Reliability of Low-Cost, Sensor-Based Fine Dust Measurement Devices for Monitoring Atmospheric Particulate Matter Concentrations

Currently, low-cost, sensor-based fine dust measurement devices are commercially available in South Korea. This study evaluated the reliability of three such devices—Yi Shan A4, Plantower PMS7003, and Plantower PMS7003—in comparison to long-term consecutive monitoring systems for discharge and prevention facilities regarding fine dust control. The performance of these devices for concentration intervals over time was examined through real-time comparison using a GRIMM (Model: 11-A, dust spectrometer from Grimm Technologies) as a reference; this included a correction factor (C-Factor), calculated by a gravimetric method and an equivalence test. For comparison, the reference and target devices were installed in a chamber with fine dust concentrations of 2 µg/m3, with temperature and humidity maintained at 20 °C and 40%, respectively. The fine particulate matter (PM)2.5 concentrations were classified into five intervals: ≤40 µg/m3, 40–80 µg/m3, 80–120 µg/m3, 120–160 µg/m3, and 200–230 µg/m3. Statistical analysis was performed using data obtained from national stations for monitoring and controlling fine dust released from facilities under high fine dust loading conditions. The results showed that the measurements of all target devices, which were corrected according to the reference device, provided accurate values at PM2.5 concentrations of ≥40 µg/m3. The statistical analysis results suggest that the evaluated devices are more reliable than the conventional numerical-analysis-based monitoring system


Introduction
Both human life and the environment are significantly affected by air quality. Numerous studies have attempted to reveal the relationship between air pollution and the quality of life and health. Air pollution deteriorates pulmonary functions, aggravates respiratory and cardiovascular problems, and increases asthma rates [1]. In addition to health problems, air pollution is also detrimental to vegetation [2], visibility, the global climate, and many other factors that impact human life.
Since the enactment of the Clean Air Conservation Act in South Korea in 1990, the Ministry of Environment has revised the act multiple times to ensure appropriate and sustainable conservation of Appearance research project.

Measurement Devices
Three low-cost, sensor-based measurement devices were tested, which are currently used to measure fine dust levels in South Korea. These devices are manufactured by E3 (Yi Shan A4 sensor, SK techx, Seoul, Korea), MAXFOR Technology (Plantower PMS7003 sensor, SK techx, Seoul, Korea), and SK Techx (Plantower PMS7003 sensor, SK techx, Seoul, Korea). The sensors are referred to as '101', '201', and '301', respectively. Table 1 presents detailed specifications of each device and sensor type. Device 101 is an instrument shelter type with an implemented A4 sensor. Both 201 and 301 use the PMS7003 sensor of Plantower and present the same instrument shelter type as 101. Device 201 is designed to be operated by drawing air in through simple inhalation exposure.
To compare the trends of output values of the target measurement devices, using cost-effective sensors, regarding concentration interval in real time, a GRIMM (dust spectrometer from Grimm Technologies), based on multi-channel light scattering, was used as the reference device in a realtime test. Before the test, the correction factor (C-factor) of the reference device was calculated using a gravimetric method and by performing an equivalence test. In the gravimetric method, two minivolume air samplers were used in a chamber of specific dimensions. The C-factor was reflected in the reference device. Particles of Japan Dust Class 10 were used as the test particle for the equivalence test. The details of the equivalence test, including its results, have already been reported by Air Quality Sensor Performance Evaluation Center (AQ-SPEC) [19].

Test Method
The comparison test between the reference device and the target measurement devices using cost-effective particle sensors was performed in the following procedure: The chamber was ventilated using clean air generated by removing particles through a HEPA (high efficiency particulate air) filter (SK techx, Seoul, Korea). After the fine dust level was maintained at or below 2 µ g/m 3 in the chamber, clean air was injected and the ejection was stopped. Once the air flow was at a steady state, both the reference and target devices were operated. The temperature and relative humidity in the chamber were maintained at 20 °C and 40%, respectively. The measurement interval was set to 1 min. The research project.

