基于改进型小波神经网络的往复泵故障诊断  

Fault Diagnosis Technology of Reciprocating Pumps Based on Improved Wavelet Neural Network

作　　者：赵志华[1] 吴力[2] 

机构地区：[1]东北石油大学电气信息工程学院,黑龙江大庆163318 [2]大庆油田天然气分公司培训中心,黑龙江大庆163412

出　　处：《自动化技术与应用》2013年第9期76-78,共3页Techniques of Automation and Applications

摘　　要：为了对往复泵的故障进行正确诊断,提出了基于改进型小波神经网络的往复泵故障诊断方法。以往复泵单个泵缸内的压力信号作为系统特征信号通过小波包分解来提取故障特征向量,同时将此特征向量作为改进型神经网络的输入,利用改进型神经网络对故障做进一步的精确实时诊断。文中对小波神经网络采用的优化算法是:动量因子和学习率自适应调整相结合的梯度下降法,该方法可以提高学习速度并增加算法的可靠性。通过对往复泵液力端多故障诊断实例的检验表明,该系统故障诊断正确率达到了93%以上。A method is proposed based on improved wavelet neural network, in order to accurately judge the reciprocating pump fault type. This paper uses the reciprocating pump cylinder pressure signal as the characteristics of the system signal by wavelet transform to extract fault features vector, and at the same time, this feature vector takes as improved neural network＇s input, then uses improved neural network to determine the diagnosis the type of the fault. In this paper, the wavelet neural network optimization algorithm is the gradient descent method of momentum factor and learning speed adaptive adjustment. This method can improve the learning speed and increase the reliability of the algorithm. Diagnosis of faults of fluid end on a reciprocating pump proves the system fault diagnosis accuracy reached 93%.

关 键 词：小波神经网络 往复泵 故障诊断 

分 类 号：TP277.3[自动化与计算机技术—检测技术与自动化装置]