小波变换和人工神经网络在荧光测温信号处理中的应用研究
发布时间：2018-08-11 17:27
【摘要】：随着温度测量在科学研究和工业控制过程中显得越来越重要,人们对测温仪的测量精度、应用范围要求也越来越高,因此荧光测温技术应运而生,迅速得到人们的青睐并且取得长足的发展。与其他类型的光纤温度传感器相比,荧光光纤温度传感器具有很多优势,既可以避免交叉灵敏度、光纤的损耗、环境的辐射、发射波的带宽等因素带来的测温精度影响,又具有稳定性好,可靠性高,使用寿命长、生产简单、成本低等特点。本文首先详细阐述了荧光测温工作机理,选择Ca2Mg Si2O7:Au+作为荧光材料,搭建了基于荧光寿命的荧光光纤测温系统。在荧光寿命法测温中,由于温度只与荧光寿命成直接关系,因此只需要获得荧光寿命即可,而噪声干扰是影响荧光寿命的分析计算关键问题。本文提出一种改进的小波阈值去噪,既保留传统的软硬阈值函数优点,又能对两者的不足之处改良,从而达到最优的信噪分离效果;通过matlab仿真分析,对比去噪后信号的信噪比,阐述了小波基、层数、阈值的选择的重要性。在实际测量中,由于去噪后的荧光信号呈现非指数形式,需要通过数据拟合来建立模型得到荧光寿命。通过对比传统的拟合数据方法的缺点,本文采用人工神经网络进行拟合,并采用小波基函数代替sigmod激活函数,结合BP算法和遗传算法的改进算法进行网络的学习,不仅提高了拟合精度和收敛速度,而且避免了较大的局部误差。将新阈值小波去噪应用在荧光信号中,筛选出合适的小波基、阈值和层数,通过matlab仿真对比去噪后的信号,可以看出信号既可以避免出现震荡点,减少了有用信息的损失,变得更加光滑,同时提高了信号的信噪比;对比几种拟合方法得到的荧光寿命,说明了人工神经网络优越性,采用小波神经网络拟合数据,不仅提高了信号曲线的拟合精度,减小了荧光寿命测量误差,同时也提高了温度测量精度。实验结果表明了本文提出方法的有效性。
[Abstract]:With the increasing importance of temperature measurement in the process of scientific research and industrial control, the measurement accuracy and application range of the thermometer are becoming more and more demanding, so the fluorescence temperature measurement technology emerges as the times require. Quickly get people's favor and make great progress. Compared with other kinds of optical fiber temperature sensors, fluorescent optical fiber temperature sensors have many advantages, such as avoiding the influence of cross sensitivity, optical fiber loss, ambient radiation, bandwidth of emission wave, etc. It also has good stability, high reliability, long service life, simple production, low cost and so on. In this paper, the working mechanism of fluorescence temperature measurement is described in detail. Ca2Mg Si2O7:Au is selected as the fluorescent material, and a fluorescent fiber temperature measuring system based on fluorescence lifetime is built. In the measurement of temperature by fluorescence lifetime method, because the temperature is only directly related to the fluorescence lifetime, it is only necessary to obtain the fluorescence lifetime, and the noise interference is the key problem in the analysis and calculation of the influence of the fluorescence lifetime. In this paper, an improved wavelet threshold de-noising is proposed, which not only retains the advantages of the traditional soft and hard threshold function, but also improves the shortcomings of both, so as to achieve the optimal separation effect of signal and noise. Compared with the signal-to-noise ratio (SNR) of the de-noised signal, the importance of the selection of wavelet basis, number of layers and threshold is expounded. In the actual measurement, because the de-noised fluorescence signal presents a non-exponential form, it is necessary to establish a model to obtain the fluorescence lifetime through data fitting. By comparing the shortcomings of the traditional fitting method, this paper uses artificial neural network to fit, and uses wavelet basis function instead of sigmod activation function, and combines BP algorithm and genetic algorithm to learn the network. It not only improves the fitting accuracy and convergence speed, but also avoids large local errors. The new threshold wavelet denoising is applied to the fluorescence signal, and the suitable wavelet basis, threshold value and layer number are screened out. By comparing the de-noised signal with matlab simulation, it can be seen that the signal can not only avoid the oscillation point, but also reduce the loss of useful information. Compared with the fluorescence lifetime obtained by several fitting methods, the advantage of artificial neural network is explained. Wavelet neural network is used to fit the data, which not only improves the fitting accuracy of signal curve, but also improves the accuracy of signal curve fitting. The measurement error of fluorescence lifetime is reduced and the precision of temperature measurement is improved. The experimental results show the effectiveness of the proposed method.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN911.7;O174.2;TP183
