An Aeroengine Gas Path Anomaly Detection Method in The Case of Sample Imbalance

Abstract

In the process of aeroengine anomaly detection, there is always an unbalance distribution among the samples of gas path performance parameters, that is, the number of normal samples is much larger than the number of abnormal samples. In addition, this imbalance will worsen with time, which leads to the classifier paying too much attention to normal samples in the process of model training. Thus, the recognition rate of abnormal samples will reduce significantly. To solve the above problems, an adaptive decision threshold support vector machine (ADT-SVM) is proposed and applied to the anomaly detection of aeroengine. Firstly, this paper analyzes the influence of the unbalanced training data on the performance of the traditional classification model. Then the concept of decision threshold is introduced and introduced into support vector machine for anomaly detection. Finally, an adaptive method is proposed to calculate the decision threshold based on the equal expected number of samples, and the performance of the adaptive threshold and the traditional default threshold SVM is compared through experiments, which show that the adaptive threshold is effective in solving the problem of the classification performance degradation of unbalanced gas path performance parameters.