光栅在轴承声发射信号测量中的应用研究

发布时间：2018-05-25 19:17

本文选题：轴承 + 声发射　；参考：《沈阳工业大学》2017年硕士论文

【摘要】：与核能发电和火力发电相比,风力发电更绿色环保、资源充足。随着风力发电行业迅猛发展,风机装机量逐年递增。如何及时发现风机故障,准确判断故障类型,保证风力发电机组安全可靠运行成为世界各国主要研究的问题。本文以风机轴承为研究对象,搭建风机轴承故障监测系统,利用该系统对轴承运行状态进行实时监测和故障诊断研究。该系统利用光栅传感器提取轴承声发射信号。当轴承发生故障时,声发射现象会引起轴承表面位移变化;光栅传感器会将位移变化量转换成电信号;AD9467采集传感器输出的电信号;STM32F429对信号进行滤波、细分和故障诊断。故障发生时,该系统有报警提示功能。此外,该系统还具有数据存储功能和网络传输功能。光栅传感器的分辨力决定了轴承故障声发射信号识别能力。为提高系统故障诊断的精准性,本文对莫尔条纹信号进行高倍细分。首先,通过智能小波阈值降噪方法对信号进行小波分解和降噪。其次,针对莫尔条纹信号含直流电平、幅值不等、相位不正交等现象,进行莫尔条纹信号补偿可有效提高细分精度。再次,本文对基于L-M的BP神经网络莫尔条纹信号细分方法进行了深入研究。通过增加新的判断条件来改进L-M算法,根据本次训练结果误差和上一次训练结果误差关系可以得到新的权值。该方法能够提高神经网络训练速度和结果精度。将其结果与RBF神经网络莫尔条纹细分方法所得结果进行对比,实验结果表明,基于改进L-M的BP神经网络莫尔条纹信号细分方法速度更快、误差波动范围更小。然后,对细分后得到的位移值做频谱分析可得不同频率对应的幅值。最后,通过比较幅值和轴承临界故障时的幅值可以判断轴承是否有故障。仿真结果和实验测试结果表明,基于改进L-M的BP神经网络莫尔条纹信号细分方法可以实现20000细分,分辨力达到1nm,能够识别纳米级轴承裂纹故障的声发射信号。表明光栅传感器通过莫尔条纹信号细分后可以用于提取轴承故障声发射信号,通过小波神经网络故障诊断方法能判断出轴承裂纹故障。
[Abstract]:Compared with nuclear power generation and thermal power generation, wind power generation is greener and more resources. With the rapid development of the wind power generation industry, the volume of wind turbines is increasing year by year. How to find out the blower fault in time, accurately determine the type of fault and ensure the safe and reliable operation of the wind turbine is the main problem in the world. This paper is based on the wind turbine shaft. As the research object, a fault monitoring system for the bearing of the fan is built, and the system is used to monitor and diagnose the bearing state of the bearing in real time. The system uses a grating sensor to extract the acoustic emission signal of the bearing. When the bearing occurs, the acoustic emission will cause the change of the displacement of the bearing surface, and the grating sensor will change the displacement. The electrical signal is converted into an electrical signal; AD9467 takes the electrical signal output by the sensor; STM32F429 filters, subdivides and diagnoses the signal. When the fault occurs, the system has alarm and prompt function. In addition, the system also has data storage function and network transmission function. The resolution of the grating sensor determines the identification of the acoustic emission signal of the bearing fault. In order to improve the accuracy of the system fault diagnosis, the moire stripe signal is subdivided in high times. First, wavelet decomposition and noise reduction are carried out by the intelligent wavelet threshold denoising method. Secondly, the moire fringe signal compensation can be effectively proposed for the moire fringe signal containing the DC level, the amplitude is unequal and the phase is not orthogonal. Thirdly, the L-M based BP neural network Moire stripe signal subdivision method is deeply studied in this paper. By adding new judgment conditions to improve the L-M algorithm, new weights can be obtained according to the error of this training result and the error relationship of the previous training results. This method can improve the training speed of neural network and the training speed of neural network. The result is compared with the results obtained from the RBF neural network moire fringe subdivision method. The experimental results show that the moire fringe signal subdivision method based on the improved L-M neural network is faster and the range of error fluctuation is smaller. Then, the spectrum analysis of the displacement values obtained after the subdivision can be obtained with the corresponding amplitude of different frequencies. Finally, by comparing the amplitude and the amplitude of the critical fault of the bearing, the fault of the bearing can be judged. The simulation results and the experimental test results show that the 20000 subdivision method based on the improved L-M BP neural network moire fringe signal subdivision can be realized and the resolution can reach 1nm. After subdivision of the moire fringe signal, the light grating sensor can be used to extract the acoustic emission signal of the bearing fault, and the fault of the bearing crack can be judged by the wavelet neural network fault diagnosis method.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM614;TP212

【参考文献】