基于深度信念网络的民航发动机状态监测

Civil Aviation Engine Health Condition Monitoring Based on DBN Deep Learning Theory

摘要:

民航飞机发动机设备构造精密、复杂,其监测系统收集的数据中蕴含了丰富的故障信息；传统发动机状态诊断依靠数据统计分析和机器学习模型,但其在深入理解与归类信号特性方面的表现难以尽如人意；此外近年兴起了多层神经网络降维算法——深度学习理论,其通过模拟人脑分析过程建立由浅入深的算法模型,数据处理效果较好；将民航发动机自身特点与深度学习理论有机结合提出了基于深度信念网络发动机状态监测方法；其优势在于克服了传统方法人工提取数据特征的不确定性与状态分类陷入局部最优的缺陷,可对发动机参数进行自主学习与特征提取；实验结果表明该算法具有出色的特征提取能力与分类准确率,能够准确识别发动机的不同状态。

Abstract:

Civil aircraft engine has precise and complicated structure. The data collected by monitoring systems contain abundant fault message. Traditional methods of monitoring engine’s health condition are based on data statistics and machine learning model. However, its performance on deep-understanding and classifying characteristics of massive data didn’t meet the requirement as we had expected. In addition, as the dimension reduction method of Neural Networks, deep learning, flourishing in recent years, builds up algorithm model which is able to process data effectively by simulating the structure of human brain. Combining the characteristics of engine with deep learning theory, the paper put forward a new method of monitoring engine’s health condition. The advantageous conditions of the method include overcoming the uncertainty of characteristic extraction and deficiency of partial response. It’s able to learn and classify the characteristics automatically. Result of the test shows that the method can not only extract characteristics from massive data, but also obtain high identification accuracy of different health conditions of engine.