Research on anomaly detection method of aero engine blade based on HOG and SVM
基于HOG与SVM的航空发动机叶片的异常检测方法研究

摘要

可靠性和安全性是航空发动机的重要发展方向，而涡轮叶片是重要的航空发动机热端部件，有效的叶片诊断方法能够提前预警航空发动机的异常状态，降低维护成本，减少重大灾难发生的风险。本文基于HOG与SVM，提出了一种航空发动机涡轮叶片异常检测的方法。首先，通过实验获取航空发动机叶片高温状态下的热成像图片。然后，通过HOG将航空发动机涡轮叶片的高温热成像图将进行梯度直方图处理，得到叶片热成像图片的低维特征向量。然后将特征向量输入SVM模型训练得到叶片异常检测分类器。最终，使用测试图片对异常检测分类器效果进行验证，在小样本情况下，能够达到100%的识别准确率。本文所提出的基于HOG与SVM的航空发动机叶片异常检测方法，为航空发动机故障诊断及涡轮叶片异常检测的智能化提供了参考。

Abstract

Reliability and safety are important development directions of aero engines, and turbine blades are important hot end components of aero engines. Effective blade diagnosis methods can warn of abnormal state of aero engines in advance, reduce maintenance costs, and reduce the risk of major disasters. In this paper, based on HOG and SVM, a method of aero engine turbine blade anomaly detection is proposed. Firstly, thermal imaging images of aero engine blades at high temperature were obtained through experiments. Then, the high temperature thermal image of aero engine turbine blade is processed by gradient histogram through HOG, and the low-dimensional feature vector of the thermal image is obtained. And then the feature vector is input into SVM model to train the blade anomaly detection classifier. Finally, the test images are used to verify the effect of the anomaly detection classifier, and the recognition accuracy rate can reach 100% in the case of small samples. The method of aero engine blade anomaly detection based on HOG and SVM proposed in this paper provides a reference for aero engine fault diagnosis and intelligent turbine blade anomaly detection.