Measurement Devices
Three low-cost, sensor-based measurement devices were tested, which are currently used to measure fine dust levels in South Korea. These devices are manufactured by E3 (Yi Shan A4 sensor, SK techx, Seoul, Korea), MAXFOR Technology (Plantower PMS7003 sensor, SK techx, Seoul, Korea), and SK Techx (Plantower PMS7003 sensor, SK techx, Seoul, Korea). The sensors are referred to as '101', '201', and '301', respectively. Table 1 presents detailed specifications of each device and sensor type. Device 101 is an instrument shelter type with an implemented A4 sensor. Both 201 and 301 use the PMS7003 sensor of Plantower and present the same instrument shelter type as 101. Device 201 is designed to be operated by drawing air in through simple inhalation exposure.
To compare the trends of output values of the target measurement devices, using cost-effective sensors, regarding concentration interval in real time, a GRIMM (dust spectrometer from Grimm Technologies), based on multi-channel light scattering, was used as the reference device in a realtime test. Before the test, the correction factor (C-factor) of the reference device was calculated using a gravimetric method and by performing an equivalence test. In the gravimetric method, two minivolume air samplers were used in a chamber of specific dimensions. The C-factor was reflected in the reference device. Particles of Japan Dust Class 10 were used as the test particle for the equivalence test. The details of the equivalence test, including its results, have already been reported by Air Quality Sensor Performance Evaluation Center (AQ-SPEC) [19].

Test Method
The comparison test between the reference device and the target measurement devices using cost-effective particle sensors was performed in the following procedure: The chamber was ventilated using clean air generated by removing particles through a HEPA (high efficiency particulate air) filter (SK techx, Seoul, Korea). After the fine dust level was maintained at or below 2 µ g/m 3 in the chamber, clean air was injected and the ejection was stopped. Once the air flow was at a steady state, both the reference and target devices were operated. The temperature and relative humidity in the chamber were maintained at 20 °C and 40%, respectively. The measurement interval was set to 1 min. The research project.

Measurement Devices
Three low-cost, sensor-based measurement devices were tested, which are currently used to measure fine dust levels in South Korea. These devices are manufactured by E3 (Yi Shan A4 sensor, SK techx, Seoul, Korea), MAXFOR Technology (Plantower PMS7003 sensor, SK techx, Seoul, Korea), and SK Techx (Plantower PMS7003 sensor, SK techx, Seoul, Korea). The sensors are referred to as '101', '201', and '301', respectively. Table 1 presents detailed specifications of each device and sensor type. Device 101 is an instrument shelter type with an implemented A4 sensor. Both 201 and 301 use the PMS7003 sensor of Plantower and present the same instrument shelter type as 101. Device 201 is designed to be operated by drawing air in through simple inhalation exposure.
To compare the trends of output values of the target measurement devices, using cost-effective sensors, regarding concentration interval in real time, a GRIMM (dust spectrometer from Grimm Technologies), based on multi-channel light scattering, was used as the reference device in a realtime test. Before the test, the correction factor (C-factor) of the reference device was calculated using a gravimetric method and by performing an equivalence test. In the gravimetric method, two minivolume air samplers were used in a chamber of specific dimensions. The C-factor was reflected in the reference device. Particles of Japan Dust Class 10 were used as the test particle for the equivalence test. The details of the equivalence test, including its results, have already been reported by Air Quality Sensor Performance Evaluation Center (AQ-SPEC) [19].

Test Method
The comparison test between the reference device and the target measurement devices using cost-effective particle sensors was performed in the following procedure: The chamber was ventilated using clean air generated by removing particles through a HEPA (high efficiency particulate air) filter (SK techx, Seoul, Korea). After the fine dust level was maintained at or below 2 µ g/m 3 in the chamber, clean air was injected and the ejection was stopped. Once the air flow was at a steady state, both the reference and target devices were operated. The temperature and relative humidity in the chamber were maintained at 20 °C and 40%, respectively. The measurement interval was set to 1 min. The research project.

Measurement Devices
Three low-cost, sensor-based measurement devices were tested, which are currently used to measure fine dust levels in South Korea. These devices are manufactured by E3 (Yi Shan A4 sensor, SK techx, Seoul, Korea), MAXFOR Technology (Plantower PMS7003 sensor, SK techx, Seoul, Korea), and SK Techx (Plantower PMS7003 sensor, SK techx, Seoul, Korea). The sensors are referred to as '101', '201', and '301', respectively. Table 1 presents detailed specifications of each device and sensor type. Device 101 is an instrument shelter type with an implemented A4 sensor. Both 201 and 301 use the PMS7003 sensor of Plantower and present the same instrument shelter type as 101. Device 201 is designed to be operated by drawing air in through simple inhalation exposure.
To compare the trends of output values of the target measurement devices, using cost-effective sensors, regarding concentration interval in real time, a GRIMM (dust spectrometer from Grimm Technologies), based on multi-channel light scattering, was used as the reference device in a realtime test. Before the test, the correction factor (C-factor) of the reference device was calculated using a gravimetric method and by performing an equivalence test. In the gravimetric method, two minivolume air samplers were used in a chamber of specific dimensions. The C-factor was reflected in the reference device. Particles of Japan Dust Class 10 were used as the test particle for the equivalence test. The details of the equivalence test, including its results, have already been reported by Air Quality Sensor Performance Evaluation Center (AQ-SPEC) [19].

