【作者】 饶泓；

【导师】 扶名福； 黄菊花；

【作者基本信息】 南昌大学 ， 材料加工与工程， 2009， 博士

【摘要】 液压机在现代生产中扮演极为重要的角色,液压机故障轻则引起生产线停产,重则造成安全事故,因此需对液压机工作状态作出准确判断。但是液压机系统构成复杂,采集信息具有多样性、随机性、复杂性和关联的层次性,日常运行时,需要监测的量有10多种,很多特征量之间是有相关性的,加上采集信息手段受各种因素影响,从而造成了信号的随机性和不确定性,对液压机的故障诊断造成了极大的困难,因此本文研究先进的液压机故障诊断方法,以获取液压机的准确工作状态。本论文将液压机的故障诊断分成两部分进行,一部分是液压动力子系统,另一部分是液压控制子系统。重点讨论了液压动力系统关键部件液压泵和液压控制子系统的故障诊断。由于液压泵工作环境恶劣,泵出口监测信号通常杂乱无章,容易被噪声信号淹没,单一传感器提取的时、频特征信息常呈现出较强的模糊性,采用常规信号处理方法难以有效提取故障特征。因此,需充分利用多传感器的信息源,以获得对设备状态的可靠估计。本文在液压泵的故障诊断中综合利用了泵壳三个方向的振动信号,并辅以液压泵外泄口温度信号,在两个层次分别对振动信号进行空间融合诊断、对振动网络诊断结果和温度诊断结果进行融合,获得对液压泵故障的准确诊断。而液压机控制子系统涉及的设备较多,采集的特征量非常繁杂,难以获得有效的故障诊断规则,故障诊断非常困难。因此对液压机控制子系统采用基于粗糙集理论的故障规则提取算法,通过属性约简和决策网络的构造,提取清晰规整的故障规则,根据这些故障规则,从液压控制子系统的表征就可以容易地推测出故障原因。最后,设计和实现了一套基于B/S结构的液压机远程在线监测与故障诊断系统,将本文所提出的先进故障诊断方法引入该系统,可在线获取设备现场数据,远程传送特征数据,实现远程故障诊断。工厂的实际应用表明该系统有效解决了液压机故障诊断难、诊断效率低的问题,获得较好的诊断效果。只要导入其它大型机电设备的知识库,该监控系统就可以方便地应用到其它大型设备的状态监测和故障诊断中。本文的主要创新点如下:(1)提出了PARD-BP(PARD,Pruning Algorithm based Random Degree基于随机度的剪枝算法)神经网络故障诊断方法,该方法在随机度的基础上,利用分治算法的思想对BP神经网络隐层的冗余节点进行剪枝,获取精简的网络结构,使网络具有更好的泛化性能,使故障诊断结论更可信;(2)提出了基于PSO(PSO,Particle Swarm Optimization粒子群优化算法)的H-BP多级神经网络故障诊断方法,该方法利用PSO计算上的优势,首先对Hopfield网络权值矩阵进行优化,再利用Hopfield对故障特征数据进行预处理,最后通过BP网络实现故障诊断。该方法可有效解决BP网络易陷入局部最小的问题,可有效提高神经网络的诊断精度;(3)提出了液压泵两级多源信息融合故障诊断模型,充分利用了多传感器的资源,最大限度发挥系统资源利用率。该模型采用PARD-BP神经网络进行各方向振动信号诊断后,进行一级振动子网诊断融合;再利用H-BP神经网络进行温度信号诊断;利用两种信号的诊断结果作为独立证据并构造概论分配函数,进行第二级D-S决策级融合。将数据融合技术应用于液压泵的故障诊断,一定程度上能获得精确的状态估计,增加置信度,提高诊断容错性和鲁棒性。(4)提出了基于粗糙集理论的液压控制系统故障诊断规则提取方法。为提高故障规则的提取效率,对粗糙集理论中的约简算法进行了优化,缩短了故障规则的提取时间:同时为了有效滤除噪声和处理不一致性规则,在准确度的基础上引入了规则覆盖度的概念,对提取的规则进一步评价,最终提取出有效的诊断规则。(5)将智能故障诊断理论引入到“基于B/S结构的液压机在线监测与故障诊断系统”中,与传统故障诊断系统相比,能消除信号噪声过大而导致的误诊和漏诊现象,并能实现设备状态信息的实时监控。更多还原

