Calibration Model of a Low-Cost Air Quality Sensor Using an Adaptive Neuro-Fuzzy Inference System.

Kemal Alhasa,Md Khan,Imran Yusuff,Tengku Tengku Hassim,Sawal Md Ali,Murnira Othman,Fatimah Ahamad,Mohd Latif,Nor Ezani,Mohammad Iqbal Faruque,Azizan Abu Samah,Yusri Yusup,Kadaruddin Aiyub,Mohd Mohd Nadzir,Haris Abd Hamid,Popoola Olalekan

doi:10.3390/s18124380

Abstract

Conventional air quality monitoring systems, such as gas analysers, are commonly used in many developed and developing countries to monitor air quality. However, these techniques have high costs associated with both installation and maintenance. One possible solution to complement these techniques is the application of low-cost air quality sensors (LAQSs), which have the potential to give higher spatial and temporal data of gas pollutants with high precision and accuracy. In this paper, we present DiracSense, a custom-made LAQS that monitors the gas pollutants ozone (O3), nitrogen dioxide (NO2), and carbon monoxide (CO). The aim of this study is to investigate its performance based on laboratory calibration and field experiments. Several model calibrations were developed to improve the accuracy and performance of the LAQS. Laboratory calibrations were carried out to determine the zero offset and sensitivities of each sensor. The results showed that the sensor performed with a highly linear correlation with the reference instrument with a response-time range from 0.5 to 1.7 min. The performance of several calibration models including a calibrated simple equation and supervised learning algorithms (adaptive neuro-fuzzy inference system or ANFIS and the multilayer feed-forward perceptron or MLP) were compared. The field calibration focused on O3 measurements due to the lack of a reference instrument for CO and NO2. Combinations of inputs were evaluated during the development of the supervised learning algorithm. The validation results demonstrated that the ANFIS model with four inputs (WE OX, AE OX, T, and NO2) had the lowest error in terms of statistical performance and the highest correlation coefficients with respect to the reference instrument (0.8 < r < 0.95). These results suggest that the ANFIS model is promising as a calibration tool since it has the capability to improve the accuracy and performance of the low-cost electrochemical sensor.

Highlights

Poor air quality has been linked to human health effects with increased associated diseases and symptoms since the rapid growth period associated with the 4th industrial revolution [1,2]
A simple equation was used to translate the signal to mixing ratio and was calibrated by adding a correction in order to achieve the minimum difference against the gas standard
After the laboratory experiment was performed, field tests were undertaken to investigate the performance of the EC sensor when measuring gases in ambient conditions

Summary

Introduction

Poor air quality has been linked to human health effects with increased associated diseases and symptoms since the rapid growth period associated with the 4th industrial revolution [1,2]. In Malaysia these gas pollutants are usually found at significantly high concentrations over urban areas such as the Klang Valley, as reported by Latif et al, Banan et al., Ahamad et al and Ismail et al [4,5,6,7]. In line with these findings, it is important for local authorities to continuously monitor these pollutants. Alternative air pollution monitoring approaches have emerged such as low-cost air quality measurement techniques [11]

Objectives

Methods

Conclusion