基于灵敏度特征的混杂系统在线多故障诊断与预测方法研究

This project aims at the study of online diagnosis and prognosis methods for hybrid systems with unknown mode changes and multiple faults of different detectabilities. It analyses the correlation characteristic between fault occurrence and mode change on diagnosis and prognosis results, and explores the dynamic rule of hybrid systems with unknown mode changes and multiple faults. Firstly, the trend of residual is explored to establish the fault sensitivity signature, and a reliable waiting time scheme for monitoring and an activating condition for fault estimator are designed based on the analysis of the correlation between fault occurrence and mode change, and then a planning based diagnosis and prognosis algorithm is proposed. Secondly, the mode change sensitivity signature is built and the dynamic rule of hybrid systems under multiple faults is explored under unknown mode changes, and a real-time mode tracking method is developed. Based on these findings, the dynamic interaction method between mode tracker and fault estimator is investigated, and a dual design algorithm of health monitoring and mode tracking is proposed. Finally, simulation and experimental verifications are carried out on a switched analogy circuit system and an electric vehicle steering system. The objective of this project is to introduce new analysis methods and technical means, to effectively reduce the impact of the unknown mode changes and the correlation between fault occurrence and mode change on the health monitoring system, which in turn improves the performance of diagnosis and prognosis algorithms. The study will provide theoretical and technical support for the fault diagnosis and prognosis of hybrid systems.

本项目针对具有未知模式变化和可检测性各异的多故障混杂系统开展在线诊断与预测方法研究，分析故障发生与模式变化的关联特性对诊断和预测结果的影响，探索未知模式变化和多故障下混杂系统的动态规律。首先，探寻残差变化趋势以建立故障灵敏度特征，并在分析故障发生与模式变化关联特性的基础上，设计可靠的监测等待时间方案和故障辨识器激活条件，提出基于规划的故障诊断与预测算法。其次，在未知模式变化下，构建模式变化灵敏度特征并探索多故障下混杂系统动态规律，建立在线模式跟踪方法；以此为基础，研究模式跟踪器和故障辨识器的动态交互方法，提出健康监测和模式跟踪双重设计算法。最后，结合切换模拟电路和电动车转向系统，开展仿真和实验验证。本项目旨在引入新的分析方法和技术手段，有效减少未知模式变化、故障发生与模式变化的关联等因素对健康监测系统带来的影响，提升诊断和预测算法性能。该研究将为混杂系统故障诊断与预测提供理论和技术支撑。

本项目以永久故障下混杂系统为研究对象，基于灵敏度特征方法开展多故障在线诊断与预测方法研究。混杂系统在工业生产中普遍存在，永久故障下混杂系统的故障诊断与预测是当前控制领域的研究热点，具有广泛的应用前景和实践价值。灵敏度特征方法由于考虑了残差的变化趋势，可以提供更多可区别的特征，在提高多故障隔离性方面具有一定的优势。本项目主要研究内容包括以下几个方面：首先，研究不同类型永久故障情况下统一故障建模方法，该模型不但考虑了突变故障和缓变故障，而且考虑了模式变化对缓变故障演化的影响，分析故障发生与模式变化关联特性并设计可靠的监测等待时间方案和故障辨识器激活条件，提出基于规划的故障诊断与预测算法。其次，构建模式变化灵敏度特征并探索多故障下混杂系统动态规律，利用残差变化趋势构建全局故障灵敏度特征矩阵，并将独立和非独立残差进行组合建立组合矩阵，提高永久故障下多故障隔离性能。接着，研究健康监测和模式跟踪双重设计方法，在参数不确定性情况下设计了基于模式依赖自适应阈值的鲁棒故障检测方法，减少模式变化和常数阈值引起误报和漏检等不利影响。最后，将理论成果应用到切换模拟电路和非线性机电系统中，验证研究内容的有效性和可行性。本项目按照原定计划，顺利完成了预期目标。本项目的重要成果主要体现在以下两个方面：首先，考虑了故障检测时间和模式变化时间之间的关联特性对故障辨识的影响，设计了故障辨识器激活条件和基于规划的故障诊断与预测算法；其次，利用残差变化趋势构建全局故障灵敏度特征矩阵，并将独立和非独立残差进行组合建立组合矩阵，可以提高多重永久故障下的隔离性。在本项目的资助下接受和发表了研究论文18篇（期刊论文6篇，会议论文12篇），其中包括IEEE汇刊论文5篇；授权国家发明专利2项。所研究的成果可以为永久故障下混杂系统故障诊断与预测提供理论和技术支撑。