基于自组织模糊神经网络的汽轮发电机组振动故障诊断系统

针对目前汽轮发电机组振动故障的模糊诊断系统精度较低的问题,将模糊理论与神经网络技术结合,构造了适合于大型复杂系统故障诊断的自组织模糊神经网络系统的体系结构.在此基础上,提出了一种新型的样本模糊自组织方式,利用快速傅里叶变换(FFT)分析方法和图形分析法提取足够多的故障征兆,再采用聚焦式模糊分段算法对故障征兆进行模糊化处理,然后建立合理的训练样本库,并将经过确认的故障数据增加到标准案例库中,以提高整个系统的诊断能力.其中,聚焦式模糊分段算法对故障征兆中正常与不正常的转折部分进行非线性聚焦,即离正常与不正常的分界值越近时,故障征兆的特征抽取密度越大,使得原来模糊的分界部分被清晰化,大大提高了诊断精度.最后,以108DAI专用检测系统作为硬件支持,设计和实现了600 MW汽轮发电机组常见振动故障的模糊诊断系统,并利用现场故障数据验证了该系统的有效性.

Fault Diagnosing System for Turbo-generator Unit Based on Self-organized Fuzzy Neural Network

In view of the lower accuracy of the vibration fault diagnosing system for turbo-generator units, a self-organized fuzzy neural network system is built by combining the fuzzy set theory with the neural network technology. In particular, an effective fuzzy self-organized method for training samples of the neural network is presented. The fault symptoms are firstly extracted from fault signals via the FFT algorithm and graphics analysis, discretized by a focusing quantization algorithm and then fuzzified by means of the given fuzzy sets. The standard sample database for the diagnosing neural network is established. Because the focus center of the focusing quantization algorithm is placed in the transition area from a normal state to an abnormal one of fault symptoms, the resolution near the focus should be enhanced so that the diagnostic accuracy of the fuzzy neural network diagnosing system can be appreciably improved. In addition, the fault data confirmed by actual application is added to the standard fault case database in order to improve the accuracy of the diagnosing system as a whole. Finally, supported by a special detecting system, a vibration fault diagnosis system for a 600 MW turbo-generator unit is designed and implemented. More importantly, its running results in a thermal power plant of Guangdong Province show that the new diagnosing system can meet the requirements of fault diagnosis of large turbo-generator units with the accuracy varying from 96% to 100%.