基于ART—2神经网络的故障诊断系统

对ART－2模拟量模式识别神经网络算法的改进能使其有效地适用于故障诊断，神经网络的预处理场F0具有增强对比度，抑制基底噪声的能力，神经网络的F2场根据优先级选择竞争节点。网络模式识别的结果作为专家系统输入的类别及其选取概率。专家系统搜索由知识库构造的关系树并依规则进行推理。本文方法结合神经网络和专家系统的优点，以得到可信的故障诊断结果。本文还给出了验证系统正确性的自动液压机故障诊断实例。

FAULT DIAGNOSIS SYSTEM BASED ON ART-2 NEURAL NETWORK

The algorithm of ART 2 analog pattern recognition neural network is improved for efficient fault diagnosis. The preprocessing field F 0 of the network has the ability of enchancing contrast and attenuating base noises. The active F 2 field of the network is designed to select competitive nodes according to their priorities. The pattern recognition results of the network are some classes together with the choice probabilities which are the inputs of expert system. The expert system works on way of searching the relationship trees constructed of knowledge base and reasoning by the rules. The method presented in this paper combines the advantages of neural network and expert system to achieve believable results of fault diagnosis. Diagnosis examples of automatic hydraulic press to verify the system are also given in the paper.