基于LVQ神经网络风电机组齿轮箱故障诊断研究
2014-07-19丁硕常晓恒巫庆辉魏洪峰杨友林
丁硕 常晓恒 巫庆辉 魏洪峰 杨友林
摘 要: 针对风电机组齿轮箱故障诊断技术的不足,提出一种基于LVQ神经网络的故障诊断方法,利用小波分析方法对某风电机组齿轮箱正常状态、磨损故障和断齿故障状态下的振动信号进行降噪处理,在时域和频域内提取了5个特征参数对所建立的模型进行训练。为了检验模型的实际诊断能力,与标准BP神经网络的诊断结果进行对比。仿真结果表明:基于LVQ神经网络的故障诊断速度更快、准确率更高、泛化能力更强,验证了所提出方法的实用性和有效性。
关键词: LVQ神经网络; BP神经网络; 风电机组; 齿轮箱; 故障诊断
中图分类号: TP183 文献标识码: A 文章编号: 1004?373X(2014)10?0150?03
Abstract: In view of the deficiency in fault diagnosis technique of wind turbine gearbox, a fault diagnosis method based on LVQ neural network is proposed. Wavelet analysis is used to de?noise the vibration signals of a wind turbine gearbox in its normal condition, wear fault condition and tooth breakage condition. Five characteristic parameters are extracted in the time domain and frequency domain to train the established model. To test its practical diagnosis ability, the diagnosis result of the model is compared with that obtained by a standard BP neural network. The simulation results show that the diagnosis method based on LVQ neural network has a faster diagnosis speed, higher accuracy and stronger generalization ability. The method proposed in this paper was verified to be practical and effective.
Keywords: LVQ neural networks; BP neural networks; wind turbines; gearbox; fault diagnosis
0 引 言
风力发电技术是目前国际上可再生能源领域发展最快的技术手段之一,齿轮箱是风力发电机组故障率最高的部件,风电机组齿轮箱安装空间狭小,而且又位于高空塔顶作业,一旦发生故障,维修非常困难。人工神经网络的快速发展为解决非线性复杂系统的故障诊断问题提供了一种新的解决途径。在众多的人工神经网络类型中,反向传播(Back Propagation,BP)神经网络是应用最为广泛的一类网络。但是,标准BP网络的收敛速度较慢,而且可能陷入局部极小值[1?4]。学习向量量化法( Learning Vector Quantization,LVQ) 神经网络是在监督状态下对竞争层进行训练的一种学习算法。竞争层自动学习并对输入向量进行分类, 这种分类的结果仅仅依赖于输入向量之间的距离。如果两个输入向量特别相近, 竞争层就把它们分在同一类,从而能较好地克服标准BP网络训练时间长及计算复杂度高等缺点[5]。……
