基于神经网络的旋转机械故障诊断研究

FAULT DIAGNOSIS FOR ROTATING MACHINERY BASED ON NEURAL NETWORKS

一般对特定的基于多层感知器的故障诊断问题 ,很难确定神经网络的结构。在分析了多层感知器对故障的识别和诊断能力后 ,采用由小到大和由大到小的方法确定神经网络隐层数与隐层单元数。研究了基于神经网络和振动频谱的旋转机械故障诊断方法 ;研制了一个基于该方法的智能故障诊断系统 ,该系统集网络学习、故障诊断、数据库管理和数据查询为一体。在该智能诊断系统中采用了知识子块的概念 ,系统界面友好 ,交互性强。将其应用于某个大型风机的故障诊断中 ,结果表明该系统操作方便 ,诊断结果准确可靠 ,且具有很强的鲁棒性 ,对某些情况可以实现自动诊断 ,进而证明了该诊断方法的有效性。

Generally,it is difficult to determine in advance a suitable network structure when a multi layer perceptron neural networks is used for a special fault diagnosis problem. For this reason,the fault diagnosis and pattern recognizing ability of multi layer perceptron was analyzed, the method of increasing or decreasing the numbers of hidden layer and hidden unit was adopted to determining the structure (including the numbers of hidden layer and hidden unit) of the neural networks. The fault diagnosis method based on this approach for rotating machinery was studied, and the intelligent fault diagnolsis system was developed. The system consist of networks learning, fault diagnosis,data management and result inquiring. The knowledge sub chunk was adopted and used in it; system has friend interface, powerful interchange and reliable diagnosis results. This system was used in fault diagnosis of the blower, it shows that this system is convenient to manipulate with precise result and powerful intelligence, and in some case the diagnosis can automatically be carried out. It is proved that this method is an effective approach to build fault diagnosis system for the rotating machinery.