尹文哲, 等. 基于变分模态分解和希尔伯特变换的转子非平稳信号故障特征识别[J] . 哈尔滨工程大学学报, 2024, 45(5): 825 − 832. [13] 张俊, 张建群, 钟敏, 等. 基于PSO-VMD-MCKD方法的风机轴承微弱故障诊断[J] . 振动、测试与诊断, 2020, 40(2): 287 − 296,418. [14] 刘光星, 马一豪. 基于LSTM-RF的电动钻机绞车齿轮箱故障诊断[J] . 振动与冲击, 2024, 43(21): 156 − 162,230. [15] 金江涛, 许子非, 李春, 等. 基于深度学习与混沌特征融合的滚动轴承故障诊断[J] . 控制理论与应用, 2022, 39(1): 109 − 116. -
下载:
点击查看大图
计量
- 文章访问数: 18
- HTML全文浏览量: 13
- PDF下载量: 1
- 被引次数: 0
