开放环境下的辩论诊断方法及其应用研究

Distributed dynamic software systems in open environments (e.g. software agent and Web service) are a sort of special Distributed Discrete Event System (DDES) that features dynamic, autonomy and mass participation. With the interaction of distributed components, fault can be carried and propagated to a remote component and manifests itself there, even worse in open environments where mass contribute to building blocks. Such DDESs pose new challenges to current software diagnosis techniques, in particular Model-Based Diagnosis (MBD), because they often demand for precise functioning models of the system and/or its components, or even have access to the source code where some marks might be added to control the execution flow. However, these needs might be an issue as the pre-defined fault modes, complete normal/abnormal behavior model, and the access to the agent’s or service’s source code are frequently impossible or undesirable in practice. Therefore, it is crucial to tackle issues of diagnosing an exception in dynamic open environments..The goal of this project is to design a framework for diagnosing distributed dynamic software systems without relying on a priori knowledge of their structures or interactions. Relaxing the restriction of traditional fault diagnoses on complete behavior model, this project thinks of fault diagnosis as an argument game among suspected components, and thus explores the theory and algorithms of argumentation-based distributed diagnosis based on learned model and man-made partial model. To this end, this project mainly investigates (1) the diagnosis theory and method for DDESs, which include argumentation diagnosis framework and its semantics, diagnosis models, uncertainty in diagnosing, and diagnosability, and (2) the application of argumentation-based fault diagnosis to services and agents, which includes model generation, diagnosis algorithms, argumentation and distributed diagnosis. With the increasing demand of safe and trustworthy distributed dynamic software systems, our project works on a fundamental research topic and has not only an important research meaning but also a great value for software quality.

开放环境下的分布式动态软件系统，如软件Agent、Web服务，是一类特殊的分布式离散事件系统（DDESs），其动态性、自主性、广泛参与性对当前软件诊断技术，特别是基于模型诊断，提出了诸多挑战。针对这类DDESs难以预先定义故障模式、难以定义完备的正常或异常行为模型、没有组件源代码可用等问题，本项目将放宽传统故障诊断的模型完备假设，在学习产生的隐态模型和人工创建的部分模型的基础上探讨具有辩论协商特点的分布式诊断理论和算法。为此本项目主要研究：（1）辩论诊断理论和方法，包括辩论诊断框架及其语义、诊断模型、诊断不确定性和可诊断性；（2）辩论诊断在分布式动态软件系统中的应用研究，通过模型生成、诊断算法、分布式辩论诊断系统等方面的工作而建立基于辩论的服务故障的诊断方法。在日益凸显的分布式动态软件系统的可靠性和安全性需求下，本课题是一个前沿性的应用基础研究工作，具有深层次的研究意义和重要的应用前景。

开放环境下的分布式动态软件系统，如软件Agent、Web服务，对当前软件诊断技术提出了诸多挑战。针对其难以预先定义故障模式、难以定义完备的正常或异常行为模型、没有组件源代码可用等问题，本项目放宽传统故障诊断的模型完备假设，在学习产生的隐态模型和人工创建的部分模型的基础上探讨具有辩论协商特点的分布式诊断理论和算法。为此本项目主要研究：（1）辩论诊断理论和方法，包括辩论诊断框架及其语义、诊断模型、诊断不确定性和可诊断性；（2）辩论诊断在分布式动态软件系统中的应用研究，通过模型生成、诊断算法、分布式辩论诊断系统等方面的工作而建立基于辩论的服务故障的诊断方法。.本项目围绕相关研究内容开展了深入的研究，在论文发表方面取得了丰硕的研究成果，主要是在IEEE Trans国际权威期刊发表高水平论文2篇、CCF A类会议1篇、SCI检索期刊论文27篇、热点论文1篇、高被引论文4篇。在研究成果论著当中，项目负责人作为第一作者的论文4篇，作为通讯作者的论文22篇。在专利申请方面，项目组设计开发了一系列关于故障预测、Bug指派、质量优化以及健康诊断与管理策略，申请提交了中国专利6项。.本项目基于理论研究成果，围绕故障预测与健康管理技术的实际需求，基于多Agent架构研发了光学装备健康管理与维护软件系统，应用于中国科学院长春光学精密机械与物理研究所真实装备健康诊断应用之中，特别在军队光学装备保障维护业务场景之中实施部署并落地应用，使其技术保障能力显著提高。