Abstract: Support vector machines (SVM) can be used for novelty detection. This paper applies SVM to detect the faults of PW4056 engine. First, three deviations of engine cruise data are analyzed, then short term deviation is chosen as the input of the detection model. Second, model selection is conducted using validation method, and the effect of model parameters on detection accuracy is analyzed. Finally, the result of detection model is studied and novelty index is defined to measure the severity of fault. The training and validation of detection model use engine run data, and the study shows that this model is feasible and the accuracy reaches 90%. However, the quality of data must be improved if the better result is wanted.

Key words: aerospace propulsion system, fault detection, support vector machines, hyper-sphere model