SIA OpenIR  > 工业信息学研究室
基于符号时间序列分析技术的电机故障诊断方法研究
Alternative TitleResearch of Faults Diagnosis for Induction Motor Based on the Technology of Simbolic Time Series Analysis
胡为1,2
Department工业信息学研究室
Thesis Advisor胡静涛
ClassificationTM38
Keyword符号时间序列分析 故障诊断 相对熵 感应电机 Hmm
Call NumberTM38/H53/2009
Pages118页
Degree Discipline机械电子工程
Degree Name博士
2009-01-19
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract随着经济建设的快速发展和电气化程度的不断提高,电机已被广泛应用于工业、农业、国防及人们日常生活的各个领域。从全球范围看,电机的用电量平均占世界用电总量的50%以上、占工业用电量的70%左右,然而在电机消耗的电能中有相当一部分被浪费掉了,其中电机带故障运行是造成电机运行效率偏低,能源浪费严重的主要原因之一。电机在线监测及故障诊断系统对于减少由于电机故障引发的人员、财产的损失,减少由于故障引发的异常状态而导致的能源浪费有着重要的现实意义。在电机故障危害产生前发现故障并进行维护是电机故障诊断的核心思想,在保证电机故障诊断系统准确性的同时,系统的快速性与鲁棒性显得尤为重要。基于此,本论文从寻求系统的快速、稳定的性能入手,提出了基于符号时间序列分析的感应电机故障诊断框架,重点研究了计算代价小、噪声干扰不敏感的诊断方法,以期提高感应电机故障诊断系统的快速性与鲁棒性。论文的主要工作有: 1. 论文首先构建了一个基于符号时间序列分析的电机故障诊断框架,将电机故障诊断分解为信号预处理、符号区间划分、符号统计分析三部分,有机地融合了统计分析、信号处理、信息论、模式识别等理论和方法,利用符号时间序列分析技术在强噪声中准确识别系统状态模式的良好性能,可以有效地解决电机故障诊断问题,并实现电机故障诊断量化分析,是对探索电机在线监测与诊断新方法的一次有益的尝试。 2.引入提升小波对信号进行前期处理,并针对常规提升小波固定预测滤波器的局限性,提出了基于梯度信息的自适应提升小波预测方法。该方法中预测滤波器并不是固定的,而是利用梯度的信息来确定预测算子。根据信号的陡峭程度选择预测算子可以更准确地预测信号,从而使原始信号中的平滑特征和陡峭特征可以在小波变换中完好地保留下来。仿真实验及实验室实验结果表明该方法可以有效地保留信号中蕴含的重要的特征信息,对于以提取、识别信号中特征信息为主的故障诊断技术来说具有非常重要的意义。 3.针对所采集现场信号的非均匀分布特点,论文提出了一种自适应符号化划分方法,既可以确保符号在数据密集区间和数据稀疏区间的合理分配,提高符号的利用率,又可以灵活地适应信号的特征,增强诊断系统对微弱故障信号的敏感度。故障诊断实验表明该方法简单有效,实现了故障初期的快速诊断,并且较平均区间划分方法有着更高的计算效率、更明显的诊断效果。 4.将相对熵的概念引入基于符号时间序列分析的电机故障诊断框架中,针对电机故障严重程度量化分析问题,提出了基于模糊相对熵及加权模糊相对熵的符号统计分析方法,并将该方法应用于感应电机的故障诊断与识别,建立了电机故障诊断模型。该方法可以更合理、充分地利用信息丰富的符号区间所蕴含的故障信息,实现了电机故障诊断与故障严重程度的识别。实验结果验证了该方法的合理性、有效性和可靠性。 5.将隐马尔可夫模型(HMM)引入到基于符号时间序列分析的电机故障诊断框架中,构造了基于HMM的电机故障诊断模型,并对HMM阶数选取问题给出了一个基于符号出现不确定信息熵的HMM阶数选取原则,使得模型在满足精度要求的同时,又尽可能地减少模型的计算代价,有效地提高了故障诊断的效率及可靠性。实验结果表明基于HMM的电机故障诊断方法有效地实现了电机转子断条故障、匝间短路故障的诊断与量化分析。
Other AbstractElectric machines have been widely applied to industry, agriculture, national defence, and all kinds of fields of human daily life accompanying that the economic construction rapid progresses and the degree of electrization enhances. From the world, over 50% the electric energy in the world are consumed by electric machines, which means that they consume 70% the industry electric energy or so. However quit a few of the consumed electric power are wasted, and one of the main reasons is that motors operate in a low efficiency owing to motor faults. The online mornitoring and faults diagnosing system for electric machine is important and significant for reducing the loss of human and property resulting from motor fauts. The core idea is that to find faults and maintain motor ahead of schedule before the faults happened. Besides the accuracy of the faults diagnosis system for motor, the rapidity and robustness of the system are also very important. This dissertation proposes the motor faults diagnosis and recogonition system based on the technology of symbolic time series analysis in order to improve the rapidity and robustness of the system. The dissertation primarily researchs the diagnosis algorithm of less calculating cost and insensitivity to noise, which improves the rapidity and robustness of the system, and relealized the faults diagnosis for motors. What the dissertation main contributes for are summarized as follows: 1. In this study, a frame of motor faults diagnosis based on symbolic time series analysis is first proposed, the frame decomposes the fault diagnosis system into three parts which is signal preprocessing, symbols region partitioning and the statictic analysis of symbols. The frame merges some theories and methods such as statistic analysis, signal processing, information theory, and pattern recognizition. Symbolic time series analysis has a good performance of recognizing the partten of system among the strong noise, also it can solve the problems of motor faults diagnosis effectively and realize the quantity analysis for motor faults diagnosis. Symbolic time series analysis is a significant try for the theory of motor monitoring online. 