DBR光纤激光拍频结合BP神经网络的温度传感研究

发布时间：2018-05-12 22:37

  本文选题：拍频 + 光纤激光器　； 参考：《河南师范大学》2017年硕士论文

【摘要】：近年来,光纤光栅传感技术在各个领域有广泛的应用,如环境、农业、地质探测、太空等,其传感解调方法一直是人们关注的焦点。普遍商用的方法是采用光纤F-P腔扫描等光干涉解调,这些方法技术复杂,成本较高。为降低系统成本,人们提出了外差式解调的方法。这一技术普遍采用分布反馈Bragg光纤激光器(DFB)结构,利用其双折射在探测器上形成的拍频实现传感解调。该方法中,实现稳定的拍频需要较复杂的技术。继而,人们提出采用分布反射激光器(DBR)结构,利用光纤激光器的腔长变化所形成的多纵模拍频实现传感解调。本文提出了一种实现光纤光栅温度传感解调的方法,通过测量激光拍频得出,同时对测量的温度传感数据进行优化用所构建的三层BP神经网络模型。该方法分别采用线性啁啾光栅(CFBG)和传感光纤光栅(FBG)作为光纤激光系统的反馈腔镜,测量激光器拍频随传感光栅温度的变化实现温度传感。在之前的传感解调系统中,我们往往以啁啾光纤光栅(CFBG)的时延特性做为参照标准,假设啁啾光纤光栅(CFBG)具有理想的线性时延,然而在实际应用中,由于制作工艺的限制,啁啾光纤光栅(CFBG)时延并非完全线性,且存在抖动。按照线性时延处理测试结果,存在明显的系统误差。因此在测量前应根据系统中所使用啁啾光纤光栅(CFBG)的具体时延特性曲线标定相应的拍频频率以降低该系统误差。温度测量误差是由啁啾光纤光栅(CFBG)非线性时延及时延抖动本身的特性引起的,为了使误差更小,我们需要选择一种算法来处理所得的温度数据。BP神经网络算法具有良好的容错和非线性映射能力,可逼近任意非线性函数,解决复杂参量之间的非线性对应关系[1]。利用BP神经网络算法搭建了三层BP神经网络模型,实验中,重复测量10次,得到10组拍频频率/温度数据。将所测频率数据中的9组确定为训练校正集,然后作为网络输入值送入所建模型的输入层,而输出值为相对应的实际温度值,对网络参数值进行训练,最终使参数值达到最佳网络结构。用剩下的一组作为测试样本集进行检验,此组数据的温度灵敏度和相关系数分别为37.89KHz/℃和99.767%,对该组数据训练温度校正及预测,其相关系数达到99.95%。通过实验,我们可以得出用三层BP神经网络算法对实验所得数据进行检验,能极大的改善实验系统的测量精度。本文基于激光拍频结合BP神经网络算法实现温度传感,使拍频解调这一技术更实用化。
[Abstract]:In recent years, fiber Bragg grating sensing technology has been widely used in various fields, such as environment, agriculture, geological exploration, space and so on. The common commercial methods are optical fiber F-P cavity scanning equal-optical interference demodulation. These methods are complex and expensive. In order to reduce the system cost, heterodyne demodulation is proposed. The distributed feedback Bragg fiber laser (DFB) structure is widely used in this technique, and the beat frequency formed by its birefringence on the detector is used to demodulate the sensor. In this method, complex techniques are needed to achieve stable beat frequency. Then, a distributed reflection laser (DBR) structure is proposed to demodulate the sensor using multi-longitudinal-mode beat frequency formed by the variation of the cavity length of the fiber laser. In this paper, a demodulation method of fiber Bragg grating (FBG) temperature sensing is proposed, which is obtained by measuring the laser beat frequency and optimizing the measured temperature sensing data using the three-layer BP neural network model. In this method, the linear chirped grating (CFBG) and the sensing fiber Bragg grating (FBGG) are used as the feedback mirrors of the fiber laser system, respectively, and the laser beat frequency is measured with the change of the temperature of the sensing grating to realize the temperature sensing. In previous sensing and demodulation systems, we often take the time delay characteristics of chirped fiber grating (CFBG) as the reference standard, and assume that chirped fiber grating (CFBG) has ideal linear delay. However, in practical applications, due to the limitation of fabrication process, The time delay of chirped fiber grating (CFBG) is not completely linear and jitter exists. According to the test results of linear delay processing, there is obvious systematic error. Therefore, the corresponding beat frequency should be calibrated according to the specific time-delay characteristic curve of chirped fiber Bragg grating (CFBG) used in the system before measurement to reduce the error of the system. The temperature measurement error is caused by the characteristic of the nonlinear delay and jitter of the chirped fiber Bragg grating (CFBG). We need to select an algorithm to process the temperature data. BP neural network algorithm has good fault-tolerant and nonlinear mapping ability, it can approximate any nonlinear function and solve the nonlinear correspondence between complex parameters [1]. A three-layer BP neural network model was built by using BP neural network algorithm. In the experiment, 10 sets of beat frequency / temperature data were obtained by repeated measurement. Nine groups of the measured frequency data are determined as the training correction set, then the input values of the network are fed into the input layer of the established model, and the output values are the corresponding actual temperature values, so the network parameters are trained. Finally, the parameter value reaches the optimal network structure. The temperature sensitivity and correlation coefficient of the data are 37.89KHz/ 鈩，

本文编号：1880447