Hybrid Model-Driven Spectroscopic Network for Rapid Retrieval of Turbine Exhaust Temperature

被引：6

作者：

Fu, Yalei ^{[1
]}

Zhang, Rui ^{[1
]}

Xia, Jiangnan ^{[1
]}

Gough, Andrew ^{[2
]}

Clark, Stuart ^{[2
]}

Upadhyay, Abhishek ^{[2
]}

Enemali, Godwin ^{[1
]}

Armstrong, Ian ^{[2
]}

Ahmed, Ihab ^{[3
]}

Pourkashanian, Mohamed ^{[3
]}

Wright, Paul ^{[4
]}

Ozanyan, Krikor ^{[4
]}

Lengden, Michael ^{[2
]}

Johnstone, Walter ^{[2
]}

Polydorides, Nick ^{[1
]}

McCann, Hugh ^{[1
]}

Liu, Chang ^{[1
]}

机构：

[1] Univ Edinburgh, Sch Engn, Edinburgh EH9 3JL, Scotland

[2] Univ Strathclyde, Dept Elect & Elect Engn, Glasgow G1 1XQ, Scotland

[3] Univ Sheffield, Dept Mech Engn, Sheffield S10 2TN, England

[4] Univ Manchester, Sch Elect & Elect Engn, Manchester M13 9PL, England

来源：

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT | 2023年 / 72卷

基金：

英国工程与自然科学研究理事会;

关键词：

Deep neural network (DNN); exhaust gas temperature (EGT); gas turbine engine (GTE); signal processing; wavelength modulation spectroscopy (WMS); WAVELENGTH-MODULATION SPECTROSCOPY; TOMOGRAPHY;

D O I：

10.1109/TIM.2023.3328086

中图分类号：

TM [电工技术]; TN [电子技术、通信技术];

学科分类号：

0808 ; 0809 ;

摘要：

Exhaust gas temperature (EGT) is a key parameter in diagnosing the health of gas turbine engines (GTEs). In this article, we propose a model-driven spectroscopic network with strong generalizability to monitor the EGT rapidly and accurately. The proposed network relies on data obtained from a well-proven temperature measurement technique, i.e., wavelength modulation spectroscopy (WMS), with the novelty of introducing an underlying physical absorption model and building a hybrid dataset from simulation and experiment. This hybrid model-driven (HMD) network enables strong noise resistance of the neural network against real-world experimental data. The proposed network is assessed by in situ measurements of EGT on an aero-GTE at millisecond-level temporal response. Experimental results indicate that the proposed network substantially outperforms previous neural-network methods in terms of accuracy and precision of the measured EGT when the GTE is steadily loaded.

引用

页码：1 / 10

页数：10

共 30 条

[21] Adaptively Optimized Gas Analysis Model with Deep Learning for Near-Infrared Methane Sensors [J].