Test Method
The comparison test between the reference device and the target measurement devices using cost-effective particle sensors was performed in the following procedure: The chamber was ventilated using clean air generated by removing particles through a HEPA (high efficiency particulate air) filter (SK techx, Seoul, Korea). After the fine dust level was maintained at or below 2 µ g/m 3 in the chamber, clean air was injected and the ejection was stopped. Once the air flow was at a steady state, both the reference and target devices were operated. The temperature and relative humidity in the chamber were maintained at 20 °C and 40%, respectively. The measurement interval was set to 1 min. The To compare the trends of output values of the target measurement devices, using cost-effective sensors, regarding concentration interval in real time, a GRIMM (dust spectrometer from Grimm Technologies), based on multi-channel light scattering, was used as the reference device in a real-time test. Before the test, the correction factor (C-factor) of the reference device was calculated using a gravimetric method and by performing an equivalence test. In the gravimetric method, two mini-volume air samplers were used in a chamber of specific dimensions. The C-factor was reflected in the reference device. Particles of Japan Dust Class 10 were used as the test particle for the equivalence test. The details of the equivalence test, including its results, have already been reported by Air Quality Sensor Performance Evaluation Center (AQ-SPEC) [19].

Test Method
The comparison test between the reference device and the target measurement devices using cost-effective particle sensors was performed in the following procedure: The chamber was ventilated using clean air generated by removing particles through a HEPA (high efficiency particulate air) filter (SK techx, Seoul, Korea). After the fine dust level was maintained at or below 2 µg/m 3 in the chamber, clean air was injected and the ejection was stopped. Once the air flow was at a steady state, both the reference and target devices were operated. The temperature and relative humidity in the chamber were maintained at 20 • C and 40%, respectively. The measurement interval was set to 1 min. The standard powder (Japan Dust) was injected into the chamber and maintained at a certain level. Under this condition, the output values of the reference and target devices were obtained. The test was carried out by classifying the fine dust (PM 2.5 ) concentrations into five intervals: 40 µg/m 3 and below (1st interval), 40-80 µg/m 3 (2nd interval), 80-120 µg/m 3 (3rd interval), 120-160 µg/m 3 (4th interval), and 200-230 µg/m 3 (5th interval). The measurement was conducted for 150 min for each concentration interval to acquire the values of PM 2.5 and PM 10 . The obtained output values were utilized for data analysis.

Evaluation Method
For the output values of each device, descriptive statistics were initially calculated for each concentration interval. To correct the output values of the low-cost, sensor-based measurement devices, a linear regression analysis was conducted by classifying fine dust into PM 2.5 and PM 10 . In each interval of PM 2.5 and PM 10 , 75% of the data obtained were randomly extracted, and a linear regression equation was derived using the output values of the GRIMM device as dependent variables and those of the target devices as independent variables. The equation was validated by using the remaining 25% output values, which were not used to derive the equation.
The precision of a measurement device using a particle sensor should be evaluated based on accuracy and repeatability [20]. Accuracy is the closeness between output values of the target device and measurements of the reference device. Repeatability means dispersion [21]. As there was no correction curve for output values of the target devices, this study utilized linear correlations derived from the linear regression equation to evaluate the accuracy of the devices. In other words, when the output value of a target device was lower or upper than that of the reference device, the accuracy was evaluated to be low. Repeatability was identified based on the variation of the output value of a target device for each particle concentration interval. Finally, the correction factors of each target device, which were calculated using the linear regression equation, were applied to correct the remaining 25% of output values, which had not been used to derive the equation, and the results were compared with the output values of the reference device to evaluate the accuracy of each device under test. In other words, residuals between the output values of the reference device and the corrected values of each sensor-based device were obtained and then divided by the output values of the reference device. The average absolute value thus calculated was used to evaluate the accuracy of each sensor-based device after correction.

