基于粗糙集理论的齿轮箱故障诊断研究

发布时间：2018-05-14 02:25

本文选题：局域波分解 + 粗糙集　；参考：《中北大学》2013年博士论文

【摘要】：齿轮箱是机械系统的重要传动部件,故障发生率较高,其振动信号呈现非线性非平稳的特点,故障程度、部位和类型等对特征参量的影响很大。在对齿轮箱进行监测与故障诊断时,若监测点选择不当就不能采集到有效的故障信息,从而导致故障发生部位不易确定,敏感特征参量提取困难、故障模式识别率低。局域波分解方法将非平稳时变信号自适应地分解展开并映射到时频分析平面,能够同时展示信号的时域和频域信息的全貌；粗糙集理论的属性约简技术能够优化故障特征参量集,提取出敏感的故障特征参量；最小二乘支持向量机的函数逼近效果良好,模式识别能力强,本文在采用局域波分解法处理故障信号以及深入研究粗糙集理论的基础上,将粗糙集与最小二乘支持向量机相结合,建立了基于粗糙集支持向量机的齿轮箱智能故障诊断系统。本文的主要研究内容与结论如下： (1)在分析齿轮箱振动特性的基础上,提出采用局域波分解技术对齿轮箱故障信号进行处理并提取了初始的故障特征参量集。在局域波分解过程中,采用镜像延拓与窗函数相结合的方法缓解了端点效应问题,采用总体经验模态分解方法有效解决了模态混叠问题,实验结果表明这两种方法在齿轮箱故障信号分解中取得了较好的效果。根据衡量故障特征参量集的指标,提出采用每个工况的均方根有效值衡量故障特征参量集的稳定性,采用每个特征参量在六个工况之间的最小均值差衡量故障特征参量集的敏感性。实验中分别提取了基于EEMD的归一化能量特征参量集和基于EEMD的近似熵特征参量集,通过实际计算结果表明,前者与后者的敏感性基本一致,但是稳定性要优于后者,因此本文采用了基于EEMD的归一化能量特征参量集进行齿轮箱故障诊断。 (2)提出一种基于改进Naive Scaler算法的全局动态寻优离散化算法。通过对Naive Scaler算法过程进行改进,确保能够得到所有保证不可分辨关系的断点；通过断点均分样本集、逐渐增加断点的方法动态地从候选集中选择断点集,保证了整个信息系统分类能力不变的条件下断点个数最少。通过与其它算法对比,实验结果表明该算法得到的断点个数较少,体现了其在连续属性离散化方面的优越性。 (3)提出一种基于条件等价类的属性约简算法。该算法在核属性集的基础上,直接针对核属性的条件类中不能正确划入决策类的类,在核属性之外的其余条件属性中找到能够区分该类的属性,并添加到核属性集中,从而得到最小属性约简集。而基于启发式信息的属性约简算法无法保证所求约简集一定是最小属性约简集,实验结果表明该算法计算复杂度较低,提高了约简效率。 (4)提出采用粗糙集的属性约简技术对故障监测点进行优化配置。该方法将六个故障监测点的最小属性约简集融合成一个大决策表进行属性约简,根据每个监测点的故障特征参量在最终约简集中出现的频次判定相应监测点的分类能力,实验结果表明该方法不需要对监测对象建模,也不需要对其进行动力学分析,而是直接对监测到的振动信号进行处理,根据各个测点的故障特征参量与故障种类之间的关联程度选择最佳测点,是一种行之有效的测点优化配置方法。 (5)基于粗糙集理论提取决策规则的过程没有学习归纳的能力,故障模式识别率较低。粗糙集理论的属性约简技术能够提取敏感的故障特征参量,最小二乘支持向量机的模式识别能力强,因此为了充分利用两者在特征参量提取与模式识别方面的优势,构建了基于粗糙集支持向量机的智能故障诊断系统。理论与实践都表明该系统在一定程度上提高了齿轮箱故障诊断性能,为非线性非平稳故障信号的处理与识别提供了一种较通用的解决方案。
[Abstract]:The gearbox is an important transmission part of the mechanical system, the failure rate is high, the vibration signal is nonlinear and non-stationary, the degree of fault, the position and type of the gear box have great influence on the characteristic parameters. When the gear box is monitored and fault diagnosis, the effective fault information can not be collected if the monitoring points are chosen unproperly, thus guiding the gear box. The fault location is not easy to be determined, the extraction of sensitive characteristic parameters is difficult and the fault pattern recognition rate is low. The local domain wave decomposition method adaptively decomposes and maps the nonstationary time-varying signal to the time frequency analysis plane, and can simultaneously display the full features of the time and frequency domain information of the signal, and the attribute reduction technique of rough set theory can be optimized. The fault feature parameter set is extracted and the sensitive fault feature parameters are extracted. The function approximation of the least squares support vector machine is good, and the pattern recognition ability is strong. Based on the local wave decomposition method to deal with the fault signal and the rough set theory, the rough set is combined with the least square support vector machine to establish the base. Gearbox intelligent fault diagnosis system based on rough set support vector machine. The main research contents and conclusions in this paper are as follows:
(1) on the basis of analyzing the vibration characteristics of the gearbox, the local wave decomposition technique is used to deal with the gear box fault signal and extract the initial fault feature parameter set. In the local wave decomposition process, the end point effect problem is alleviated by the combination of the mirror extension and the window function, and the general empirical mode decomposition method is adopted. The model aliasing problem is effectively solved. The experimental results show that the two methods have achieved good results in the breakdown signal decomposition of the gearbox. According to the index of the fault characteristic parameter set, the stability of the fault characteristic parameter set is measured by the mean square root of each working