Research on Model-Based diagnosis of Dynamic Systems

本研究项目针对以下4类动态系统开展了基于模型的故障诊断方法及其应用研究：(1)状态空间模型描述的集总参数线性系统(2)黑箱表示的过程(3)电厂等大系统(4)分布参数系统。(1)利用并行卡尔曼滤波和广义似然比(GLR)检验建立了一个可在线执行的分层诊断系统。此外，将数值积分或无限脉冲数字滤波方法与递推辨识算法相结合，提出了几种直接估计连续时间线性系统物理参数的方法。通过…仿真验证了该估计器对故障诊断问题的适用性(2)首先，为了研究现有的多种用于故障诊断的递推辨识算法的收敛和数值不稳定性，在个人计算机上制作了一个CAD系统软件包。其次，针对自适应控制系统，提出了一种以Kullback判别信息为检测指标的在线诊断方法。此外，还提出了一种利用模糊推理来判断检测到的系统参数变化是否是由于TA故障引起的诊断方法。特别是，该方法可以被认为是一种新的诊断方法，称为基于知识的诊断，它有效地将有关系统运行行为和故障模式的现有知识融入到基于模型的诊断中。(3)提出了火电厂可控除氧器系统的两种诊断方法。一种是基于线性化摄动模型和GLR检验的方法，另一种是基于仿真器的小参数调整方法，并以实际火电厂模型为例进行了验证。(4)针对抛物型和/或双曲型分布参数系统，提出了几种状态变量、系统参数和未知边界条件的估计与辨识方法。另一方面，它们在诊断问题上的适用性还在研究中。较少

In this research project, developments of model-based failure diagnosis methods and their application studies have been carried out for the following 4-types of dynamic systems,(1) Lumped parameter Linear Systems Described by State-Space Model(2) Processes with Black-Box Type Representation(3) Large-Scale Systems like an Electric Power Plant(4) Distributed Parameter Systems.Research results of the project obtained for each system are summarized as follows :(1) A hierarchical diagnosis system which can be performed on line has been developed by using parallel Kalman filter and generalized likelihood ratio (GLR) test. Furthermore, several methods for directly estimating physical parameters of continuous time linear systems have been proposed by a combinational use of the numerical integration or infinite impulse digital filter approach and the recursive identification algorithms. The applicability of the estimator to failure diagnosis problem has also been confirmed through the simulatio … More n.(2) First, in order to investigate the convergence and the numerical instability in many kinds of existing recursive identification algorithms, which are used for failure diagnosis, a soft-ware package of CAD system has been manufactured on the personal computer. Next, for adaptively controlled systems, a method of on-line diagnosis method has been proposed by using Kullback discrimination information as the detection index. Furthermore a diagnosis method using fuzzy inference to decide whether the detected system parameter changes are due to a failure or not tas also been developed. In particular, the method can be considered as a new one, so called "Knowledge Aided Model-Based Diagnosis", which effectively incorporates available knowledge about the system operating behavior and the failure modes into the model-based diagnosis.(3) Two diagnosis methods of controlled deaerator system in the thermal power plant has been developed. The one is a method using a linearized perturbation model and GLR test, and the other is a simulator-based method with small parameter adjusting scheme, and their effectiveness has been confirmed for the model of a practical thermal power plant.(4) Several methods for estimation and identification of state variables, system parameters and unknown boundary conditions have been proposed for parabolic and/or hyperbolic type of distributed parameter systems. On the other hand their applicability to diagnosis problems are on being investigated. Less