当前位置：首页 > 过刊浏览->2023年45卷2期
基于卷积神经网络的智能船舶组合导航系统故障检测算法
An fault detection algorithm for intelligent ship integrated navigation system
- DOI:
- 作者:
- 刘军坡, 吴兆麟, 曹玉墀, 何庆华, 郭沐壮
LIU Jun-po, WU Zhao-lin, CAO Yu-chi, HE Qing-hua, GUO Mu-zhuang
- 作者单位:
- 大连海事大学 航海学院，辽宁 大连 116000
School of navigation, Dalian Maritime University, Dalian 116000, China
- 关键词:
- 智能船舶;组合导航系统;故障检测
intelligent ship; integrated navigation system; fault detection
- 摘要:
- 子滤波器中存在的故障会影响组合导航系统的性能，甚至会污染其他子导航系统。针对现有基于残差观测器故障检测算法的检测阈值、观测器参数以及滑动窗口宽度等依赖经验，不够智能，且对软故障不敏感的问题，本文提出一种基于残差频域信息的卷积故障检测网络，通过小波变换对${\ \chi ^2}$检测的残差信号进行预处理，而后通过短时傅里叶变换获取频域故障信息特征，将其代入卷积神经网络中进行训练。实验结果表明，该算法不仅能够在数据量有限的前提下检测出渐变软故障，且更加智能化，故障检测准确率提高。
The faults in the sub filter will affect the overall performance of the integrated navigation system, and even pollute other sub-systems. To solve the problem that the existing fault detection algorithm based on residual observer rely on experience, such as detection threshold, observer parameters and sliding window. Moreover, it is not intelligent enough and is not sensitive to soft fault. In this paper, a convolution fault detection network based on residual frequency domain information is proposed. The detected ${\ \chi ^2}$ residual signal is processed by wavelet transform, and then the frequency domain fault information features are obtained by short-time Fourier transform, which are substituted into convolutional neural network for training. Experiment results show that the algorithm can not only detect soft faults with limited amount of data, but also be more intelligent and improve the fault accuracy.
2023,45(2): 155-158 收稿日期：2022-03-08
DOI：10.3404/j.issn.1672-7649.2023.02.027
分类号：U675.7
作者简介：刘军坡(1982-),男,硕士,副教授,研究方向为船舶自动避碰、船舶安全理论与技术、船舶运动控制
