【作者】 周理；

【导师】 蒋新华；

【作者基本信息】 中南大学 ， 交通信息工程及控制， 2013， 博士

【摘要】 大型铝型材挤压生产线(Large Aluminum profile Extrusion Production Line, LAEPL)是由工频炉、挤压机主机、主辅机液压控制系统和油泵站、主辅机电气控制系统和监测装置、运锭机、推锭器、挤压筒加热及空气冷却系统装置等多种机械部件、油压单元、电气单元协调工作的复杂过程控制工业系统,故障种类多。当前LAEPL的故障诊断手段主要依赖人工经验,因此,对其进行智能故障诊断技术的研究具有重要意义。本文综合运用集成故障建模技术、液压故障诊断技术、传感器故障诊断技术、非线性系统状态估计理论、智能故障诊断理论等先进理论和算法,全面研究LAEPL的故障诊断关键技术,并开发了LAEPL状态监测与故障诊断系统。本文开展了集成故障建模方法研究,提出一种集成时序Petri网与混杂键合图(TPN-HBG)的LAEPL故障诊断模型,全面描述连续变量和离散事件,以满足LAEPL对诊断模型的要求。在集成故障诊断模型中,设计了系统层和设备层的故障诊断方法。其中在系统层,基于TPN设计了LAEPL故障诊断方法；在设备层,针对大型铝型材挤压机(Large Aluminum profile Extrusion Machine, LAEM)这一多液压源,多回路的复杂液压系统,提出一种集成混杂键合图与时间因果图(HBG-TCG)的故障诊断方法,用于检测与诊断LAEM这一关键设备的故障。针对LAEM的电液伺服系统的故障诊断,提出一种基于混杂系统模型及多元线性回归的故障模型,利用混杂系统理论抽象电液伺服系统的工况与故障,建立基于混杂系统的电液伺服系统模型,利用多元线性回归算法辨识系统参数,建立故障观测模型；实验表明,该故障诊断方法能有效应用于液压故障参数的监测和早期故障预报。LAEPL上的传感器数量、种类众多。因此,本文针对LAEPL故障诊断模型下的传感器故障诊断,提出一种基于最小折扣因子的证据不确定性修正算法,正确区分传感器故障、设备故障和环境造成的干扰。结合线性系统理论中观测器的设计方法,构造了用以实现传感器故障检测与隔离的残差产生器,利用几何理论中的不变子空间理论,通过特征空间分割,实现故障特征的解耦；利用空间投影运算,实现传感器故障的检测与隔离；该方法能对单故障实现检测与隔离,而且对多传感器并发故障同样具有很好的检测与隔离效果。在铝型材挤压状态估计研究方面,通过对挤压过程热力学理论分析,提出使用不确定性自回归滑动平均模型把挤压过程中的非线性连续状态表示成一个线性系统和非线性扰动,采用自适应摸糊神经网络与一种连续、单调并可逆的一一映射相结合的方式逼近非线性扰动。在铝型材挤压状态估计研究方面,通过对挤压工艺的分析,充分掌握控制参数与输出温升的关系,设计基于隐马尔科夫的离散状态估计模型,使用贝叶斯公式确定最有可能的离散状态,预测铝锭发生液相-固相突变的概率。在本文所研究的状态估计与故障诊断方法的基础上,开发了LAEPL故障诊断系统,结合铝型材挤压的工艺分析与故障诊断的要求,通过硬件升级,软件编程,研究开发了LAEPL远程监测与故障诊断系统,并已成功应用于某铝业有限公司的55MN铝型材挤压监测现场,实现了铝型材生产状态实时监测与故障诊断。更多还原

【Abstract】 Large aluminum extrusion production line (LAEPL) is a complex industrial process control system, which is a combination of a variety of mechanical parts, hydraulic union and electrical unit including the frequency furnace, extrusion machine, the main auxiliary oil pump station, hydraulic control system, electrical control system of main auxiliary equipment,the monitoring device, etc. The rate of failures in the LAEPL is high for it is very difficult to diagnose these faults in LAEPL from people’s practical experience, thus the research in intelligent fault diagnosis technique is of vital importance and very significant.This paper is a tentative research on key technologies of fault diagnosis in LAEPL by using a series of technologies. These technologies include integrated fault modeling technology, hydraulic fault diagnosis technology, sensor fault diagnosis technology, nonlinear system state estimation theory, intelligent fault diagnosis theory and other advanced theories and algorithms. On the basis of theoretical research, a fault diagnosis system in LAEPL is initiated and developed in this paper.This paper is expected to put forward a fault diagnosis model integrated with TPN-HBG by conducting some researches on the integrated fault modeling method, and based on hybrid system is supposed to be integrated in this model. This model can overall reflect continuous variables and discrete events, and would satisfy the requirement of diagnostic model in the complex process control system. Based on the integrated fault diagnosis model, two fault diagnosis methods in the system layer and the device layer are designed. To be more specific, a LAEPL fault diagnosis method in the system layer was designed. A multi-loop complex hydraulic system will be put forward in the device layer. This system is a kind of HBG-TCG based fault diagnosis method for LAEM (Large Aluminum profile Extrusion Machine) with multi hydraulic source for detecting and diagnosing the key equipment failure-LAEM.In order to diagnose electro-hydraulic servo system of LAEM, a brand-new model based on hybrid systems and multiple linear regression models is proposed in this dissertation by using multi sensor information fusion, a model based on hybrid systems and multiple linear regression models. The condition and the fault of electro hydraulic servo system are abstracted by using hybrid system theory, and the electro hydraulic servo system model of hybrid system is further established. Also the parameters of the system are identified by the use of multiple linear regression algorithms. Experiments show that two fault diagnosis methods are effective in fault diagnosis of hydraulic parameters monitoring and fault forecasting.For the diversity in the quantity and the type of sensors on LAEPL, this paper will propose good evidence to show the uncertainty correction algorithm on the basis of the smallest discount, and this algorithm can distinguish the sensor faults, equipment failure and environmental disturbance. In this paper, a residual generator is constructed to realize the sensor fault detection and isolation through combining the theory of linear system observer. Also by using the geometric theory of the invariant subspace, the fault feature decoupling will be achieved through feature space partition. The sensor fault detection and isolation system will be further realized by using space projection operation. By doing these, not only the single fault can be detected and isolated by the method, but also multi sensor fault can be detected and isolated effectively.In the field of state estimation research in aluminum extrusion, the extrusion processes is represented as a linear system through the analysis of the extrusion thermodynamic theory, and a nonlinear disturbance model is proposed by using ARMR model. By combining an adaptive-network-based fuzzy inference system with "one-to-one mapping", a compensator for unmodeled dynamics is constructed. Based on the research on state estimation in aluminum extrusion and the analysis of extrusion process, the relationship between the control parameters and the temperature rise is derived A discrete state estimation model based on Hidden Markov is designed; the most likely state can be determined by using the Bays formula. This method can be used to correctly estimate the extrusion process as well as the liquid-solid phase mutation probability prediction.On the basis of the research in the state estimation and fault diagnosis method, the fault diagnosis system for LAEPL will be developed in this paper, and the LAEPL running status monitoring and fault diagnosis system will also be designed for the actual need. This system has been successfully applied in55MN aluminum extrusion practice monitoring of a aluminum CO.,LTD, and it can also be used in the real time state monitoring and fault diagnosis.更多还原