【作者】 赵国良；

【导师】 于学兵；

【作者基本信息】 大连理工大学 ， 动力工程（专业学位）， 2018， 硕士

【摘要】 动力组为机车柴油机提供动力,是保证其正常工作运行的重要部分,对于动力组故障的诊断显得尤为重要;而轴承在机车中也同样占据着重要的位置,也极易发生故障,尤其是滚动轴承的结构更复杂、故障类型更多。随着现代智能化方法的发展,故障诊断和健康管理系统(PHM)的诊断方法也更加多元,本文主要是利用故障树法和神经网络法对机车柴油机动力组和滚动轴承的故障进行诊断研究。故障树法采用图形化方法来描述故障原因,使得故障的因果关系更加直观。本文就根据故障树法的分析步骤和规则对柴油机动力组的常见故障进行分析,并据此建立相关的故障树。神经网络的特点使得它在对故障的诊断识别方面有较好的应用,本文采取应用最多的前向型网络对动力组和轴承故障进行诊断研究。以柴油机缸套磨损故障为例对BP神经网络进行网络设计,建立网络模型并对其进行训练和测试仿真,实现对故障的诊断识别;针对BP网络的缺陷提出优化改进措施,并对各改进方法进行比较,结果表明,采用LM优化算法的网络性能和诊断效果最好,可以达到诊断的要求;为避免BP网络固有的缺陷采用RBF神经网络进行故障诊断仿真,采用两种方法设计网络模型,并对其诊断结果进行比较。总的来说,RBF网络的训练速度更快、网络性能更好、诊断精度也更高。最后针对机车滚动轴承的表面损伤类故障,利用小波包分析理论对振动信号进行分析处理得到故障征兆特征,然后再采用RBF网络对其诊断仿真,结果显示这两种方法的结合对故障的识别效果非常好,可以准确地识别出轴承的各种状态。更多还原

【Abstract】 The power assembly provides power for the diesel engine of locomotive,it is an important part of ensuring diesel engine’s normal operation.So,the fault diagnosis of power assembly is important.The bearing also occupies an important position in the locomotive,and it is also prone to fail,especially rolling bearing’s structure is more complex and there are more types of faults.With the development of modern methods of intelligent diagnosis,the diagnosis methods of fault diagnosis and health management system(PHM)are more diversified.This paper used the fault tree and neural network to analyze the faults of power assembly and rolling bearings.The fault tree is a graphical method to describe the faults,the causality of fault became more intuitive.This paper analyzed the common faults of power assembly according to the steps and rules of the fault tree analysis,then established the fault tree.The characteristics of the neural network indicate that it is suitable for the pattern recognition of faults.In this thesis,feed forward neural networks wildly applied is used to diagnose the faults of power assembly and bearings.The fault of cylinder liner wear is cited here,the model of network is established,then trained and tested the network,finally the diagnosis is accomplished by BP network.In the meantime,there are some measures to optimize the BP network,then compared these improvement measures.The results proved that using LM optimization algorithm is the best improvement measure,and it can meet the requirements of fault diagnosis;RBF neural network can avoid the inherent defects of the BP network.In this thesis,two methods are used to design the RBF network model,then compared their results.In general,the training speed of RBF network is faster than BP network,and the performance is better,the diagnostic accuracy is higher.Finally,the wavelet packet analysis theory is used to analyze the vibration signal of locomotive rolling bearing’s surface damage,then the fault is diagnosed by RBF network.The results proved that the combination of wavelet packet analysis and neural network is very effective in fault diagnosis,and the fault status of bearing can be accurately identified.更多还原