为了实现产品的自动装配，智能机器人装配系统除了需要关于产品的信息之外，还需要关于机械手、装配工具、周边装置等几何和物理特征，工作空间布局等大量信息及装配工艺的经验知识。这些数据和知识不仅种类繁杂，数据量大，冗余度高，而且难以量化，因而使其难为应用程度有效地存储、存取和充分利用。如何有效地建模、存储和管理这些数据知识，为装配规划、调度、系统仿真及实时执行提供必要的信息，是我们面临的主要问题。在现有工作基础上，我们建立了一个智能数据库管理系统，该工作可分为以下六个方面：1) 参照DBMS的层次划分、装配系统自身的特点以及我们以前工作的基础，我们将数据模型分为语义模型、实体模型、存储模型和程序设计模型四个不同的层次。该数据模型的特点是：即能方便用户建模现实世界的复杂对象，又能反映系统设计者的意图，即从存储管理层来扩展关系数据库系统，使之具有直接支持复杂对象的存储和存取能力，同时还能方便应用程序的存取，可以认为该模型为装配数据和知识建模、存储和利用提供了第一的框架。2) 给出了装配智能数据库管理系统的体系结构，并分析了它与Pro/Engineer几何造型系统、SAROT原型装配系统的关系及其工作原理，体现了灵活的Client/Server体系结构的优越性。该系统可分为接口、装配数据库和核心功能模块三部分。3) 接口分为用户接口、身体信息建模工具和应用程序接口。其中，用户接口又分为数据定义语言(DDL)和数据操作语言(DML)。我们参照EXPRESS语言和SQL语言，分别设计了类EXPRESS语言和扩展SQL语言，作为该系统的DDL和DML。选用Pro/Engineer作为提供几何图形和特征的造型系统。文章重点讨论了类EXPRESS语言的设计及其处理，它与实体模型的语义是一致的。4) 装配数据库分为数据字典、数据库和规则库三部分，文章讨论了三者的存储结构及其组织结构关系，分析了复杂对象（包括几何图形和几何特征）及其语义的存储。5) 核心功能模块的设计是装配智能数据库管理系统设计最为复杂的一部分，是实现物理数据独立性的关键。它分为事务管理、规则管理、对象管理和存储管理四部分。我们主要开展了下列工作：·着重讨论了相应的并发控制和事务调度算法；·给出了复杂对象推理的框架，并证明了转化后的逻辑系统与一阶谓词逻辑的等价性；·讨论了对象操作算子----扩展代数操作算法及对象合并、分解原理；·存储管理模块考虑了对多种存取路径的支持，索引的建立与维护，以及数据缓冲区的管理。文章重点讨论了存储空间的分配/回收算法。6)文章最后讨论了三个应用实例，说明了建立智能数据库系统的必要性、可行性及其优越性。

During the automatic assembly of products, the intelligent robotic assembly systems need not only the information for products, but also great amount of other related data and knowledge, such as the geometric and physical properties of manipulators, assembly tools and equipment, information of work space, and the empirical knowledge about assembly technology. Since these data and knowledge have the following characteristics: mixed, great amount, much redundancy, and difficult for quantification, they are difficult to be stored, accessed and utilized in applications. Therefore, the main problem facing us is how to model, store and manage these data and knowledge, so as to supply necessary information for assembly planning, scheduling, graphic simulation and real-time execution. The main work described in this dissertation is to build an intelligent database management system for intelligent robotic assembly systems (ADBMS) on the basis of our previous work, which is summarized as follows:1) Referring to classification of levels of DBMS, and considering the characteristics of assembly systems and our previous work, a data model is carefully designed, which can be divided into four different levels: semantic model, entity model, storage model and programming model. It is convenient both for users to model the complex objects of real world, and for applications to access these data and knowledge stored in the assembly database; and at the same time, it also reflects the intentions of designers, i.e.extending a relational database system in the storage management level to directly support the storage and access of complex objects. So, we can say, the model provides a unified framework to manage these assembly data and knowledge.2) A system architecture of ADBMS is proposed, whose working principle and the relationship with Pro/Engineer solid modeling package and SAROT prototype assembly system are analyzed. It reflects the advantages of flexible Client/Server architecture.The system consists of three components: interface, assembly database and kernel function module.3) Interface can be classified into user interface(UI), geometric information modeling auxiliary tool and application programming interface(API). UI consists of two components: data definition language(DDL) and data manipulation language(DML). Referring to the standard EXPRESS and SQL language as DML. Pro/Engineer solid modeling package is selected for providing geometric images and features for assembly systems. The design and processing of DDL is discussed in detail, which is consistent with the entity model.4) An assembly database includes three components: data dictionary, database and rule base. The storage structures and the physical organization of them are discussed carefully. Especially, the storage of complex objects (including geometric images and features) and their semantics is analyzed.5) Kernel function module is one of the most complicated components of ADBMS, which is the key to realize the physical data independence. It consists of four components, the transaction management, rule management, object management and storage management. We have done the following work:·The concurrency control and transaction scheduling algorithms are proposed;·the framework for complex object reasoning is given, and equivalence between the logic system mapped from symbol object model and first-order predicate logic is proven;·the object operators, i.e.algorithms of extended algebra operations, and the principle of merging and decomposition of objects are discussed in detail;·the principle of supporting multiple access paths, the creation and maintenance of indices, and the management of data buffers are analyzed. We especially focus on discussing the algorithms of the allocation and collection of storage space.6) Three application examples are given, which show the necessity, feasibility and advantages of building intelligent database systems for intelligent robotic assembly systems.