[Abstract]:With the increasing importance of temperature measurement in the process of scientific research and industrial control, the measurement accuracy and application range of the thermometer are becoming more and more demanding, so the fluorescence temperature measurement technology emerges as the times require. Quickly get people's favor and make great progress. Compared with other kinds of optical fiber temperature sensors, fluorescent optical fiber temperature sensors have many advantages, such as avoiding the influence of cross sensitivity, optical fiber loss, ambient radiation, bandwidth of emission wave, etc. It also has good stability, high reliability, long service life, simple production, low cost and so on. In this paper, the working mechanism of fluorescence temperature measurement is described in detail. Ca2Mg Si2O7:Au is selected as the fluorescent material, and a fluorescent fiber temperature measuring system based on fluorescence lifetime is built. In the measurement of temperature by fluorescence lifetime method, because the temperature is only directly related to the fluorescence lifetime, it is only necessary to obtain the fluorescence lifetime, and the noise interference is the key problem in the analysis and calculation of the influence of the fluorescence lifetime. In this paper, an improved wavelet threshold de-noising is proposed, which not only retains the advantages of the traditional soft and hard threshold function, but also improves the shortcomings of both, so as to achieve the optimal separation effect of signal and noise. Compared with the signal-to-noise ratio (SNR) of the de-noised signal, the importance of the selection of wavelet basis, number of layers and threshold is expounded. In the actual measurement, because the de-noised fluorescence signal presents a non-exponential form, it is necessary to establish a model to obtain the fluorescence lifetime through data fitting. By comparing the shortcomings of the traditional fitting method, this paper uses artificial neural network to fit, and uses wavelet basis function instead of sigmod activation function, and combines BP algorithm and genetic algorithm to learn the network. It not only improves the fitting accuracy and convergence speed, but also avoids large local errors. The new threshold wavelet denoising is applied to the fluorescence signal, and the suitable wavelet basis, threshold value and layer number are screened out. By comparing the de-noised signal with matlab simulation, it can be seen that the signal can not only avoid the oscillation point, but also reduce the loss of useful information. Compared with the fluorescence lifetime obtained by several fitting methods, the advantage of artificial neural network is explained. Wavelet neural network is used to fit the data, which not only improves the fitting accuracy of signal curve, but also improves the accuracy of signal curve fitting. The measurement error of fluorescence lifetime is reduced and the precision of temperature measurement is improved. The experimental results show the effectiveness of the proposed method.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN911.7;O174.2;TP183
【参考文献】
相关期刊论文 前10条
1 孙德敏,张利,王永,秦琳琳,司炜,段增强;基于乘积型最小二乘法的传感器特性拟合[J];传感技术学报;2002年04期
2 张友俊,汤伟中,郭军伟,周泓;非接触型荧光屏光纤温度传感器[J];传感器技术;1999年03期
3 陈华丽,李裕能;基于小波变换的自适应阈值消噪法[J];电力科学与工程;2003年03期
4 刘志刚,王晓茹,钱清泉;小波网络的研究进展与应用[J];电力系统自动化;2003年06期
5 曾U喺，
本文编号：2177712
本文链接：https://www.wllwen.com/kejilunwen/yysx/2177712.html
