摘要
为保障飞机的飞行安全,做到预防性维修,提升飞机的飞行安全及任务出勤率,需要对飞机结构出现的疲劳裂纹进行及时检测并修理。基于支持向量机理论,建立了支持向量机回归预测模型,并应用该模型对B737飞机水平尾翼健康信息的特征值(小波包分解系数提取的能量)进行了故障预测研究。为建立最佳支持向量机模型,选用了支持向量机四种常用的核函数分别对特征值进行了预测。同时还对支持向量机预测模型与神经网络预测模型(BP神经网络预测模型)的预测结果进行了比较与分析,研究表明,应用支持向量机所设计的预测模型准确率比较高,可以较好地对飞机水平尾翼的裂纹故障进行预测。
In order to protect the safety of the aircraft flights and to do preventive maintenance,upgrading of aircraft flight safety and mission attendance,we need to predict the aircraft stabilizer crack fault which may occur during the work time.This paper illustrates the related knowledge about statistical learning theory and support vector machine(SVM),and designs a SVM regression prediction model based on SVM theory,and we apply the model to do failure prediction research work about the characteristics(wavelet packet characteristics from the energy factor which contain the health information) of the B737 aircraft stabilizer.In order to choose the best SVM model,we use four kind of nuclear functions which are commonly used in support vector machines to do failure prediction research work.In order to illustrate the merit of the SVM,we also design the BP neural network forecasting model to forecast the characteristics,and we compare and analyze the result of SVM model and neural network forecasting model(BP neural network forecasting model).The experiments show that the application model based on support vector machines theory has a higher accurately forecast rate than others,it can achieve a better level of the prediction of aircraft stabilizer crack fault.
出处
《中国民航大学学报》
CAS
2012年第3期37-41,46,共6页
Journal of Civil Aviation University of China
关键词
水平尾翼
故障预测
支持向量机
核函数
神经网络
horizontal stabilizer
failure prediction
SVM
nuclear function
neural network
