SIA OpenIR  > 数字工厂研究室
基于数据驱动的自主式水下潜水器故障检测方法研究与应用
Alternative TitleResearch and Application of Fault Detection Method for Autonomous Underwater Vehicle Based on Data Drive
杨宗圣
Department数字工厂研究室
Thesis Advisor郭大权
Keyword自主式水下潜水器 故障检测 主元分析 注意力机制 大数据
Pages83页
Degree Discipline控制理论与控制工程
Degree Name硕士
2020-05-26
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract海洋资源丰富,我国对海洋资源的开发日益迫切,而自主式水下潜水器(Autonomous Underwater Vehicle,AUV)作为进入海洋进行科学研究的智能化装备装备,也越来越引起研究人员的关注。潜水器每次出海将会花费巨大的人力、物力和财力,如果发生故障将会造成巨大的损失:一方面,故障数据使采集的环境数据失去分析意义,航次失效造成巨大经济损失;另一方面,潜水器中的隐含故障难以发现,造成潜水器损坏甚至丢失。本文首先介绍了潜水器的结构,并且对国内外的研究现状进行了整理,并总结了当前研究存在的问题和不足。然后,本文以“潜龙二号”AUV的资源勘测系统和运动控制系统为研究对象,并依托该潜水器的实际海洋航行数据为训练样本,开展了AUV故障检测方法研究。主要工作有:(1)针对“潜龙二号”AUV 的资源勘查系统传感器数据有着多重变量相关性、故障类型多样、受运行状态和环境变化影响数值变化大以及噪声强等问题, 提出一种新的基于多块信息提取的主元分析 (PCA) 故障检测方法.首先,针对变量之间的多重相关性,通过滑窗和相关系数的方法提取变量间相关性信息;然后,根据变化率在不同运行状态和环境下基本稳定的特点,对于不同类型故障,分别提取变化率信息和变化率信息的各阶统计量累积误差信息;最后,基于提取的特征信息建立3 个子块,对每个子块分别建立PCA 模型并进行检测,将检测的结果通过中值滤波去噪后,用贝叶斯推断进行融合.通过对“潜龙二号”实际运行数据进行检测,验证了该方法的有效性。(2)针对“潜龙二号”AUV的运动控制系统数据有着时空耦合、地理关联、多模态和混合特性共存等问题,利用神经网络提取高维信息,提出一种基于注意力机制的时间卷积网络(Temporal Convolutional Networks,TCN)。一方面,将解决文本分析问题的TCN扩展到潜水器故障检测领域,充分利用时序数据的历史信息,提取数据特征;另一方面,为了提高模型准确度,降低冗余信息影响,通过注意力机制来计算每个特征的权重。最后,为了对潜水器数据进行分析,开发了潜水器数据辅助分析软件,提高分析效率。但是随着潜水器潜次的增加,采集到的潜水器数据越来越多,传统的数据分析软件难以对这些数据进行全面的分析,并且故障检测与故障诊断算法的运行时间随着数据量增加而增加,难以满足潜水器研究人员的效率要求。因此,对基于大数据技术潜水器数据分析平台进行了架构和界面设计,不仅可以全面的集成数据展示、数据分析、故障检测和诊断等功能,并且能够帮助潜水器研究人员对潜水器数据进行高效的分析,而且对平台的可拖拽式算法库进行了实现,为平台的实现奠定了基础。
Other AbstractWith abundant Marine resources, the development of Marine resources in China is increasingly urgent. Autonomous underwater vehicle (AUV), as an intelligent equipment for scientific research into the ocean, has attracted more and more attention of researchers. Every time a submersible goes out to sea, it will cost huge manpower, material resources and financial resources. If it fails, it will cause huge losses. On the other hand, it is difficult to find the hidden faults in the submersible, which causes the submersible to be damaged or even lost. This paper first introduces the structure of the submersible, and summarizes the research status at home and abroad, and summarizes the problems and shortcomings of the current research. Then, taking the resource survey system and motion control system of "Qianlong 2" AUV as the research object, and relying on the actual sea navigation data of the submersible as the training samples, this paper studies the AUV fault detection method. The main work includes: (1) Aiming at the problems of AUV of "Qianlong 2" in the actual navigation process, such as multivariable correlation, multiple fault types, large numerical variation influenced by operation status and environmental changes, and strong noise, a new PCA fault detection method based on block information extraction is proposed. Firstly, according to the multiple correlations among variables, the correlation information between variables is extracted by sliding window and correlation coefficient method; secondly, according to the basic stable characteristics of change rate in different operating states and environments, for different types of faults, the cumulative error information of each order statistics of change rate information and change rate information is extracted separately; Finally, three sub-blocks are built based on the extracted feature information, and PCA model is built and tested for each sub-block respectively. After de-noising by median filtering, the detected results are fused by bayesianinference. Thevalidityofthemethodisverifiedbytestingtheactualoperationdataof "Qianlong2". (2)Aiming at the problems of spatiotemporal coupling, geographical correlation, multimodality and hybrid characteristics of the motion control system data of qianlong-2 AUV, a temporal convolutional network based on attention mechanism is proposed by using neural network to extract high-dimensional information Network, TCN). On the one hand, TCN which solves the problem of text analysis is extended to the field of submarine fault detection, making full use of the historical information of time series data to extract data features; on the other hand, in order to improve the accuracy of the model and reduce the impact of redundant information, the weight of each feature is calculated by attention machine. Finally, in order to analyze the data of the submersible, the auxiliary analysis software of the submersible data was developed to improve the analysis efficiency. However, with the increase of submersible dives, more and more submersible data are collected, so it is difficult for traditional data analysis software to comprehensively analyze these data, and the running time of fault detection and fault diagnosis algorithm increases with the increase of data, which is difficult to meet the efficiency requirements of submersible researchers. Therefore, the submersible data analysis platform based on big data technology architecture and interface design, can not only fully integrated data display, data analysis, fault detection and diagnosis, and other functions, and can help researchers submersible to the analysis of the submersible data efficiently, and the platform can drag-and-drop algorithms library for implementation, has paved the way for the platform implementation.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/27122
Collection数字工厂研究室
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
杨宗圣. 基于数据驱动的自主式水下潜水器故障检测方法研究与应用[D]. 沈阳. 中国科学院沈阳自动化研究所,2020.
Files in This Item:
File Name/Size DocType Version Access License
基于数据驱动的自主式水下潜水器故障检测方(4581KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[杨宗圣]'s Articles
Baidu academic
Similar articles in Baidu academic
[杨宗圣]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[杨宗圣]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.