Fundamental Study on Safety Assessment of Intelligent and Integrated Systems

本研究的主要目的是发展一种方法论来评估利用高度复杂的信息处理技术操作和控制的智能系统的安全性。在项目的第一年，由于智能系统的可靠性和安全性取决于其存储知识的正确性，因此在基于规则的故障诊断系统中，开发了两种维护存储规则可靠性的支持方法。在第二年，基于错误风险准则，开发并研究了几种动态和静态系统的故障诊断方法。其中一种方法考虑了1)检测项目的监测值序列，2)监测值的误差，3)系统的可靠性特征，4)系统异常事件造成的损失的主观评估。第三年，重点关注操作和决策中的人为因素和人为错误传播过程。首先，针对层次分析法(AHP)中广泛使用的两两比较矩阵，提出了一种新的一致性指数，以求取方案的相对重要性。通过对AHP过程的计算机模拟，与传统指标进行比较，检验了该指标的性质。其次，为了分析人为错误对执行给定任务的多个成员团队的影响，提出了一个基于人类行为模型的一般框架，并以一个三人团队的加油任务为例说明了所提出的方法的细节。第三，讨论了由主要决策者及其下属组成的层次化团队的决策。在利用期望误差风险准则推导出一般决策规则之后，详细分析了它们的性质。

The main objective of this research is to develop methodology to assess the safety of intelligent systems which are operated and controlled by use of highly sophisticated information processing technology.In the first year of the project, since the reliability and safety of the intelligent systems depend on the correctness of their stored knowledge, two types of supporting methods to maintain the reliability of stored ruyes in rule-based failure diagnosis systems are developed.In the second year, several failure diagnosis methods for dynamic and static systems are developed and investigated based on an error-risk criterion. One of the methods developed, for example, considers 1) sequences of monitored values of inspection items, 2) errors in monitored values, 3) reliability characteristics of the system, and 4) subjective evaluation of losses caused by system abnormal events.In the third year, human factors together with human error propagation processes in operation and decision making are focused on. Firstly, a new consistency index is proposed to the pair-wise comparison matrix which is extensively used in the Analytic Hierarchy Processes (AHP) for obtaining the relative importance of alternatives. By computer simulation of AHP processes, the properties of the proposed index are examined in comparison with the conventional index. Secondly, to analyze the effect of human errors in a team of multiple members who execute a given task, a general framework based on a human behavior model is proposed and an illustrative example of a tank-filling task for a team of three members is made to show the details of the proposed method. Thirdly, decision making by a hierarchical team which consists of a primary decision maker and his subordinates is discussed. After general decision making rules are derived by use of an expected error-risk criterion, their properties are exploited in detail.