Fault diagnosis for engine based on feature fusion

In order to achieve higher fault recognition rate of engine, the paper proposed a multi-class feature fusion method which combined B&B 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, B&B 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.