【作者】 吕琛；

【导师】 宋希庚；

【作者基本信息】 大连理工大学 ， 动力机械及工程， 2002， 博士

【摘要】 故障诊断技术对于减少生产实践中机械设备事故的发生，降低由此造成的经济损失意义重大。多年来科研人员在此领域进行了大量的研究工作，取得了很多宝贵的成果；本文在认真研究与总结前人成果的基础之上，从理论分析与实验研究等方面对故障诊断的测试手段与诊断方法这两个领域进行了研究。 本文的主要工作为2100型柴油机的主轴承磨损故障诊断与状态监测。内燃机往复式活塞和曲柄连杆机构的运动方式，决定了它具有冲击特点的工作过程，并且在实际应用中这种过程还是变工况的，它所产生的振动与噪声信号十分杂乱，不像旋转机械那样具用简洁的频谱特征。其内部的故障信号往往被淹没在正常的工作信号中而难以发现，所以，如何在杂乱的工作信号中提取表征故障的特征参数，是状态监测与故障诊断的关键所在。 首先，本文在振动测试方法已较成熟的基础上，提出了采用机体外噪声测量的方法来采集含有主轴承磨损故障信息的噪声信号，避免了主轴承振动测量中，需要将机体拆卸来安装加速度传感器，以及需要选择好传感器位置的麻烦。采用噪声测量方法具有易于现场测试、及时性强、测量方便的特点。作者通过调整主轴瓦的间隙状态来模拟内燃机实际工作中的间隙磨损状态，测得各磨损状态下的噪声信号。在数据的处理方面，提出了阶次这一特征分量在机械故障诊断中的作用，以及其原理和各种具体实现方法。而时域同步平均可以有效衰减与回转频率无关的干扰，提取与工作状态直接相关的周期信号。在比较几种传统方法后，提出了采用高速数字信号处理器DSP电路和插值算法来进行噪声信号的时域同步平均处理。克服了基于锁相环电路采样方法跟踪精度低、速度慢和使用不便的缺点。 在故障特征频率提取方面，提出了采用基于噪声和正交小波分析检测主轴承磨损故障的一种新方法。利用小波分析将测得的机体噪声信号变换到时频域，选择合适的频带加以提取并进行包络谱分析提取包含特征频率的包络，通过分析提取后的0.5倍频和2倍频分量，发现两个特征频率可准确反映主轴承的磨损状况。并说明了这种方法的有效性和应注意问题。将各个磨损间隙状态下得出的特征频率幅值送入建立好的神经网络模型进行训练后，得出了可辨别各故障状态的网络模型。 此外，在诊断理论方面，本文提出了新的内燃机故障诊断方法—基于小波包和图像处理的诊断方法。首先对噪声信号进行小波包分解，得到含有待诊断柴油机主轴承间隙故障信息的噪声信号时频分布图，并由此定义图像。然后对图像进行平均，得到主轴承在不同间隙状态下的“标准”时频分布图，建立了柴油机主轴承间隙图像匹配的故障诊断模型，匹配时所用的参数为两幅图像之间的欧氏距离。结果表明此方法简单有效，充分利用了噪声信号及其小波包分解所包含的状态信息，为内燃机的状态监测与故障诊断开辟了一条新的途径。 在上述故障诊断理论与测试手段的基础上，编写了基于局域网的机械故障诊断与状态监测软件。软件功能包括了服务器和客户端的信号采集与监测、数字滤波、FFT谱分析、包络谱分析、小波分析、小波包分析、神经网络训练与模式识别、数据库管理、网络通信等模块。其中客户端除了采集信号外，其它的一切功能(信号处理与分析、故障识别)均提交到服务器处理；即实现“瘦客户机”的概念。数据采集模块可同时对多路振动、噪声、温度、压力等信号进行采集。程 大连理工大学博＿士学位论文序还特别加强了噪声采集的功能，除了通过声级计采集噪声外，还编制了利用计算机声卡进行采集的模块，噪声采集中可对噪声自适应地按照分贝级采集并存储为各种格式的音频文件或数据文件。在各种信号的监测中，可以实时地监测多路数据、定制显示范围、查看数据值。同时增加了色带报警与自定义语音报警模块。 基于上述软件，本文提出了基于Web的远程设备监测与维护系统。虽然现在有许多成功的基于Web的应用，但是专门为设备维护设计的基于Web的系统很少。一个主要的原因是开发这样一个系统存在许多困难，因为它包括大型数据库、信号处理、设备监测和实时数据通信。更重要的是同时具有设计基于Web虚拟仪器和故障诊断经验的专家很少。然而，基于Web的虚拟仪器的益处是明显的，用于设备监测和分析的昂贵硬件仪器可以被低廉的基于软件的虚拟仪器代替。通过在Internet服务器上安装虚拟仪器，当前用于工厂的冗余仪器可以精简到最小。因为每种虚拟仪器只有一套拷贝安装在服务器上，所以维护工作变得简单。通过Web的交互式维护变得容易。本文给出了这样一套基于Web的维护系统的设计框架。它包括：远程数据传感和记录，采用虚拟仪器的信号处理与分析、智能故障诊断、以及Web交互式维护平台。整个平台可以采用跨操作系统的JAVA语言OSP、JavaBean、JavaApplet、JDBC）或 MATLAB Web Server来设计虚拟仪器和诊断数据库等模块。 最后本文中提出利用包含有故障信息的原始模拟信号（振动加速度传感器信号、精密倍频程声级计输出的噪声信号）经过前置信号处理和神经网络运算，得出VLSI电路输?更多还原