【Abstract】 Hydraulic press plays a very important role in modern manufacture, the failures of hydraulic press often lead to the stop of the production line, even the occurrence of the accident. So it is necessary to determine the work status of hydraulic press correctly. However, the hydraulic press is a complex system, with diversity, randomness, complexity, and the relevancy of information gathering. More than 10 kinds of signal need to be monitored, together with the means of acquisition information are easily affected by variety of factors,which result in the signal’s uncertainty and causes the great difficulties of hydraulic press’s fault diagnosis.Thus,the thesis researches on the advanced fault diagnosis method to obtain accurate working status of the hydraulic press.In this paper, the hydraulic press fault diagnosis is divided into two parts, one part is the hydraulic press power system, and the other part is the hydraulic press control system. We focused on discussion about hydraulic pumps of the power system and hydraulic control system’s fault diagnosis.Due to poor working environment of the hydraulic pump, monitoring signal of the pump’s output port is usually chaotic, and vulnerable to be interfered by the noises. The frequency characteristics of the information extracted from single sensor are often ambiguous. The conventional signal processing methods can not extract fault features effectively. As a result, it is necessary to take advantage of the different information from multi-sensor to obtain reliable estimates of the state of equipment. This paper utilizes the three directions’s vibration signals of the hydraulic pump shell comprehensively,supplemented by temperature signals of the hydraulic pump output port. The method fuses the vibration signals in space to diagnose in first level and fuses the vibration and temperature diagnostic result in second level to diagnose the hydraulic pump accurately.Hydraulic press control systems of the hydraulic press involve many equipments, the collection of feature is very complicated. It is hard to obtain effective fault diagnosis rules and fault diagnosis is very difficult. Therefore, based on rough set theory , the fault rule extraction algorithm for the hydraulic control system was proposed, which extracted clear and regular fault rules through attribute reduction and decision-making network construction. It is easy to speculate the reasons for failure from the characters of the hydraulic control system with the extracted fault rules.Finally, a set of remote on-line monitoring and fault diagnosis system of the hydraulic press based on B / S structure was developed,which adopted the advanced fault diagnosis theory that proposed in this paper. It can get on-line data of equipment working state , transmit long-distance characteristics data and realize remote fault diagnosis. The practical application in the plant showed that the system solved the problem of the difficulty and inefficient in hydraulic press diagnosis and obtained better diagnostic results. The system can also be easily applied to other large equipment condition monitoring and fault diagnosis.The main innovation points are as follows:(1) The PARD-BP(PARD, Pruning Algorithm based Random Degree)neural network fault diagnosis method was introduced in this thesis,which pruned the redundant nodes of the BP neural network on the basis of random degree and division algorithm to obtain the reduced network structure. The reduced neural network has better generalization performance and the fault diagnosis conclusion is more credible;(2) The H-BP multi-stage neural network fault diagnosis method based on particle swarm optimization was proposed in this paper, which takes the advantages of the PSO’s caculation ability to optimize the weight matrix of the Hopfield network, pre-processes the characteristics of the failure by the optimized Hopfield network, and fault diagnoses by the BP neural network. This method can effectively solve the problem that BP network is easy to fall into local minimum and can improve the accuracy of diagnosis effectively;(3) The 2-level multi source information fusion fault diagnosis model was proposed in this paper,which took full advantage of multi-sensor to maximize utilization of system resources.The model took PARD-BP neural network to carry out the 3-direction vibration singal diagnosis and fuses the diagnostic results in the first level;and then use the H-BP neural network for temperature singnal diagnosis;at last ,fused diagnosis at the second level taking the two diagnosis result as the independent evidences and constructing the mass function with the two signal’s diagnosis results.Applied data fusion technology to the hydraulic pump fault diagnosis can obtain accurate state estimates to a certain extent and increase the confidence and improve the robustness of diagnosis;(4) In order to improve the efficiency of faults diagnosis rule extraction in hydraulic press control system,the reduction algorithm of rough set theory is optimized to shorten the rules’ extraction time;at the same time,in order to effectively filter out noise and to deal with inconsistent rules,the concept coverage was introduced on the basis of the accuracy concepts to further evaluate the extracted rules and finally extracted the effective diagnostic rules.The examples proved the effectiveness of the method;(5) Intelligent fault diagnosis theory has been introduced to the system " The large-scale electrical and mechanical equipment on-line monitoring and fault diagnosis system based on B / S structure ".Compared with the traditional fault diagnosis system, the on-line sytsem with the intelligent fault diagnosis theory can avoid misdiagnosis and wrong diagnosis caused by excessive signal noise , and can monitor the status of the equipment real-time.更多还原