现代防御技术 ›› 2018, Vol. 46 ›› Issue (5): 122-128.DOI: 10.3969/j.issn.1009-086x.2018.05.19

• 综合保障性技术 • 上一篇    下一篇

模拟电路故障的深度学习融合模型诊断方法

王应晨1, 段修生1,2, 单甘霖1   

  1. 1.陆军工程大学,河北 石家庄 050003;
    2.石家庄铁道大学,河北 石家庄 050003
  • 收稿日期:2018-01-23 修回日期:2018-02-14 出版日期:2018-10-30 发布日期:2020-11-25
  • 作者简介:王应晨(1995-),男,山西长治人。硕士生,主要研究方向为人工智能、装备测试与故障诊断。通信地址:050003 河北省石家庄市新华区和平西路97号二系火控教研室 E-mail:791191357@qq.com

Deep Learning Fusion Model Diagnosis Method of Analog Circuit Fault

WANG Ying-chen1, DUAN Xiu-sheng1,2, SHAN Gan-lin1   

  1. 1. Army Engineering University,Hebei Shijiazhuang 050003,China;
    2. Shijiazhuang Tiedao University,Hebei Shijiazhuang 050003,China
  • Received:2018-01-23 Revised:2018-02-14 Online:2018-10-30 Published:2020-11-25

摘要: 针对传统智能故障诊断方法因装备电路复杂和工作环境噪声等因素引起的诊断困难问题,提出了基于降噪自编码器和高斯深度信念网络的融合模型,来实现模拟电路的故障诊断。首先,降噪自编码器用于处理原始信号的噪声并学习低层特征;然后,深度信念网络基于所学习的低层特征来提取深层特征;最后将融合的深度特征融入softmax分类器中,对智能诊断模型进行训练。融合模型在模拟电路上进行了故障诊断实验,结果表明,所提方法具有优越的诊断性能。

关键词: 模拟电路, 故障诊断, 自适应学习率, 深度信念网络, 降噪自编码器, 鲁棒性

Abstract: Aiming at the difficulty of diagnosis caused by the complexity of equipment circuit, noise of working environment and big data of traditional intelligent fault diagnosis method, a fusion model based on noise reduction self-encoder and deep belief network is proposed to realize the fault diagnosis of analog circuit. The denoising autoencoder is used to process the random noise of the original signal and to learn the low-level features. The deep belief network extracts the deep features based on the learned low-level features. The fused depth features are incorporated into Softmax classifier for intelligent diagnosis model training. The new diagnostic model is validated on typical analog circuits. The results show that the proposed method has superior diagnostic performance.

Key words: analog circuit, fault diagnosis, adaptive learning rate, deep belief network, denoising autoencoder, robustness

中图分类号: