Multi-parameter demodulation for temperature, salinity and pressure sensor in seawater based on the semi-encapsulated microfiber Mach-Zehnder interferometer. (15th June 2022)

• Record Type: Journal Article
• Authors: Liu, Jichao; Hou, Yunfei; Wang, Jing; Zhong, Guoqiang; Zhang, Lihui; Zhuang, Funa; Yu, Lijun; Wang, Shanshan
• Is Part Of: Measurement. Volume 196(2022)
• Publication Date: 2022-06-15
• Subjects: Multi-parameter sensing -- Temperature, salinity and pressure in seawater -- Support vector regression; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.111213
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: An sample-fabricated multi-parameter sensor based on the polydimethylsiloxane semi-encapsulated in-line microfiber Mach–Zehnder interferometer (MMZI) for simultaneous measurement of temperature, salinity and pressure in seawater is proposed and realized with typical sensitivities of −2312 pm/℃, 631 pm/‰, and 3775 pm/MPa, respectively. Based on the theoretical analysis, encapsulation dependent temperature sensitivity is revealed, which is much different with those traditional MMZIs. In signal demodulation, both sensitivity matrix method and machine learning method are used and compared with results measured by commercial conductivity-temperature-depth (CTD) system. Using the machine learning method, accuracy of multi-parameter demodulation is improved significantly than that of sensitivity matrix method in salinity and pressure measurement. Non-linear dependence of sensor sensitivity on surroundings is revealed, by which differences between two demodulation methods (sensitivity matrix method and machine learning method) are analyzed. Abstract: Semi-encapsulated microfiber Mach-Zehnder interferometer is designed and developed for temperature, salinity and pressure (TSP) sensing in seawater. Based on the theoretical analysis, sensing experiment is performed with typical sensitivities of −2312 pm/℃, 631 pm/‰, and 3775 pm/MPa, respectively. To demodulate the signal with cross-sensitivity, sensitivity matrix method (SMM) and machine learning method (MLM) are used, respectively. By 25 tests under arbitrary TSP, relatively low errors of about 19.14 %, 4.01 % and 15.75 % are obtained based on the support vector regression (SVR) model. In addition, non-linear dependence of sensitivity on surrounding, SMM accuracy on sensing dips, stability of prediction and influence of the dataset used in training are also investigated. Finally, combination of SMM and MLM is realized, which shows relatively good performance for TSP measurement with errors of 10.67 %, 5.25 % and 16.76 %, respectively.