Abstract: There is high complexity and risk in caprolactam plant，so it is necessary to develop an online safety operation guidance system with fault diagnosis as core，which can help operator’s operation. In this paper，a fault detection and diagnosis model for the entire caprolactam plant was built based on qualitative and quantitative fault diagnosis methods，such as signed directed graph (SDG)，fuzzy logic，principal component analysis (PCA)，Artificial Neural Network(ANN) and expert system. Based on the characteristics of process and failure mode of caprolactam plant，the expert knowledge database with root reason，propagation path and treatment measures of the fault was built by combining Hazard and Operability Analysis results with expert experience. At last，a Caprolactam Plant Safety Operation Guidance System was established and used in the caprolactam plant with good performance.

Key words: signed directed graph, model, principle component analysis, artificial neural network, expert system, hazard and operability analysis

摘要： 由于己内酰胺装置具有高复杂性及高危险性，有必要开发一套在线诊断故障的安全运行指导系统，辅助操作人员的操作。本文基于定性的符号有向图（signed directed graph，SDG）、专家系统，结合模糊逻辑、主元分析（principle component analysis，PCA）、神经网络（artificial neural network，ANN）等多种定量故障诊断方法，建立了整个装置的异常监测与诊断模型；根据己内酰胺装置的工艺特点及故障模式，结合危险与可操作性分析结果和专家经验，建立了导致装置发生故障的原因、传播路径以及处理措施的专家知识库；开发了己内酰胺装置安全稳定运行指导系统，并在己内酰胺装置上进行了工业应用，取得了良好效果。

关键词: 符号有向图, 模型, 主元分析, 神经网络, 专家系统, 危险与可操作性分析