基于SA-2DCNN的涡轮叶片故障诊断方法

FAULT DIAGNOSIS METHOD OF GAS TURBINE BLADES BASED ON SA-2DCNN

为提高燃气轮机故障诊断中预测模型的准确度,提出一种结合了自注意力(Self-Attention,SA)机制的二维卷积神经网络(Two-Dimensional Convolutional Neural Network,2DCNN)诊断方法。相比传统的故障特征提取和样本分类两阶段模式,该方法将两者合二为一:将涡轮传感器的振动信号转为格拉姆角场(Gramian Angular Field,GAF),实现振动信号从一维序列到二维图像的变换,利用SA-2DCNN的卷积层、池化层、注意力层和全连接层分别进行特征构建和样本分类,得到与振动信号对应的故障类型。以某电厂燃气轮机数据进行实验,结果表明SA-2DCNN模型具有强大的特征提取能力,分类准确率达到95.1%,相比传统模型提升了0.05~0.10,能更好地应用于叶片故障诊断。

In order to improve the accuracy of prediction model in gas turbine fault diagnosis,a fault diagnosis method based on two-dimensional convolutional neural network(2DCNN)combining self-attention(SA)mechanism is proposed.Compared with the traditional two-stage pattern of fault feature extraction and sample classification,this method combined the two into one.The vibration signal of turbine sensor was transformed to the Gramian angular field(GAF),which converted the signal from sequence to image.Through convolution layer,pooling layer,attention layer and full connection layer of SA-2DCNN,the feature was extracted and the sample was classified,and the fault type corresponding to the vibration signal was obtained.The experiment was performed on gas turbine data from a certain power plant.The results show that this method has powerful feature extraction capabilities.Its classification accuracy achieves 95.1%,which is 5%-10%higher than that of traditional models,which can be better applied to gas turbine fault diagnosis.