condition, and each characteristic parameter is used between the six working conditions. The minimum mean difference is used to measure the sensitivity of the fault characteristic parameter set. In the experiment, the normalized energy characteristic parameter set based on EEMD and the approximate entropy characteristic parameter set based on EEMD are extracted respectively. The results show that the former is basically the same as the latter, but the stability is better than the latter, so this paper uses the EEMD The normalized energy characteristic parameter set is used for gearbox fault diagnosis.
(2) a global dynamic optimization discretization algorithm based on the improved Naive Scaler algorithm is proposed. By improving the Naive Scaler algorithm process, we ensure that all the breakpoints that guarantee the unresolved relation can be obtained; the breakpoint is gradually increased by dividing the sample set by the breakpoint, and the breakpoint set is dynamically selected from the candidate set. The number of breakpoints is the least under the condition that the ability of information system classification is constant. By comparing with other algorithms, the experimental results show that the number of breakpoints obtained by the algorithm is less, which reflects the superiority of the algorithm in the discretization of continuous attributes.
(3) an attribute reduction algorithm based on conditional equivalence class is proposed. On the basis of the kernel attribute set, the algorithm can not be correctly classified into the class of the decision class in the condition class of the kernel attribute, and find the attributes that can distinguish the class from the rest of the kernel attribute, and add it to the kernel attribute set, thus the minimum attribute reduction is obtained. The attribute reduction algorithm based on the heuristic information can not guarantee that the reduction set must be the minimum attribute reduction set. The experimental results show that the computational complexity of the algorithm is low, and the reduction efficiency is improved.
(4) the attribute reduction technique of rough set is proposed to optimize the fault monitoring point. This method combines the minimum attribute reduction set of six fault monitoring points into a large decision table for attribute reduction, and determines the classification ability of the corresponding monitoring points according to the frequency of the fault characteristic parameters of each monitoring point in the final reduction concentration. The experimental results show that the method does not need to model the monitoring object and does not need to carry on the dynamic analysis to it, but directly handles the monitored vibration signal, and selects the best point according to the relation between the fault characteristic parameters and the types of the fault. It is an effective method to optimize the distribution of the measured points.
(5) the process of extracting decision rules based on rough set theory does not have the ability to learn induction, and the rate of fault pattern recognition is low. The attribute reduction technique of rough set theory can extract sensitive fault feature parameters, and the least squares support vector machine has strong pattern recognition ability, so it makes full use of both feature parameters extraction and pattern recognition. In other aspects, an intelligent fault diagnosis system based on rough set support vector machine is constructed. Both theory and practice show that the system improves the fault diagnosis performance of the gear box to a certain extent and provides a more general solution for the processing and recognition of nonlinear and non-stationary fault signals.

【学位授予单位】：中北大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TH132.41;TH165.3

【参考文献】