Next Article in Journal
Durability Tests of a Fiber Optic Corrosion Sensor
Previous Article in Journal
Scattering Removal for Finger-Vein Image Restoration
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Comparison of the Characteristics of Small Commercial NDIR CO2 Sensor Models and Development of a Portable CO2 Measurement Device

1
Plant and Environmental Sciences, Department of Environmental Health Science, University of Shizuoka, 52-1 Yada, Shizuoka 422-8526, Japan
2
National Institute for Agro-Environmental Sciences, 3-1-3 Kannondai, Tsukuba 305-8604, Japan
3
Institute for Environmental Sciences, University of Shizuoka, 52-1 Yada, Shizuoka 422-8526, Japan
*
Author to whom correspondence should be addressed.
Sensors 2012, 12(3), 3641-3655; https://doi.org/10.3390/s120303641
Submission received: 6 February 2012 / Revised: 6 March 2012 / Accepted: 13 March 2012 / Published: 16 March 2012
(This article belongs to the Section Chemical Sensors)

Abstract

: Many sensors have to be used simultaneously for multipoint carbon dioxide (CO2) observation. All the sensors should be calibrated in advance, but this is a time-consuming process. To seek a simplified calibration method, we used four commercial CO2 sensor models and characterized their output tendencies against ambient temperature and length of use, in addition to offset characteristics. We used four samples of standard gas with different CO2 concentrations (0, 407, 1,110, and 1,810 ppm). The outputs of K30 and AN100 models showed linear relationships with temperature and length of use. Calibration coefficients for sensor models were determined using the data from three individual sensors of the same model to minimize the relative RMS error. When the correction was applied to the sensors, the accuracy of measurements improved significantly in the case of the K30 and AN100 units. In particular, in the case of K30 the relative RMS error decreased from 24% to 4%. Hence, we have chosen K30 for developing a portable CO2 measurement device (10 × 10 × 15 cm, 900 g). Data of CO2 concentration, measurement time and location, temperature, humidity, and atmospheric pressure can be recorded onto a Secure Digital (SD) memory card. The CO2 concentration in a high-school lecture room was monitored with this device. The CO2 data, when corrected for simultaneously measured temperature, water vapor partial pressure, and atmospheric pressure, showed a good agreement with the data measured by a highly accurate CO2 analyzer, LI-6262. This indicates that acceptable accuracy can be realized using the calibration method developed in this study.

1. Introduction

Carbon dioxide (CO2) is a trace gas in the atmosphere that causes progressive global warming via the greenhouse effect. Since the first observation station for carbon dioxide was established in 1958 at Mauna Loa, Hawaii, the CO2 concentration has been measured globally. The CO2 concentration is increasing slowly but continuously with a typical seasonal fluctuation.

CO2 concentration in urban areas has also been monitored in order to quantify the CO2 emission from cities and to investigate the degree of its contribution to the regional carbon budget. Idos et al. [1] measured atmospheric CO2 concentration in Phoenix (AZ, USA), and found that the concentration at the center of the city (555 ppm) was greater than that in the surrounding rural area (370 ppm). This was attributed to anthropogenic CO2 emissions, in particular from vehicular exhaust (79.9%), at the center of the city. They termed the high CO2 concentration observed in the central regions of cities as the “urban CO2 dome”.

So far, urban CO2 has been monitored in many cities, including Chicago, IL, USA [2]; Edinburgh, UK [3]; Marseille, France [4]; Copenhagen, Denmark [5]; Tokyo, Japan [6]; Essen, Germany [7]; Mexico City, Mexico [8]; Basel, Switzerland [9]; and Cairo, Egypt [10]. Most of the urban CO2 data were obtained using an observation tower or a vehicle, such as an automobile.

However, the above studies using a few towers or one vehicle provide CO2 concentration data only for a small number of fixed points [10] or representative points along the route of the vehicle [7]. A relatively detailed distribution of CO2 concentration was obtained by moving a vehicle along various routes [1], but this takes several hours and may not show temporal CO2 distribution within a short period of time.

To collect CO2 concentration data from many points during a limited measurement period, it is necessary to employ a multi-point observation method using as many fixed stations or observation vehicles as possible. For multi-mobile observation, many CO2 sensors should be prepared and therefore, low-cost but accurate CO2 sensors are required. If bicycles or electric bikes are to be used for the observations, small and lightweight sensors are preferable.

