摘要
航空发动机故障诊断技术对避免飞行事故和降低飞行器运行成本是十分重要的。提出一种BP网络对某型飞机发动机进行故障诊断,但是由于BP网络收敛速度较慢而且容易陷入局部极小值,特别是BP网络通常只能给出一个解,受训练样本病态影响大。因此通过对BP网络的改进,建立了L-M算法神经网络的飞机发动机故障诊断模型。实验表明,该网络在一定程度上克服了BP网络存在的的问题,在逼近能力、分类能力和学习速度等方面均优于BP网络。为机务人员提供了有效的、科学的发动机故障诊断方法,该种评估手段较好地解决了发动机故障诊断问题,在飞行安全中发挥着越来越大的作用。
Flight Simulation and flight training data intelligent evaluation have been used wildly in pilots training. This paper is based on DirectX technology under a certain type of aircraft 3D model, implemented a certain type of aircraft flight simulation platform. Based on this platform, we simulated actual flight courses and obtained flight data from a special flight course. According to expert＇s and special-class pilot＇s flying experience, we extracted feature vectors as key parameters, and input them into neural network model. After the neural network learning, more accurate flight evaluation results can be achieved. The algorithm greatly improves the efficiency of flight training data evaluation, reduces the man-made errors, corrects the deviation of the flight, and increases levels of pilot＇s flight training. Considering slow convergence of BP neural network, calculated results affected by the initial value, poor stability, easy defects such as a local minimum, we applied L-M algorithm instead of gradient descent algorithm to neural network training. The establishment of the L-M algorithm based on the flight simulation data model has been developed. The research shows that results that are generated by L-M model are significantly better than the other three layers BP neural network models.
出处
《电子设计工程》
2012年第11期89-92,共4页
Electronic Design Engineering
关键词
飞机发动机
故障诊断
神经网络
L—M算法
plane engine
failure diagnose
neural networks
Levenberg-Marquardt algorithm
