基于Bi-LSTM网络的游标传感器输出解调技术  

Output Demodulation Technology of Vernier Effect Sensor Based on Bi-LSTM Network

作　　者：曾心 郭茂森 张昕 丁晖[1] 胡红利[1] ZENG Xin;GUO Mao-sen;ZHANG Xin;DING Hui;HU Hong-li(School of Electrical Engineering,Xi'an Jiaotong University,Xi'an 710049,China)

机构地区：[1]西安交通大学电气工程学院,陕西西安710049

出　　处：《光谱学与光谱分析》2025年第5期1257-1263,共7页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(52177009)资助。

摘　　要：针对光学游标传感器输出解调难的问题,提出基于双向长短时记忆(Bi-LSTM)网络的光谱数据预测技术。利用Bi-LSTM网络对数据序列的预测能力,实现了宽光谱范围的光谱数据预测,从而解决了游标传感器由于工作光谱范围有限的光源或光谱扫描技术,而导致游标传感器难以实现输出解调的技术难题。采用该方法,只要采集有限波长范围的传感器输出光谱,利用训练好的Bi-LSTM模型就能够在较宽的波长范围内准确预测传感器输出光谱的包络曲线,从而极大降低了对游标传感器工作光谱范围的技术要求。介绍了Bi-LSTM网络用于游标传感器输出解调的基本原理和实现过程,实验证明了该方法对游标传感器输出光谱数据预测的准确性,其预测曲线与实际光谱包络在波峰处的波长最大误差~0.02 nm,幅值最大误差仅为0.058%。验证了Bi-LSTM网络对具有不同包络周期的游标传感器输出解调的泛化性,针对不同包络周期的游标传感器输出光谱,其最大预测误差为0.02 nm,最大均方根误差(RMSE)为9.72×10^(-5),证明了所训练的Bi-LSTM网络对不同包络周期的游标传感器输出光谱都具有准确的“预测性”和“跟踪度”。研究表明,实际工作中只要光源的波长范围能够覆盖游标传感器的1/2个光谱包络周期(绝大多数情况下可以满足),利用Bi-LSTM网络能够在宽光谱范围内,实现对传感器输出光谱的准确预测,从而极大降低了对游标传感器的工作光源(或其他光谱扫描技术)的光谱范围的要求。本研究解决了游标传感器的输出解调光谱范围过宽的难题,具有理论及实际应用意义。To solve the problem of output demodulation of optical vernier sensors,this paper proposes a spectral data prediction technology based on a bidirectional long short-term memory(Bi-LSTM)network.By utilizing the predictive ability of the Bi-LSTM network for data sequences,a wide spectral range of spectral data prediction has been achieved,thus solving the technical problem of cursor sensors having difficulty achieving output demodulation due to the limited working spectral range of light sources or spectral scanning techniques.By using this method,as long as a limited wavelength range of sensor output spectra is collected,the trained Bi-LSTM model can accurately predict the envelope curve of the sensor output spectra over a wide wavelength range,greatly reducing the technical requirements for the working spectral range of the vernier sensor.The paper introduces the basic principle and implementation process of the Bi-LSTM network for output demodulation of vernier sensors.The experiment proves the accuracy of this method in predicting the spectral data output of vernier sensors.The maximum wavelength error between the predicted curve and the actual spectral envelope at the peak is about 0.02 nm,and the maximum amplitude error is only 0.058%.In addition,the paper also verified the generalization of the Bi-LSTM network for demodulating the output spectra of cursor sensors with different envelope periods.For the output spectra of cursor sensors with different envelope periods,the maximum prediction error was 0.02 nm,and the maximum root mean square error(RMSE)was 9.72×10^(-5),proving that the trained Bi-LSTM network has accurate“predictability”and“tracking”for the output spectra of cursor sensors with different envelope periods.Comprehensive research papers have shown that in practice,as long as the wavelength range of the working light source can cover half of the spectral envelope period of the vernier sensor(which can be met in most cases),the Bi-LSTM network can accurately predict the output spectrum of t

关 键 词：光学游标传感器 自由光谱范围 光谱预测 Bi-LSTM网络 

分 类 号：O439[机械工程—光学工程] TP183[理学—光学]