汽轮机故障诊断中的信息熵融合

Information entropy fusion in fault diagnosis of steam turbine shaft systems

在汽轮机轴系振动故障模拟试验的基础上,对大量故障模拟试验数据进行计算,建立了典型故障的4种信息熵样本.采用概率神经网络对故障信号的4种信息熵特征进行融合研究,并将融合结果与最小距离分类器的分类效果进行了对照分析.研究表明,概率神经网络可实现对训练样本100%的正确识别率,对“陌生”样本的正确识别率也超过80%,其识别效果远远超过最小距离分类器.可见,概率神经网络综合了贝叶斯分类器和神经网络的优势,在汽轮机故障模式分类方面具有明显的优势,利用概率神经网络融合信号的信息熵特征实现汽轮机轴系故障模式识别是一种可行有效的方法.

Four information entropy samples, singular spectrum entropy, power spectrum one, wavelet energy spectrum one and wavelet space state spectrum one were calculated as information entropy data, after the faulty signals were collected from the rotor test rig. Probability neural networks （PNN） was explored to fuse the four information entropy samples. Research shows that with the advantages of Bayes classifier and neural networks, PNN have good classification ability of typical vibration faults of turbine, the classification accuracy is 100 % for training data, 80 % for fresh data. Compared with the classification accuracy of the improved minimum distance classifier, PNN have more high solutions, which is a practical fusion diagnosis method for typical fault identification of turbine rotors.