2. Lifting wavelet is introduced into the frame to process the signal. In order to improve the limitation of fixed predicting filter of routine lifting wavelet, the dissertation proposes a new adaptive lifting wavelet based on gradient information. In the method the predicting filter is not fixed, however, it selects the predicting operators according to the gradient information, and this method can predict the signal more accurately. Thus the satisfied smooth properties and sharp transition characteristics of the original signal can be reserved perfectly during the wavelet transforming. The results of simulation tests and laboratory tests show that the important information can be reserved effectively with this method, and the method is very significant for the fault diagnosis fields which depend on mining the useful information in the signal. 3. For the industrial signal are almost nonhomogeneous, this dissertation presents a adaptive partitioning method. The method ensures that the symbols can assigned reasonably in both more information regions and sparse information regions. The method increases the use ratio of symbols and neatly adapts it to the characters of signal, also it makes the system increase the sensitivity to weak signal. The tests of motor faults diagnosis show that the method effectively realize the quick diagnosis for motor faults, and the method has a higher calculating efficiency and the more obvious diagnosis results than the uniform partitioning method. 4. The concept of relative entropy is introduced into the frame for the statistic analysis of symbols, and the symbols statistic analysis methods based on fuzzy relative entropy and weighted fuzzy relative entropy are presented for dealing with the problem of multi character vectors composed by the probabilities of various symbols. The methods are applied to diagnose and recognize the faults for motor, also a model of motor faults diagnosis is constructed. The method based on weighted fuzzy relative entropy can realize that recognize and classify the serious degree of faults using the fault information from the regions of rich information more reasonably and more adequately, and enhance the reliability and accuracy of classifying. The results of tests verify the reasonableness, effectiveness and reliability. 5. Introduces HMM to the frame of motor faults diagnosis based on symbolic time series analysis, and constructs the model of motor faults diagnosis based on HMM. An order selective rule is proposed for solving the problem that how to select the order of HMM. The rule satisfies the accuracy of HMM meanwhile reduces the calculating cost of HMM as possible as it could, and the rule enhances the operating efficiency of HMM and improves the reliability of HMM. The results of laboratory tests show that the method of motor faults diagnosis based symbolic time series analysis realizes that quantificationaly diagnose and recognize broken bar fault and short circuit fault of motor effectively.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/386
Collection工业信息学研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院研究生院
Recommended Citation
GB/T 7714
胡为. 基于符号时间序列分析技术的电机故障诊断方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2009.
Files in This Item:
File Name/Size DocType Version Access License
10001_20051801471900(9604KB) 开放获取--Application Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[胡为]'s Articles
Baidu academic
Similar articles in Baidu academic
[胡为]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[胡为]'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.