烧结过程异常状况诊断专家系统SPADES(Ⅱ) --关键技术

知识库构造和推理机设计是实现专家系统的关键技术.烧结过程异常诊断的知识具有模糊性、不确定性与因果性强,定性知识与定量知识并存的特点.采用模糊逻辑表达烧结过程信息的模糊特性,研究了表达烧结过程异常诊断知识的综合型知识表达方式.知识库的组织包括数据库、事实库、模型库和规则库.规则库中的规则根据用途不同分组存放,同组规则按优先级别排序,这样的知识组织方式为实现高效快速推理奠定了基础;SPADES系统的知识库具有管理与维护功能,知识学习采用专家指导学习方式;烧结过程异常诊断专家系统的推理机采用多级目标推理的混合推理机总体目标推理采用过程化推理,按异常类型判断、异常原因分析、操作决策顺序进行;烧结过程异常状况的诊断采用模糊诊断确定假定异常集合,再通过反向推理验证假定异常集合中的元素,确定出现的异常类型;异常原因分析和操作决策采用正向推理.对于本系统中模糊规则的模糊性传递问题,采用基于确定因子法的不精确推理策略,以及基于模糊规则的模糊匹配方法.SPADES系统在生产现场的应用结果证明了本系统知识表达方式和推理机可以实现模糊知识的表达和高效推理.

Expert system for abnormity diagnosis in sintering process SPADES( Ⅱ ) -Key techniques

The construction of knowledge base and inference engine are key techniques in realizing an expert system. The knowledge for diagnose abnormalities in sintering process has characteristics of fuzzy,uncertainty,causality,both qualitative and quantitative. In this paper,fuzzy logic was used to represent the fuzzy characteristic and the hybrid knowledge representation method was studied.The multi base and multi hierarchy knowledge base including data base,evidence base,model base and rule base,the rules in the same base were organized to several groups according to their usefulness,the rules in the same group were lined up by their priority. Knowledge base managed in such a way was ready for rapid inference.The SPADES system has management and maintenance module and learns new knowledge with expert guiding.The universal multi goal inference model and hybrid inference strategy was put forward according to the characteristics of hybrid knowledge representation and the multi base and multi hierarchy structure of knowledge base,the general inference goal including abnormity type judging,abnormity cause analyzing and decision making were realized by procedure reasoning.The supposed abnormities were judged by fuzzy diagnosis and proved by backward reasoning,the factors that give rise to the abnormities and control decision were inferred by forward reasoning.The strategy of reasoning with uncertainty of fuzzy production rules and fuzzy matching were also investigated. The proper representation and effective inference of fuzzy knowledge are realized in the developed expert system and successfully applied to practice.