Results and Discussion
3.1. PM 2.5 Table 2 presents descriptive statistics of the output values of the reference device and the sensor-based devices for each PM concentration interval. For every interval, the output values of the reference device were higher than those of the target devices. This indicates that, if the target devices were used without correcting their output values, environmental pollution by PM 2.5 could be underestimated. Therefore, it was necessary to calculate an appropriate upward correction factor for each device and to evaluate the applicability of the devices after correction. Among the target devices, 201 showed the highest values of PM concentrations except for the 5th interval, followed by 101 and 301. In the 5th interval, 101 showed the highest output values, followed by 201 and 301. The output values of 201, which were generally higher than those of the other two devices, showed a similar increasing trend to that of the reference device from the 1st to 4th intervals but decreased in the 5th interval. Like the reference device, 101 showed an increase in output value for all the intervals; 301 also showed a similar increase except for the 4th interval. Accordingly, it appears sufficient to correct the output values of the target devices for each concentration interval instead of the entire concentration range. In addition, as 101 showed a more similarly increasing pattern to the reference device than did 201 and 301, along with the increase of concentration interval, it was evident that 101 maintained a stable sensor sensitivity for each concentration interval whereas the other devices (201 and 301) did not show a constant sensitivity across the intervals. This emphasizes the need to correct the output values of the cost-effective target devices for each concentration interval. Table 3 presents the results of the linear regression analysis that was conducted by randomly selecting 75% of the data in each concentration interval and using the output values of the reference device as dependent variables and those of 101, 201, and 301 as independent variables. The results show correction factors and correction functions of each device for each interval. Table 4 shows the relative standard deviations which were obtained by calculating residuals between the output values of the reference device and the corrected values of each target device from the remaining 25% of the data, subsequently dividing the residuals by the output values of the reference device. For 101, the correction functions of each concentration interval were 4.231x − 1.451 (1st interval), 1.212x + 49.447 (2nd interval), 0.161x + 105.825 (3rd interval), 0.276x + 33.960 (4tht interval), and 1.841x + 115.045 (5th interval) (x = PM 10 measurement value of each interval). The correction functions thus obtained were used to correct output values for each concentration interval. The relative standard deviations with respect to the output values of the reference device were calculated to be 6.7%, 0.6%, 0.3%, 1.7%, and 1.4% for the respective intervals. All deviations were below 10% and, except for the 1st interval, the deviations in the remaining intervals were even below 5%. This means that when 101 was corrected by the correction functions obtained in this study, its measurements of PM 2.5 concentration in the real atmosphere featured an error rate of less than 10%. Moreover, for PM 2.5 concentrations above 40 µg/m 3 , the accuracy of 101 improved with a relative standard deviation of less than 2%. For 201, the correction functions for each concentration interval were −0.364x + 41.098 (1st interval), −0.634x + 92.372 (2nd interval), formula 0.246x + 101.604 (3rd interval), 0.202x + 135.458 (4th interval), and 2.465x + 113.471 (5th interval). Based on these correction functions, the relative standard deviations were calculated to be 16.1%, 2.7%, 0.5%, 1.6%, and 0.7% for the respective intervals. Like 101, 201 showed an improvement of accuracy with a relative standard deviation of less than 3% for PM 2.5 concentrations exceeding 40 µg/m 3 . However, for PM 2.5 concentrations below 40 µg/m 3 , the accuracy was less reliable.
All the target measurement devices exhibited relative standard deviations of less than 10% with respect to the measurements of the reference device. When the level exceeded 40 µg/m 3 , their relative standard deviations were less than 4%, indicating an improvement of accuracy. Consequently, the measurement devices using cheap sensors are suitable for monitoring PM 2.5 levels in the air. In addition, their reliability is enhanced in areas or regions with PM 2.5 levels exceeding 40 µg/m 3 . Finally, among the target devices, 101 had the highest accuracy. Table 5 presents descriptive statistics of PM 10 outputs of the reference device and the target devices for each concentration interval that were set based on PM 2.5 levels. For every interval, like PM 2.5 , the output values of the reference device were higher than those of the target devices. This shows that, if the sensor-based devices were used without correcting their output values, environmental pollution by PM 10 could also be underestimated. Accordingly, it was necessary to calculate an appropriate upward correction factor for each device and to evaluate the reliability of the device after correction.

