SIA OpenIR  > 自动化系统研究室
汽车变速器新产品故障诊断机器学习方法研究
Alternative TitleResearch on Machine Learning for Mobile Transmission Products Fault Diagnosis
周晓锋1,2
Department自动化系统研究室
Thesis Advisor史海波
ClassificationTP18
Keyword故障诊断 变速器 阶次分析 遗传算法 支持向量机
Call NumberTP18/Z78/2012
Pages94页
Degree Discipline机械电子工程
Degree Name博士
2012-05-31
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract变速器是汽车的主要部件之一,其可靠性直接影响着汽车的整体性能指标。随着我国汽车工业的发展,为了提高生产效率,变速器新产品的故障检测工作应逐渐由自动化的计算机系统来代替传统的人工检测。因此,研发一种高效的变速器故障诊断系统迫在眉睫。该系统在提高检测效率的基础上,不但可以代替人工检测出有故障的不合格变速器产品,而且可以比人工检测更为精准的判断出故障原因,定位出故障部位,从而指导不合格变速器的维修工作并为变速器的质量跟踪和产品改进提供有力依据。 本文针对变速器新产品在线检测问题,主要进行了两个方面的研究工作:针对变速器故障诊断的理论研究和变速器故障诊断系统的设计开发验证工作。 针对变速器故障诊断的理论研究主要从以下三个方面展开: (1)变速器故障特征提取方法研究。首先分析了近期旋转机械故障诊断中几种常用的故障特征提取方法:频谱分析、小波分析、经验模式分解方法和阶次分析方法,论述了每种方法的优缺点。然后根据变速器新产品故障诊断的特点,提出了基于阶次分析的故障特征提取方法。通过等角度采样、分轴计算、角域同步平均计算、阶次谱计算,利用同步采集的转速信号将振动信号由时域信号转化为反应变速器阶次信息的阶次域信号。最后选择三个能敏感的反应故障信息的时域特征指标与阶次域特征向量集共同组成了变速器故障特征向量集,完成故障特征提取工作。 (2)变速器故障特征选择方法研究。由于特征选择后得到的特征向量集中特征数量过多,一方面会影响学习算法的效率,一方面由于冗余特征的存在会降低学习算法的准确率。因此本文提出了一种基于遗传搜索算法的特征选择策略。该方法以类内类间距离与惩罚系数的乘积做为特征评价准则,以改进的遗传算法作为搜索策略进行最优特征子集的搜索,最终得到了精简后的近似最优解。 (3)变速器故障分类方法研究。针对变速器故障分类判别中先验知识不完备、前期训练样本偏少,对实时性要求较高的特点,选取了支持向量机理论来解决变速器故障分类问题。在分析了支持向量机理论基础和原理之后,建立了二层多分类支持向量机结构模型,并针对正反例样本数量不平衡的特点,对支持向量机算法进行了改进,通过对惩罚因子的调整,解决了判别结果偏向于样本数多的类别的问题,得到了满意的判别结果。 针对变速器新产品故障诊断系统的分析设计和实施过程中遇到的问题,从系统实施方面对以下关键技术进行了研究: (1)数据同步采集。针对转速信号和振动信号的同步采集问题,提出了一种解决方案。该方案不同于以往的以振动信号采集卡作为主控卡的方式,从阶次分析计算的需要入手,以转速信号采集卡作为主控卡,根据变速器输入轴的旋转周期作为数据采集长度的依据,实现了两路信号的同步采集。 (2)振动传感器测点位置分析。通过对变速器壳体上各个采集位置采集到的信号进行分析和评价,从十几个备选位置中选择了一个最优位置作为传感器安装位置,保证了采集到的信号质量。 (3)通用的变速器结构建模和参数存储。为实现系统的通用性,设计了通用的变速器结构模型,可存储不同型号不同结构的变速器齿轮和轴承参数。设计并实现了参数管理系统用于存储变速器故障诊断系统运行所需要的各种计算和硬件参数,可满足变速器性能试验台上不同工况下的诊断工作。 (4)故障诊断系统与变速器性能试验台之间的通讯模块的设计。变速器故障诊断系统在运行中,需要实时接收变速器性能试验台发送的各种指令并作出应答,才能实现诊断过程的自动化。文本在分析了交互过程后,设计了通讯模块的工作流程,并完成了该模块的实施。 (5)给出了变速器故障诊断软件的整体设计概要并展示了各部分功能。
Other AbstractThe transmission is one of the main parts of cars, as its reliability may immediate impact the indexes of the overall performance of automobiles. Along with the development of our national automotive industry as well as to improve manufacturing efficiency, the fault detect for new transmissions should be replaced to robotized computer system gradually instead of artificial detection traditionally. So it is extremely urgent to research and develop a new kind of transmission fault diagnosis system with high efficiency. Under the basis of high detection efficiency, this system is not only to detect out the faulted or below standard transmission products that substitute for the manual labor, but also figure out the cause of the problems as well as locate the faults much more accurate than human beings, thus which can guide the maintenance work of the unqualified transmissions and provide strong evidence of quality tracking and products improving. According to the problems exist in the detection of the new transmission products online, this paper aimed at two sides to research which are theoretical studies of fault diagnosis for transmissions and design, exploitation and authentication for transmission fault diagnosis system The theoretical studies of fault diagnosis for transmissions are mainly from the following three aspects: (1) Research for feature extraction methods of transmission malfunction First of all, some commonly used methods of feature extraction of transmission malfunction which in recently selecting mechanical fault diagnosis are analyzed and expound the advantages and disadvantages of each method, namely, spectrum analysis, micro wave analysis, EMD, order analysis. Then according to features of new transmission products` fault diagnosis, we offered the methods of feature extraction of transmission malfunction on the basis of order analysis which by means of equal-angle sampling principle, split axle calculation, synchronous average angular domain calculation, sub-order spectrum calculation and make use of synchronous collection`s speed signal which will have vibration signals converted time signal to order time domain signals of reaction transmission order time