Research on Fault Diagnosis Method for Nonlinear Systems Based on Their Structures' Modeling

在本研究项目中，我们对基于模型的非线性黑箱系统故障诊断方法的发展进行了研究，取得了以下成果：1.非线性系统辨识的准ARMAX模型的发展我们发展了一个与ARMAX模型具有相同线性结构的通用输入输出型非线性黑箱系统辨识模型--拟ARMAX模型。它通过非线性非参数建模将系统的非线性特性嵌入到ARMAX模型参数中，具有足够的灵活性来描述各种类型的非线性系统。此外，该模型具有结构建模的特点，对线性系统理论框架下的系统分析和控制设计具有广泛的适用性。通过对作为基准…对象模型的船舶推进装置模型的仿真研究，证实了该模型可以有效地用于非线性系统的故障检测。更多关于故障诊断问题的RK测试。2.基于准ARMAX模型的故障检测特征提取利用现有的递推辨识方法，如预测误差法，可以估计准ARMAX模型的参数，并将正常运行期间辨识出的模型作为故障检测的参考模型。引入Kullback判别信息(KDI)作为故障检测的指标。KDI是两个辨识模型之间的失真度量，即在正常运行时获得的参考模型和监测期间在线辨识的对象模型。为了建立高性能的故障检测系统，我们对基于准ARMAX建模和辨识的特征提取过程提出了几种改进方案。通过对船舶推进基准系统的仿真研究，验证了该方法对各种故障模式的有效性。3.故障检测与隔离系统的实现准ARMAX模型本质上是一个表示未知系统的输入输出型数学模型，其参数不包含关于系统结构的任何物理信息。然而，该模型具有带权因子的多线性形式，因此可以在一定程度上反映由于结构参数变化而导致的故障模式特征在辨识模型参数中的反映。基于这种思想，我们在故障检测系统中实现了故障隔离功能。故障检测后，在由模型参数组成的特征空间中利用模式识别方法进行故障分离，根据典型故障模式的先验知识构造典型故障模式对应的参考特征集。这样，我们针对黑箱非线性系统开发了一个基于模型的FDI系统，并通过仿真研究验证了该系统的有效性。在这些研究中，我们还证实了本研究项目同事开发的几种模式聚类和识别方法可以有效地用于构建FDI系统。较少

In this research project, we have performed studies on the development of a Model-based Fault Diagnosis Method for nonlinear black-box systems and obtained following results.1.Development of A Quasi-ARMAX Modeling for Identification of Nonlinear SystemsWe developed as the general input-output type model for identifying nonlinear black-box systems "A Quasi-ARMAX Model" with the same linear structure as the ARMAX model. It has enough flexibility to describe various types of nonlinear systems by imbedding system nonlinear characteristics into the ARMAX model parameters through nonlinear nonparametric modeling. Furthermore the model has wide applicability to system analysis and control design in the framework of linear system theory due to its property of the structure modeling. It has been confirmed that the model can effectively be used for fault detection of nonlinear systems through simulation studies on the ship propulsion plant model, which was proposed as the plant model for benchma … More rk test of fault diagnosis problems. Here in the plant model, various fault mode could be realized as unexpected abrupt changes in plant configuration parameters.2.Feature Extraction for Fault Detection based on the Quasi-ARMAX ModelParameters of the Quasi-ARMAX Model could be estimated by using existing recursive identification scheme, e.g. prediction error method, and the model identified during the normal operating period was used as the reference model for fault detection. The Kullback Discrimination Information (KDI) was introduced as the index of fault detection. The KDI is a distortion measure between two identified models, i.e. the reference model obtained under normal operation and the on-line identified plant model during the monitored period. In order to establish the fault detection system with high performance, we have proposed several improvement schemes to feature extraction procedures based on the Quasi-ARMAX modeling and identification. The effectiveness of the schemes has been verified for various fault modes through simulation studies on the ship propulsion benchmark system.3.Realization of A FDI (Fault Detection and Isolation) SystemThe Quasi-ARMAX model is essentially an input-output type mathematical model for representing unknown system, therefore its parameters have not any physical information about the system structure. However the model has multi-linear form with weighting factors, so features of fault modes occurred in the plant due to changes in the configuration parameters may be reflected in some degree to the identified model parameters. Based on this idea, we have realized a fault isolation function in our fault detection system. After the fault detection, the fault isolation could be performed by using pattern recognition method in the feature space consisting of model parameters, in which reference feature sets corresponding to typical fault modes were constructed based on a prior knowledge about their fault modes. In this way we developed a model based FDI system for black box nonlinear systems and its effectiveness has been confirmed via the simulation studies. In these studies, we also confirmed that several methods of pattern clustering and recognition developed by co-workers of this research project could effectively be used to construct the FDI system. Less