基于D-S证据理论的多模型融合齿轮早期故障智能诊断方法研究

发布时间：2018-03-31 19:27

  本文选题：齿轮传动系统　切入点：重分配小波尺度谱　出处：《西安建筑科技大学》2014年博士论文

【摘要】：齿轮箱是一种量大面广的机械设备关键性基础部件,也是最易损坏的零部件之一,其运行状况直接影响到整个机器或机组设备的安全运行,因此,如何能尽早发现齿轮系统的早期故障,做到合理组织安排设备的维修,避免发生重大安全事故,造成重大的经济损失具有重大意义。机械设备的振动信号蕴含着系统(正常、故障)状态的信息,各种类型故障也有一定的规律可循,因此,采用振动信号对大型、关键机组运行状态监测和故障诊断是目前设备管理维护的主要手段。由于受齿轮传动振动响应和环境噪声的影响,齿轮早期故障的微弱信号往往被其他成分或环境噪声淹没,故障信号具有复杂的非线性、非平稳特性,采用传统的基于平稳信号假设的信号处理方法很难对其取得准确诊断,因此,研究有效去噪、消噪信号预处理技术和非平稳信号处理方法对设备故障诊断具有非常大的意义。小波分析和经验模态分解(EMD)是近年来发展起来的两种处理非平稳信号的时频方法,小波阈值去噪,形态滤波,奇异值分解技术是几种应用较多的去噪方法,两种时频方法与几种去噪方法相融合,被广泛应用于信号检测,机械故障诊断等工程领域。同时,随着设备向着高速度、高功率、高可靠性、大型化/微型、智能化、集成化的方向发展,使得传统的设备故障诊断方法和单一智能诊断方法已不能完全满足设备状态复杂性的需求,将多种智能诊断方法相融合的智能诊断技术是目前研究的热点方向。因此,本文对齿轮系统动力学和故障形成机理、小波分析理论、小波阈值去噪和重分配小波谱奇异值去噪、Hilbert-Huang变换理论、D-S证据理论、遗传算法-BP神经网络,模糊优化理论研究的基础上,提出了基于D-S证据理论的多模型融合齿轮早期故障智能诊断方法,分析齿轮典型故障信号的结果验证了该方法的有效性。对齿轮故障诊断提供了依据。本文主要工作如下:[1]本文建立了考虑摩擦、时变刚度、齿侧间隙的具有偏心直齿轮摩擦-间隙齿轮振动模型,分析考虑摩擦、齿侧间隙、偏心质量时的齿轮动力学行为以及它们的频谱特征。[2]提出了基于shannon熵优化TBP参数的重分配小波尺度谱进行SVD降噪方法,通过仿真信号分析发现该方法具有比小波尺度谱、重分配小波尺度谱更好的时频聚集性,且其时频分辨率能够同时实现最佳,具有更高的时频分布可读性。因此,该方法能够识别出强噪声背景下的机械早期故障微弱信号成分,为强噪声背景下机械早期故障微弱信号的去噪、消噪以及特征提取和故障诊断奠定了一定的理论基础。[3]将经验模式分解(EMD)方法和分形维数融合,提出了基于小波阈值去噪和EMD分形融合故障诊断方法,列出了基于EMD的分形维数的具体步骤。并将该方法应用于齿轮传动齿面磨损、断齿故障状态振动信号的故障诊断中,用关联维数均方根值替代关联维数,实现对齿轮齿面磨损和断齿等故障的准确诊断,取得了良好的效果。[4]提出了基于D-S证据理论的多模型融合智能齿轮故障诊断方法,通过实例验证:本文提出的多模型融合模型能够综合利用各单一智能模型的优点,使得区分度比单一模型有明显提高,即使单一模型出现误判,该融合模型仍然能够得到正确的诊断结果。具有较好的容错性、纠错性。
[Abstract]:The gear box is the key basic parts of machinery and equipment of a large amount of wide, one of the most easily damaged parts, its operation conditions directly affect the safe operation of the machine or equipment. Therefore, how to find fault gear system as soon as possible, to achieve a reasonable arrangement of equipment maintenance, to avoid the occurrence of major security the accident caused significant economic losses is of great significance. The vibration signals of mechanical equipment contains system (normal, fault) state information, various types of fault has certain rules, therefore, the vibration signal of the large, key unit condition monitoring and fault diagnosis are the main means of equipment management and maintenance. Because of influence of gear vibration and noise, weak signal early gear failure is often other components or environmental noise, the fault signal is complex Complex nonlinear, non-stationary characteristics, using the traditional signal processing method based on the assumption of stationary signals it is difficult to obtain accurate diagnosis, therefore, research on effective denoising, denoising signal preprocessing techniques and non-stationary signal processing method has great significance for fault diagnosis. Wavelet analysis and empirical mode decomposition (EMD two) is developed in recent years, processing non-stationary time-frequency method, wavelet threshold denoising, morphological filtering, singular value decomposition technique is widely used in several denoising methods, two kinds of time and frequency of several denoising method of integration, has been widely used in signal detection, fault diagnosis etc. engineering. At the same time, along with the equipment towards high speed, high power, high reliability, large / miniature, intelligent, integrated direction, making the equipment fault diagnosis method and the traditional single intelligent diagnosis method has not been able to Fully meet the equipment requirement of complexity, the intelligent diagnosis technology combining intelligent diagnosis methods is the current research focus. Therefore, the gear system dynamics and fault formation mechanism, the theory of wavelet analysis, wavelet threshold de-noising and reassigned wavelet singular value spectrum denoising, Hilbert-Huang transform theory, D-S theory of evidence. -BP neural network, genetic algorithm, fuzzy optimization theory, a multi model D-S evidence theory fusion incipient fault diagnosis method based on the analysis of typical fault signals of gear results verify the effectiveness of the proposed method. Provide the basis for gear fault diagnosis. The main work is as follows: This paper established [1] friction, time-varying stiffness, friction with eccentric gear clearance gear vibration model of gear backlash, considering friction, backlash, eccentric quality The gear dynamic behavior and their spectral characteristics.[2] proposed reassigned wavelet scale Shannon entropy optimization TBP parameter spectrum denoising method based on SVD, the simulation signal analysis shows that this method is better than wavelet scalogram and reassigned scalogram better time-frequency aggregation, and the time-frequency resolution can also achieve the best. Has the time-frequency distribution more readable. Therefore, the method can identify early fault weak signal components under strong background noise, as strong noise background machinery early fault signal denoising, denoising and feature extraction and fault diagnosis has laid a theoretical foundation of the.[3] (empirical mode decomposition EMD) fusion method and the fractal dimension, the wavelet threshold denoising and EMD fractal fusion fault diagnosis method based on the list of specific steps of fractal dimension based on EMD and the party. Method is applied to the gear tooth wear, broken tooth fault diagnosis fault vibration signal, using the correlation dimension of RMS value instead of correlation dimension, realize the accurate diagnosis of gear tooth wear and broken tooth fault, achieved good results with the.[4] proposed model D-S evidence theory fusion method for gear fault diagnosis based on intelligent, proved that this multi model fusion model can take advantage of the single integrated intelligent model, the discrimination is better than a single model, even a single model of misjudgment, the fusion model can still obtain the correct diagnosis result. With fault tolerance, good error correction.

【学位授予单位】：西安建筑科技大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TH132.41
，

本文编号：1692140