【作者】 董敏；

【导师】 刘才；

【作者基本信息】 燕山大学 ， 机械设计及理论， 2006， 博士

【摘要】 作为现代板带轧机核心技术的液压厚度自动控制系统(HAGC),集机、电、液于一体,结构高度集成,复杂精密,造价十分昂贵,该系统的运行状态对产品的质量和产量具有重大影响,直接决定了轧机的工作性能。HAGC系统在运行状态下是一个复杂的非线性动力学系统,具有不确定、非线性、时变性的特点,故障状况复杂,干扰因素多,系统本身的复杂性给状态监测和故障分析带来许多困难,因此,开展HAGC系统的故障诊断技术研究具有重要的理论意义和实际应用价值。本文从提取故障信息的三种独立特征:瞬态特征、稳态特征和突变特征的角度,开发了多种诊断判据。为综合多角度信息,在利用新方法的基础上研究了多种判据信息融合问题,通过仿真及故障实例验证了所提方法的正确性及可行性。通过对连轧机轧制过程的分析,提出在不需要停机和外加激励信号的情况下,将轧制咬钢过程和控制输入突变过程看作阶跃输入信号,在线获得液压压下系统瞬态响应。分析了常见故障对HAGC系统瞬态响应性能指标的影响,并通过与正常情况下参数进行对比分析,建立了数值属性描述的瞬态响应判据算法。基于系统数学模型和未知输入观测器理论,利用状态方程对系统故障进行分析,确定了不同故障影响方向,建立了HAGC系统观测器组诊断方法,通过组合逻辑来进行故障定位与分离,并提出了数值属性故障判据的算法,理论分析和试验研究证明了观测器组在提取系统稳态故障信息方面的有效性。另外,本文基于微分几何方法研究了HAGC系统非线性模型负载不确定解耦子系统的建立以及非线性观测器的设计方法,限于其可诊断故障模式,文中将其作为观测器组方法的辅助校验。基于轧制过程历史数据,分别建立了一种RBF网络轧制力预报模型和基于压下系统各环节ARMA模型的动态递归网络模型,并对网络结构和学习算法提出了优化,使其收敛速度和泛化能力得到提高。为排除输入突变引起的输出信号突变带来的网络误诊问题,通过对系统MA模型进行小波变换研究,使得故障信息得到准确定位。将小波作为神经网络跟踪异常信号的后级处理,解决了小波变换计算量大,全程监控实时性差的缺点。鉴于小波变换对突变信息的敏感性,提出将其作为故障突变信息判据,并提出了故障测度的数值算法。更多还原

【Abstract】 As a kernel technology of modern strip rolling mill-Hydraulic Automatic GaugeControl system(HAGC system), gathering mechanism, electric, and hydraulic pressuretechnique, has the attribute of complicated and high precision, high integration inconfiguration and costly in price. The working state of HAGC has the important effect onthe quality and output of rolling production and decides the performance of rolling milldirectly. HAGC system is a complicated dynamics system in running state, has thecharacteristics of uncertainty, non-linearity and time-variety. It’s complexity bring manydifficulties to state monitor and fault analysis. So, research on HAGC system faultdiagnosis technology has important theoretical and practical significance.From the perspective of extracting three independent fault signal characters:transient, steady and singularity, multi diagnosis criterions are developed in this paper.Information fusion is researched based on the new methods in order to integratemulti-angle information. The result of simulation and fault example diagnosis verify thecorrectness and feasibility of the proposed methods.Analyzing the rolling course of continuous rolling mill, it is proposed that under thecondition of needn’t work stop and external exciting signal, bitting course and abruptcontrol input course can be taken as step input signal. So, transient response can beachieved on line. The numerical property algorithm is established by contrasting theeffect of frequent faults to transient response performance index with the parameters onhealthy states.Based on system mathematical model and unknown inputs observer theory, HAGCsystem observer group diagnosis method is established after different fault directions aredecided by analyzing with fault state equation. Fault isolation is achieved by combinationlogic. Numerical algorithm is proposed which make the criteria has the numericalproperty. Theory analysis and experimental research have proved that observer groupdiagnosis method is effective in extracting fault steady information. In addition, loaduncertainty decoupling subsystem of nonlinear model for HAGC system is establishedbased on differential geometry theory and nonlinear observer is designed. Limited to thediagnosticable fault mode, it is taken as a subsidiary check to linear observer groupmethod.Based on history data of rolling course, an RBF network rolling force model anddynamic recurrent network model based on ARMA model for each loop of screw downsystem are established respectively. Network structure and learning algorithm areoptimized, then the convergence rate and generalizing capability of networks areimproved. In order to exclude the error alarm problem caused by abrupt input, systemMA model is introduced and decomposed by wavelet transform, by which accurate faultinformation can be achieved. In this paper, wavelet transform is chosen as secondarydealing to the abnormity of network monitor, which can solve the defect of largecalculation and bad real time problem of wavelet transform. Owing to the sensitivity tosingularity information, wavelet is chosen as singularity information criteria andcorresponding numerical algorithm is proposed.Considering the advantages of D-S evidence theory at dealing withmulti-information, multi-criteria fusion is carried out using D-S evidence theory.Reasonable basic probability assignment algorithm is constructed with fault measuredegree, by which the shortcoming of basic probability assignment depending on expertknowledge and has excessive subjectivity is eliminated. Diagnosis example verify thatinformation fusion reduce the uncertainty, improve the diagnosis belief degree anddiagnosis system capability is improved greatly.更多还原