摘要
BP神经网络具有优良的非线性映射能力,可以很好地描述频率特征和诊断结果之间的关系。针对BP神经网络存在局部极小值和收敛速度慢等问题,提出了一种基于Gauss-Newton的改进的BP网络。论述了Gauss-Newton神经网络的基本原理以及学习、运行过程,通过模拟运算指出了Gauss-Newton神经网络具有较快的学习速度,进而探讨了Gauss-Newton神经网络在旋转机械故障诊断中的应用,将该网络模型应用于旋转机械故障诊断,显示出Gauss-Newton网络具有诊断精度高、容错性和稳定性好的优势。
BP Neural network is effective for dealing with non-linear mapping which could satisfaetorily describe the non-linear relations between frequency character and diagnosis results. A kind of BP neural network based on Gauss-Newton Baekpropagation optimization algorithm is introudueed in detail to eomeover the disadvantages of standard BP algorithm. This paper discusses the Gauss-Newton neural networks and the basic principles of learning and operation process through simulation. It is proved that learning speed of Gauss-Newton neural network is fast. The application of the Gauss-Newton in rotating machinery fault diagnosis is also studied. The experiment shows that the Gauss-Newton network has great advantage in diagnosis accuracy, fault tolerance and stability.
出处
《北京机械工业学院学报》
2007年第4期6-9,共4页
Journal of Beijing Institute of Machinery
基金
国家自然科学基金资助项目(50375017)
北京市自然科学基金资助项目(3062008)
北京市属市管高校人才强教计划资助项目
机电系统测控北京市重点实验室开放课题资助项目(KF20061123201
KF20061123202)