PM 10
Among the three sensor-based devices, 201 showed the highest PM concentration values except for the 5th interval, followed by 101 and 301. In the 5th interval, 101 had the highest output values, followed by 201 and 301. The output values of 201, which were higher than those of the other two devices, showed a similarly increasing trend to that of the reference device for the 1st to 4th intervals but decreased in the 5th interval. Like the reference device, 101 showed an increase of output values for all intervals; 301 showed a similar increase except for the 4th interval. The same conclusion as or PM 2.5 can be indirectly drawn, namely that the output values of the sensor-based devices needed to be corrected for each concentration interval rather than for the whole concentration range. In addition, as 101 showed a more similar pattern to the reference device than did 201 and 301 regarding increasing concentration intervals, it was evident that, unlike the other devices (201 and 301), 101 maintained a stable sensor sensitivity across all concentration intervals. This further emphasizes the need to correct the output values of the target devices for each concentration interval.  Table 6 presents the results of the linear regression analysis using 75% of the PM 10 data with the output values of the reference and target devices as dependent and independent variables, respectively. Table 7 shows the relative standard deviations, which were obtained by calculating residuals between the output values of the reference device and the corrected values of each sensor-based device from the remaining 25% of the data, subsequently dividing the residuals by the output values of the reference device.
For 101, the correction functions of each concentration interval were 8.055x − 11.983 (1st interval), 2.453x + 127.747 (2nd interval), 3.296x + 180.915 (3rd interval), 2.368x + 284.705 (4th interval), and 4.248x + 293.755 (5th interval) (x = PM 2.5 measurement value of each section). The correction functions thus obtained were used to correct output values for each concentration interval. The relative standard deviations with respect to the output values of the reference device were calculated to be 9.8%, 3.2%, 0.9%, 1.7%, and 1.2% in the respective intervals. All the deviations were below 10% and, except for the 1st interval, the deviations were even below 4%. This shows that when 101 is corrected by the correction functions obtained in this study, measurements of PM 10 concentration in the real atmosphere feature an error rate of less than 10%. Moreover, for PM 10 concentrations above 40 µg/m 3 , the accuracy of 101 improved with a relative standard deviation of less than 4%.   10 concentrations below 40 µg/m 3 in the real atmosphere. However, 301 also displayed an improvement of accuracy with a relative standard deviation of less than 4.5% for PM 2.5 concentrations above 40 µg/m 3 .
As for PM 2.5 , all target devices showed low accuracy with relative standard deviations exceeding 10% for PM 10 levels below 40 µg/m 3 . However, when the level exceeded 40 µg/m 3 , the relative standard deviations decreased to less than 5%, indicating an improvement of accuracy. Consequently, the target measurement devices are suitable for monitoring PM 10 levels in areas or regions with PM 10 levels exceeding 40 µg/m 3 . Among all target devices, 101 had the highest accuracy for PM 10 measurements.

Conclusions
As part of a government-sponsored research and development project to establish a smart management support system for pollutant discharge and prevention facilities, this study evaluated the reliability of sensor-based measurement devices to overcome limitations of existing methods. To this end, data from air pollution stations for monitoring air quality and managing discharge and prevention facilities were analyzed. Also, correction functions were derived by comparing cost-effective sensor-based measurement devices with a reference device, and the accuracy of corrected values, which were obtained by applying the functions, was examined.
The low-cost sensor uses the light scattering method for measurement and thus is lower in cost than the gravimetric of beta ray absorption devices. Moreover, as the low-cost sensor not only can measure fine dust levels on a minute time basis but is also small and light, it can be made in both a fixed type and portable type. However, this sensor tends to produce over-or under-measurements compared to other methods and the densities of fine dust emissions from various sources are not the same. For this reason, an error often occurs while the amount of scattering light is converted to a mass concentration. A correction formula has been developed to reduce such errors. If the correction formula is applied, as the proposed method is less expensive than other methods, the distances between each station may be reduced at a lower cost.
The evaluation showed that, under the condition of PM 2.5 concentrations exceeding 40 µg/m 3 , all the target devices (101, 201, and 301) had a relative standard deviation of less than 5% for both corrected PM 2.5 and PM 10 levels using the correction functions. It is postulated that most facilities involve PM 2.5 concentrations exceeding 40 µg/m 3 . In this regard, the above cost-effective sensor-based measurement devices are more reliable than the numerical analysis-based monitoring method. The conventional monitoring method uses data from national air pollution monitoring stations from discharge and prevention facilities at a long distance from each other. Therefore, data from such stations may not be very trustworthy for monitoring and controlling fine dust levels. Instead, sensor-based measurement devices, which are cost effective and easy to implement, may improve the development of a smart management support system for pollutant discharge and prevention facilities.