故障诊断是人工智能的一个重要研究内容．分析了现有故障诊断方法（基于规则、基于模型、基于人工神经网、基于案例）的各自特点和局限性．提出了集成现有故障诊断方法（规则、模型、人工神经网、案例）的故障诊断系统及其系统结构．从知识获取和表示、推理和诊断、机器学习方面，说明该新系统能结合现有故障诊断方法的互补性并克服存在的局限性，并且能提高故障诊断的智能性及效率． Fault diagnosis is an important research content of artificial intelligence. The respective characteristics and limitations of the existing fault diagnosis methods (rule-based, model-based, artificial neural network based, case-based) are analyzed. A fault diagnosis system integrating the existing fault diagnosis methods (rules, models, artificial neural networks, cases) and its system structure are proposed. From the aspects of knowledge acquisition and representation, reasoning and diagnosis, and machine learning, this new system can combine the complementarity of existing fault diagnosis methods and overcome the existing limitations, and can improve the intelligence and efficiency of fault diagnosis.