information. Finally choose three possibilities of time domain feature index and order time domain feature vector sets which react stoppages sensitively to make up transmission fault characteristics vector sets jointly, then finish stoppages` feature extractions. (2) Research for feature selection methods of transmission malfunction Due to excessive number in the selected feature vector, on the one hand, it will influence the efficiency of the learning algorithm; on the other hand, the existence of redundant features can decrease the learning algorithm accuracy. Therefore this paper put forward a feature selection strategy based on the genetic algorithm which is used the product of inner and outer distance and penalty coefficient as assessment criteria and improved genetic algorithm as a search strategy to get an optimal feature subset of the search, finally get the approximate optimal solution after being cut. (3) Research for classification methods of transmission malfunction. According to the characteristics of transmissions` fault classification and discrimination which are insufficient prior knowledge, lack of early training samples and higher demand of instantaneity, we choose the theory of support vector machine to solve transmission fault classification problems. After the analysis of the theory, we built the multi-classes model structure of SVM and according to the number of positive and negative cases samples of the unbalance characteristics In light of the scientific analysis of fault diagnosis and problems in the course of implementation process for new transmission products, we present the research for the following key technique on system implementation. (1) Collection of data synchronism A solution was proposed for the synchronous acquisition of the speed signal and vibration signal. Which are different than they used to vibration signal acquisition card as a way of master card, starting from the order analysis and calculation of needs, this solution use the speed signal acquisition card as master. According to the transmission input shaft rotation cycle as the basis for data acquisition length, enables two-way synchronization of signal acquisition. (2) Analysis of vibration sensor`s point position After analysis and evaluate the signal collected on each collection location on the transmission from more than a dozen alternative location, select an optimal location as the sensor installation location to ensure the quality of the collected signal. (3) General transmission structure modeling and parameters save In order to realize the versatility of the system, the design of the general transmission structure model, that can store different models of different structure of transmission gear and bearing parameter. Design and implementation of management system for storing the parameters of transmission fault diagnosis system, which can meet the transmission performance test rig under different working condition diagnosis. (4) Design of communication module between fault diagnoses systems and test-bed of transmission performance When transmission fault diagnosis system is running, the system need to receive and answer a variety of instruction sent from the transmission performance test rig in order to achieve the automation of the diagnostic process. This paper designed the module of the work flow and completed the module implementation. (5) Providing Software`s holistic design outline of transmission fault diagnosis and showing its each part of functions
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/9325
Collection自动化系统研究室
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
周晓锋. 汽车变速器新产品故障诊断机器学习方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2012.
Files in This Item:
File Name/Size DocType Version Access License
汽车变速器新产品故障诊断机器学习方.pd(2007KB) 开放获取LicenseApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[周晓锋]'s Articles
Baidu academic
Similar articles in Baidu academic
[周晓锋]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[周晓锋]'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.