【作者】 张捷；

【导师】 李伯全；

【作者基本信息】 江苏大学 ， 机械设计及理论， 2003， 硕士

【摘要】 齿轮箱作为拖拉机的重要传动部件，对其进行状态监测和故障诊断的研究具有重要的实际意义。本文致力于拖拉机齿轮箱齿轮的自动故障诊断技术的研究。 本文在介绍了齿轮的主要故障模式和其机理的基础上选定了利用神经网络来进行故障诊断的方案。按照神经网络用于模式识别的主要步骤本文主要由三部分组成：设计并搭建齿轮箱故障的模拟系统和振动信号采集的系统；对采集到的信号进行合适的处理获得故障的特征信号；应用神经网络对故障诊断进行训练和检验。本文的重点在后两部分，这两部分的工作主要是应用自己设计的一个智能故障诊断系统软件来完成的，在软件中集成了对XML(可扩展标记语言)的支持，可以随时对工程进行标记非常适合于工程实际中的应用。而本软件中的信号处理和神经网络部分则是应用了集成的MATLAB引擎来完成的，利用这种方法既利用了MATLAB中的丰富资源又减轻了程序开发的工作量。文中还对于BP网络提出了应用主元分析进行改进的算法，经试验检验这种方法能够有效的减少神经网络的训练时间。最后又由齿轮箱模拟故障试验测得的试验数据在智能故障诊断系统软件中进行信号处理和神经网络的训练，经对训练好的神经网络检验可以看出它是完全能满足齿轮箱齿轮的故障诊断要求的。更多还原

【Abstract】 Gearbox is one of the most important elements in a tractor. It is significant to study the condition of monitoring and fault diagnosis of gearbox. This thesis aims at the investigation of automatic diagnosis techniques of gearbox.After introducing the main faults of gear and the causes of them, the project using neural-network in diagnosing is put forward. Following the steps of neural-network being used in pattern recognition, the paper is made up for three parts. Firstly, this paper illustrates the building of the indoor load-adding testing system applying on the study of diagnosis experiment and it’s working principle. Secondly this paper introducing the way in transacting the signal and getting character of fault. It is the kst one that training the neural-network and testing it The later two parts are emphases of the paper. The software programmed by myself works out the task of the later two parts. The software includes support for XML. Using it the software can mark the project. The engine of MATLAB accomplishes the task of signal processing and neural-network. Using the source of MATLAB can reduce the load of programming. In this paper, a way to improve the BP arithmetic based on PCA theory is expounded. Trough experiment it can be proved that the improvement can notable lessen time in train network. At kst after dealing with the signal collected in test-bed for simukting the broken-down gear-box, using the software a neural-network is designed. After being trained and testing the neural-network is proved to reach the require of diagnose for gear-box.更多还原