Design Logic for Construction of Dependable Control Systems

控制系统的设计要求已从高速、高精度转向实现系统的高可靠性和安全性。此外，系统与人的协调配合也成为控制系统设计中必须考虑的重要因素。从这些动机出发，本研究项目解决了以下问题：1。基于概率信息理论的故障检测与诊断2.分布式故障诊断主要针对大规模系统3.基于混合动力系统的人类行为建模在主题1中，我们提出了两个新的基于概率信息论的数学模型。第一种是一种新的隐马尔可夫模型（HMM），其中一个连续的动态模型分配给每个离散状态的HMM，我们应用该模型的故障诊断。第二个是概率产生式规则，其中概率被分配给传统的产生式规则，我们应用 ...更多信息 将该理论应用于机械手控制系统的故障诊断。此外，本文还提出了一种基于时间马尔可夫模型的故障诊断算法，并以自动化生产线为原型进行了实验验证。在课题2中，本文提出了一种基于贝叶斯网络理论的分布式诊断算法，能够结合局部诊断结果对整个系统进行诊断。在该框架中，子系统中的故障条件和观测信号之间的关系被建模的网络结构。通过将该算法应用于一个自动化生产线的原型系统，验证了该算法的有效性。在课题3中，我们采用了基于数据聚类的两种建模方法和课题1中提出的新的隐马尔可夫模型。因此，我们可以在两种情况下，离散模式之间的过渡建模确定性和随机建模的人类行为。将所提出的方法应用于驾驶行为和任务技能的分析，验证了这些方法的有效性。少

The requirements for design of control systems have been shifting from high speed and high precision to realization of high reliability and safety of the systems. Furthermore, harmonious corporation between systems and human is becoming an important factor which must be considered in the control system design. From these motivations, this research project addressed the following topics:1. Fault detection and diagnosis based on probabilistic information theory2. Distributed fault diagnosis focused on large scale systems3. Modeling human behavior based on hybrid dynamical systemsIn the topic 1, we proposed two new mathematical models based on probabilistic information theory. The first one is a new Hidden Markov Model (HMM) wherein a continuous dynamical model is allocated to each discrete state of HMM, and we applied the model to the fault diagnosis. The second one is a Probabilistic Production Rules wherein probabilities are assigned to the conventional production rules, and we applied … More the theory to the fault diagnosis of the manipulator control system. In addition, we proposed a fault diagnosis algorithm based on Timed Markov Model, and verified the availability by experiments using a prototype of automating transfer line.In the topic 2, we proposed a new distributed diagnosis algorithm capable of diagnosing the overall system behavior by combining the local diagnosis results based on Bayesian network theory. In the proposed framework, the relationships between the faulty conditions in the subsystems and the observed signals are modeled by network structure. By applying the algorithm to a prototype of automating transfer line, we verified the availability of the algorithm.In the topic 3, we employed two modeling approaches based on data-clustering and the new HMM developed in the topic 1. As the result, we can model human behaviors in both cases that the transition between the discrete modes is modeled deterministically and stochastically. The availabilities of these approaches were verified by applying the proposed methods to analysis of driving-behaviors and task skills.These achievements are or will be published in scientific journals. Less