Gaussian process based modeling and experimental design for sensor calibration in drifting environments

SENSORS AND ACTUATORS B-CHEMICAL

卷:216

Article

Geng, Zongyu¹  Yang, Feng¹  Chen, Xi²  Wu, Nianqiang³ 

[1] W Virginia Univ, Ind & Management Syst Engn Dept, Morgantown, WV 26506 USA

[2] Virginia Tech, Dept Ind & Syst Engn, Blacksburg, VA 24061 USA

[3] W Virginia Univ, Dept Mech & Aerosp Engn, Morgantown, WV 26506 USA

关键词: Sensor drift;    Sensor calibration;    Gaussian process model;    Design of experiments;    Bootstrapping;   

DOI  :  10.1016/j.snb.2015.03.071

来源: Elsevier

【 摘 要 】

It remains a challenge to accurately calibrate a sensor subject to environmental drift. The calibration task for such a sensor is to quantify the relationship between the sensor's response and its exposure condition, which is specified by not only the analyte concentration but also the environmental factors such as temperature and humidity. This work developed a Gaussian Process (GP)-based procedure for the efficient calibration of sensors in drifting environments. Adopted as the calibration model, GP is not only able to capture the possibly nonlinear relationship between the sensor responses and the various exposure-condition factors, but also able to provide valid statistical inference for uncertainty quantification of the target estimates (e.g., the estimated analyte concentration of an unknown environment). Built on GP's inference ability, an experimental design method was developed to achieve efficient sampling of calibration data in a batch sequential manner. The resulting calibration procedure, which integrates the GP-based modeling and experimental design, was applied on a simulated chemiresistor sensor to demonstrate its effectiveness and its efficiency over the traditional method. (C) 2015 Elsevier B.V. All rights reserved.

【 授权许可】

Free