Investigation of the Impact of Environmental Factors on the Electrochemical Sensor Response and Recommendation for Ambient Measurement

環境因素對電化學傳感器影響的研究及使用於大氣監測上的建議

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
  • Keith NGAN (Supervisor)
  • Zhi NING (External person) (External Co-Supervisor)
Award date25 Mar 2019

Abstract

Commercial electrochemical sensors are now being widely used by the environmental pollutant monitoring community due to their low cost and small shape factor that readily adapts into different applications. This type of sensor is sensitive not only to gas concentration variation, but also to environmental condition change and even chemical changes within the cells. Applying algorithms to data reported by sensors to consider the influence of environmental factors is especially crucial for the sensor’s proper usage under ambient conditions but is a challenging task. First, the influences of environmental factors are often not linear and are different for different sensors produced by various firms and for each specific gas. These differences can be considered at various stages in sensor design, calibration and in processing of sensor data. This paper will focus on the mathematical post processing of data from sensors.

Designing mathematical correction for sensor is challenging. Considerable complexity centers on the choice of environmental parameters to include in the sensor data correction model. For example, temperature is a straight forward metric for inclusion. A second metric commonly is relative humidity, but electrochemical sensors are essentially influenced by the mass of ambient water vapor - more commonly referred to as absolute humidity, so there is the remaining question on the best choice of the input parameters. A factor that has considerable impacts in field calibration of sensors is that electrochemical sensor with high time resolution (seconds to minutes) are often placed near the lower time resolution (minutes to hours) reference machines to provide reliable values for model development and training. The discrepancy in source data frequency may mask the variation of non-linearly related environmental factors and cause uncertainty in the sensor algorithm performance. Calibrating high time resolution sensor by low time resolution sampling may reduce the sensor model accuracy.

This thesis aims at providing an analytic evaluation of the signal responses of electrochemical sensors in ambient conditions and to provide suggestions in addressing the uncertainties in sensor correction algorithms commonly encountered in the low cost sensor applications when used to measure ambient air. A statistical method is proposed for analyzing the environmental influences on the signal of electrochemical sensors. Strategies for designing parametric models for retrieving concentration information from electrochemical sensors is then introduced. Statistical methods are applied to the field data to evaluate the performance of commercial gas sensors for different pollutants including carbon monoxide, nitric oxide, nitrogen dioxide and ozone. The designed parametric models are compared with other commonly used non-parametric models (Smoothing Spline and Kernel Regression) and machine learning models (Support Vector Regression and Random Forest). Based on the model results, this thesis also makes practical recommendations on the selection of relative humidity or absolute humidity on the performance of the sensor model, the suitable calibration interval for different types of sensors, etc.

    Research areas

  • electrochemical sensor, amperometric gas sensor, sensor response, GAM, generalized additive model