基于小波神经网络的抽油杆缺陷识别

Recognition Based on Wavelet Neural Network for Sucker Rod's Defects

为了正确识别深井泵抽油系统中抽油杆杆体的缺陷以减少油杆井下断裂等事故的发生,讨论了应用小波变换和神经网络技术进行缺陷识别的方法.使用小波与神经网络松散型结合的方法,基于小波包原理,将抽油杆的时域检测信号分解到独立的频带内,应用自适应学习速率梯度下降动量法的BP网络,将提取的频带能量作为神经网络输入,抽油杆的裂纹、腐蚀坑、偏磨、损伤及无缺陷作为神经网络待识别输出.经过实验室大量的实验数据训练和验证,结果表明,此种方法既可以正确识别抽油杆的单一缺陷,也可以识别混合缺陷.

Sucker rod is an integral part of the oil well pumping. To reduce the rupture possibility of the rod, it is important to recognize correctly its defects. Discusses the way to recognize the defects by using wavelet transform in combination with neural network. Based on the principle of wavelet package, the signals in time domain detection are decomposed and enter into every frequency band within which the energy is extracted as an input to BP neural network by the momentum-range method for adaptive learning speed gradient descent, then the output reveals the sucker rod＇s defects including crack, corrosion pits, partial wear, impairments and indefectible rod. Testing results with lots of data acquired in lab showed that the way to recognize the defects of sucker rod involves both the single and mixed defects.