论文部分内容阅读
航空发动机故障诊断中一个有挑战性的难题就是如何处理具有高维数、非线性化特点的故障数据,传统模式识别方法很难发现这类数据集的真实结构,导致故障诊断准确性不高。针对这一问题,将一种新兴的非线性维数约简技术——流形学习引入航空发动机振动故障诊断,提出基于监督流形学习理论的航空发动机特征提取与识别方法。该方法首先采用最近兴起的监督局部线性嵌入流形学习算法对蕴含在高维振动故障数据中不同故障的流形特征进行学习,映射到低维嵌入空间以实现故障的特征提取,在降维后的流形特征空间中构造分类器实现故障识别。利用航空发动机转子故障数据对方法的有效性进行了验证,结果表明,该方法显著提高了故障诊断性能,克服了传统的模式识别方法PCA和LDA的不足,并且在训练样本数为每类100的条件下,该方法的平均故障诊断正确率比PCA和LDA分别高出2.93%和7.20%。
A challenging challenge in aero-engine fault diagnosis is how to deal with fault data with high dimension and non-linear characteristics. The traditional pattern recognition method can hardly find the true structure of such a data set, which leads to fault diagnosis accuracy is not high. In response to this problem, a new nonlinear dimensionality reduction technique, manifold learning, is introduced into the diagnosis of aeroengine vibration. Aeroengine feature extraction and recognition method based on supervised manifold learning theory is proposed. This method first uses the newly arisen supervised local linear embedding manifold learning algorithm to learn the manifold features of different faults contained in the high-dimensional vibration fault data and maps them to the low-dimensional embedding space to achieve fault feature extraction. After the dimensionality reduction The manifold classifier is constructed to realize the fault identification. The effectiveness of the method is verified by using the data of aeroengine rotor fault. The results show that this method can significantly improve the performance of fault diagnosis, overcome the shortcomings of traditional pattern recognition methods such as PCA and LDA. When the number of training samples is 100 , The average fault diagnosis accuracy of this method is 2.93% and 7.20% higher than that of PCA and LDA, respectively.