钻井泵液力端故障诊断新方法

往复泵液力端故障原因及故障与征兆间对应关系复杂,为了全面地利用获取的振动信号资源,得到更全面、准确的诊断结果,将分析得到的幅值域的峭度指标、峰值指标、脉冲指标、裕度指标、波形指标和歪度等6个参数,频域的重心频率、均方根频率、频率标准差等3个参数以及32个小波包分频带能量值作为神经网络输入的备选特征向量,由此形成了液力端综合振动信号特征参数的神经网络诊断系统。为了对网络的性能进行比较,分别构建了BP网络和RBF网络。将上述特征输入向量作不同组合,分别输入该网络并进行训练诊断和效果对比,由此求得了最优诊断系统组合。利用此神经网络诊断系统,对现场实际使用的钻井泵液力端进行了多次的测试分析和调试,证明这种方法对钻井泵液力端的故障诊断是行之有效的,可取得较高的诊断准确率。

A new method for fault diagnosis of fluid end in drilling pump

The faults are resulted from some complex reasons in fluid end of drilling pump. The corresponding relationship between faults and symptoms is complex. In order to totally use information resources of vibration signal and to obtain more comprehensive and precise results, six amplitude-domain indexes such as Kurtosis, peak, pulsed, margin, wave and skewness, three frequency-domain parameters such as gravity, root mean squared and standard deviation, and 32 wavelet packet frequency band energy values were regarded as the reserve input feature vectors of the artificial neural network (ANN). An ANN diagnosis system was proposed based on synthesis feature parameters of vibration signals. In order to compare the property of network, the BP and RBF networks were established separately. The different combinations of the extracted vectors were taken as the input information of networks, and an optimal diagnosis system was obtained by diagnosis training and effective comparison. The actual tests prove that the ANN diagnosis system is effective and can get more accurate diagnosis rate in the fault diagnosis of fluid end in the reciprocating pump.