基于支持向量机决策树的航空发动机轴心轨迹识别方法

针对航空发动机转子轴心轨迹难以准确自动识别的问题，提出了基于二维形状不变矩和支持向量机（SVM）决策树的识别方法。对信号滤波降噪和倍频提纯，形成比较清晰的轴心轨迹；利用二维形状不变矩提取轴心轨迹的图形特征，得到不变矩特征向量，进而构造特征故障的训练和测试样本；采用SVM进行训练和学习，构造SVM决策树，识别故障类别，分类正确率达93.3%以上。应用实测弹支振动应力信号对该方法的准确性进行了验证，结果表明，该方法有效地解决了航空发动机转子轴心轨迹自动识别准确率低和小样本问题。

Shift Orbit Recognition Method of Aero-engines Based on SVM Decision Tree

It was difficult to automatically and exactly identify shift orbits of aero-engine rotors, a  recognition method was proposed based on two-dimensional feature invariant moments and SVM decision tree. First, the relatively clear shift orbits could be acquired after the noise reduction and frequency purification of the signals. The invariant moments were used to extract the graphic features of the shift orbits, then the invariant moment vectors were obtained, constructing the training and test samples of the characteristic faults. Finally, in order to construct SVM decision tree and identify faults, support vector machine was trained and studied. The accuracy rate of classification reaches 93.3%. The method was verified by measured signals of elastic support vibration stress. The results show that the method effectively resolve low recognition accuracy and small sample problems for shaft orbits of aero-engine rotors.