【Abstract】 Fault diagnosis is important to the reducing of machinery accidents and economic loss in manufacture and practice. Scientists have researched a lot in this area these years, and have got many precious achievements; In this paper, based on the studies of former achievements, a lot of theoretical and practical researches in test method and diagnosis theory have been made by means of theory analysis and experiment.The experiment goal of this paper is main bearing abrasion fault diagnosis and condition monitoring for 2100-BI diesel engine. The moving manner of reciprocating I.C.E. shows its working processing of shock characteristic, simultaneously generates wild vibration and noise signal and often changes its operate speed. So it’s difficult to get unique and simple spectrum unlike a rotary machine. The signals that indicate the faults of I.C.E. are usually being concealed by the normal working signal. The key work of condition monitoring and fault diagnosis should be finding out the fault feature parameters from disorder working signals.Firstly, based on the developed vibration testing method, a method for monitoring abrasion fault on main bearing of reciprocating engine using noise measurement is presented. It can avoid the trouble of vibration method that needs disassemble engines for placing acceleration sensors and selection of the position of sensors. On data processing, a order domain analysis method based on DSP technique is presented along with several concrete realizing methods. Time domain synchronous averaging can reduce the signal not coherent to rotation frequency effectively, and extract the repetitive signals directly related to machine operation condition. Both methods have some similar characteristics that this paper has introduced. After comparing with a few traditional methods, I present a type of high-speed DSP circuit and a mathematic algorithm to process order domain analysis and time domain synchronous averaging processing for vibration and noise signal. It improves the accuracy and tracking speed comparing to PLL based method.On the aspect of extracting fault feature frequency, a method for monitoring abrasion fault of main bearing using noise and orthogonal wavelet transform is presented. Wavelet analysis is used to translate noise signal of the engine into time-frequency domain. Selecting appropriate frequency band to pick up and making its envelope spectrum, then two feature frequency-0.5x speed frequency and 2x speed frequency that represent the abrasion condition can be found.In addition, a new diagnosis method of image processing technique is proposed. We illustrate the wavelet package analysis that can obtain the time-frequency distribution figure of noise signal, which can be used to define gray-image. Through image-average, we can get the standard time-frequency distribution images of different main bearing gap. Based on that, a main bearing wear fault diagnosis model using image matching is set up.A intranet-based mechanism fault diagnosis & monitoring software is also be compiled. Its main functions include: client/server signal acquisition and monitoring> FFT spectrum x envelope spectrum N wavelet /wavelet package analysis N neural networkintraining and mode recognition database manage > network communication and so on.Based on above software system, a web-based remote equipment monitoring and maintenance system is presented. Even though there are many successful Web-based applications, as today, designated Web-based system for maintenance is seldom. One of the main reasons is the difficulty in developing such a system as it involves huge database, intensive signal processing, and near real-time data communication for monitoring and control of equipment. Most importantly, the expert, who has experience in designing both the Web-based virtual instruments and automated system for equipment fault diagnosis, is difficult to be found. Nevertheless, the benefits provided by the Web-based maintenance system are substantial. Expensive hardware based instrumen更多还原