基于多神经网络协同推理的故障诊断系统的研制

DEVELOPMENT OF FAULT DIAGNOSIS SYSTEM BASED ON MULTI NEURAL NETWORK JOINTED INFERENCE

神经网络系统与传统专家系统相比，具有其优势也有其不足。由于其不足，在建立神经网络系统时，就必须首先解决一些具体的实现问题。在神经网络诊断系统中，我们采用了网络分块技术和多网络协同推理技术，同时也考虑了系统的解释功能。网络分块技术和“基于实例”的解释大大增强了系统的解释功能。在解决了具体的实现问题之后，采用先进的Ｗｉｎｄｏｗｓ 编程方法开发了神经网络故障诊断系统。本文较详细地介绍了系统的总体结构和模块功能。系统的模拟测试和现场考核表明，诊断效果不亚于基于规则的诊断专家系统，在推理速度。

Neural network system, compared with traditional expert system, has predominance, as well as its deficiency. To overcome the deficiency , a kind of self organized neural network ( Fuzzy ARTMAP ) is chosen to cooperate with the BP networks for diagnostic inference. By this way, neural networks can not only be used for forward inference , but also backward inference, and the system was provided with the ability of self study and self organization. Besides, the explanatory ability of the neural network system is also considered. Framework of the system and function of its modules are discussed in the paper. Simulating test and practical use show that the neural network system gains advantage over expert system based on rules, especially in speed of inference, acquirement of knowledge and ability of self study.