基于特征融合技术的发动机故障诊断

Fault diagnosis for engine based on feature fusion

为了提高发动机的故障识别率,设计了一种将B&B算法与广义辨别分析(GDA)相结合的多类特征融合方法。从发动机转子的振动信号中提取出频谱特征集和纹理特征集,用B&B算法删去2类特征集中信息量少的特征,并用GDA和支持向量机(SVM)分类器进行特征融合和分类识别。发动机的转子故障试验结果表明,该方法获得的融合特征包含有更多的类别信息,用于转子故障获得的识别率为98.21%,且不受分类器核参数的影响;而频谱特征、纹理特征输入SVM分类器后获得的故障识别率仅为92.86%和89.29%。该研究为发动机的故障诊断提供了一种有效、实用的特征提取方法。

In order to achieve higher fault recognition rate of engine,the paper proposed a multi-class feature fusion method which combined BB algorithm with generalized discriminant analysis（GDA） . Firstly,the spectrum feature set and texture feature set were extracted from the vibration signal of engine rotor. Subsequently,BB algorithm was used to remove the information-lacked features from these feature sets. Finally,the GDA and SVM classifier were used to implement feature fusion and fault recognition. The experiment results indicated that this method can make the fused features contain more category information,and it can reach 98.21% of fault recognition rate for engine rotor fault diagnosis,moreover,it was almost free from the kernel parameter of support vector machine（SVM） . While the spectrum features and texture features were directly inputted to SVM classifier,the fault recognition rate can be reached to only 92.86% and 89.29%,respectively. This study provides an effective and useful feature extraction method for engine fault diagnosis.