| Endurance test is an important part of engine test and it is an essential link to ensure the quality of engine production.Due to the low frequency of fault in the durability test and the accumulation of the fault samples,it is difficult to diagnose the faults under the condition of small samples.A company with a delta analyzer as monitoring equipment of engine durability test,the instrument of the core order spectrum analysis technology has been successfully applied to the fault of the gearbox,but due to the complexity of the engine structure,there is a lack of the equipment based on mature scheme.In this paper,several classical methods of engine vibration signal fault analysis are reviewed,which are successfully applied to the engine test data.Then,the monitoring method and alarm principle of delta analyzer are introduced.According to several typical faults in durability test,the data characteristics are analyzed,and the corresponding relation between fault and order spectrum is summarized.Based on the experience and conclusion of the order spectrum analysis,this paper present a feature extraction method based on order spectrum analysis,the feature vector dimension is stable and each eigenvalue has a clear physical meaning.Finally,the machine learning method is used to identify the fault types automatically.The scheme one uses PCA dimension reduction combined with support vector machine to obtain a higher fault recognition rate at the same speed working condition,but the classification accuracy is decreased under the extended fault condition.The scheme two using the proposed feature extraction method with xgboost algorithm to improve the accuracy of classification of the fault conditions,fault identification of exhaust rocker arm and the spark plug ablation falut can reach one hundred percent,and the severity of the fault can be predicted. |
