摘要
为实现传感器故障的实时检测和准确诊断,提出了基于注意力卷积自编码器(Attention-based Convolutional Auto-Encoder,ACAE)的传感器故障自动检测和诊断方案。建立6种典型传感器故障的数学模型,并由此提出故障检测和诊断模型。利用线性SVM分类器进行实时故障检测,区分正常数据和故障数据。检测到故障后,利用基于ACAE的故障分类模块进行准确故障诊断。ACAE通过2个并行特征提取器改善高区分性特征的提取性能,并采用不对称结构提高了模型效率。实验结果表明,所提方案能够完成故障快速检测和准确分类,性能优于其他深度学习方法,且能够满足嵌入式设备的资源约束,有助于提高物联网和信息物理系统的稳定性。
In order to realize the real-time detection and accurate diagnosis of sensor faults,a sensor fault automatic detection and diagnosis scheme based on Attention-based Convolutional Auto-Encoder(ACAE)is proposed.Firstly,the mathematical models of six typical sensor faults are established,and the fault detection and diagnosis models are proposed accordingly.An efficient linear SVM classifier is used for real-time fault detection to distinguish between normal data and faulty data.After a fault is detected,the ACAE-based fault classification module is used for accurate fault diagnosis.In the ACAE,two parallel feature extractors are used to improve the extraction performance of highly discriminative features,and an asymmetric structure is adopted to improve the model efficiency.The experimental results show that the proposed scheme can realize the rapid detection and accurate classification of faults,the fault classification performance is better than deep learning methods,and the resource constraints of embedded devices are well satisfied,which is helpful to improve the stability of the Internet of Things and Cyber-Physical Systems.
作者
许绘香
苏玉
岳媛
XU Huixiang;SU Yu;YUE Yuan(College of Information Engineering,Zhengzhou University of Technology,Zhengzhou 450044,China)
出处
《无线电工程》
北大核心
2023年第8期1965-1973,共9页
Radio Engineering
基金
河南省科技攻关计划项目(212102110168)
河南省大学生创新创业训练计划项目(S202111068008)
河南省教育科学“十三五”规划课题(2020YB0289)。
关键词
支持向量机
卷积自编码器
故障检测
故障诊断
物联网
嵌入式设备
support vector machine
convolutional autoencoder
fault detection
fault diagnosis
Internet of Things
embedded device