In recent years, CO2 sensors made of semiconductors [11], solid electrolytes [12,13], optic fibers [14], laser diodes [15], and non-dispersive infrared (NDIR) detectors have been developed for monitoring CO2 concentration. For atmospheric CO2 concentration measurements, NDIR sensors are widely employed since they are stable and very robust against interference by other air components, including pollutants. The NDIR sensor also has excellent durability, and therefore, it seems to be the most reliable sensor for atmospheric CO2 measurement [16].

The output of NDIR sensors is affected by temperature, atmospheric pressure, and length of use [1719]. To enhance the precision and accuracy of NDIR sensors, it is important to correct the sensors’ outputs for these factors. So far, there have been several studies on the calibration of high-cost NDIR analyzers [1618,2022]. Only a few studies focused on the performance and calibration methods of small commercial CO2 sensors [2,19,23]. There is no report on the recently developed low-cost CO2 sensors produced by several manufacturers.

In the present study, to improve the precision and accuracy of recently developed low-cost CO2 sensors, a calibration method combining offset-correction and linear correction was examined. On the basis of the results of these performance tests, we have chosen the most suitable CO2 sensor for a multi-mobile measurement device. Using this sensor, we developed a portable CO2 measurement device that can measure and record temperature, humidity, air pressure, and GPS data as well as CO2 concentration. We evaluated the accuracy of this device by measuring the CO2 concentration in a school lecture room and by comparing it with the CO2 concentration measured by a highly accurate CO2 analyzer.

2. Experiments

2.1. Characterization of CO2 Sensors

2.1.1. CO2 Sensors Used for Experiments

Five commercial models of diffusion type NDIR CO2 sensors were purchased (GMM222C: Vaisala, K30: SenseAir, S100: ELT Co., Korea, AN100: Korea Digital Co. LTD., T6615: GE Sensing & Inspection Technologies). Their measurement ranges are 0–5,000 ppm, except for the GMM222C (0–2,000 ppm). GM222C is about 200 g in weight, 155 mm in probe length, and 18.5 mm in probe diameter, and is used as a reference sensor. The outer dimensions and weight for the other four sensors are less than 82 × 50 mm and 30g, respectively. Detailed information is provided in Table 1. Three individuals for each sensor model were used in the experiment.

2.1.2. Experimental Apparatus and Procedure

CO2 concentrations of standard reference gases were measured with the individual sensors, and the outputs of each sensor were recorded. Four samples of the prepared standard gas (CO2_ref) with CO2 concentrations of 0, 407, 1,110, and 1,810 ppm (N2-based, Taiyo Nippon Sanso Corporation, Japan) were used. To operate the 13 sensors simultaneously, the individual sensors were arranged in a small box (230 × 170 × 40 mm, 1.5 L) made of polyethylene (Figure 1). This was similar to the Dynamic Enclosure Approach method [23,24]. To control the ambient temperature of the individual sensors, the box was placed in a temperature-controlled incubator (DKM600, Yamato). Universal asynchronous receiver transmitter (UART) cables from the individual sensors were passed through a small hole made on the wall of the incubator and connected to a computer. The output shielded cables of the temperature sensors (LM35DZ, National Semiconductor) and humidity sensors (CHS-UPS, TDK) were also passed outside and connected to a data logger (CR1000, Campbell Scientific). The flow rate of the standard gases into the box was controlled with a flow meter (FS-25CO2, Yamato).

To evaluate the sensors’ output dependency on temperature and length of use, CO2 concentrations of standard CO2 gases were measured with the individual sensors for different temperatures and durations. Electrical power was supplied to the individual sensors 30 min before the start of the experiment. The standard CO2 gas was sent to the sensor box at a flow rate of 1 L/min.

According to the specifications, the response times of all the sensors were less than 2 min. In order to completely replace the air in the small box with standard CO2 gas, each standard CO2 gas was supplied for 5 min to allow the sensor outputs to stabilize. Thereafter, outputs were recorded every second for 3 min. The sensor outputs against the four standard CO2 gases were recorded at different temperatures of 10 °C, 25 °C, and 40 °C on 4 different days (1, 37, 106, and 306) after the start of use.

2.1.3. Cluster Analysis for Classifying CO2 Individual Sensors

All the sensor outputs for the standard CO2 gases were analyzed by the centroid method of cluster analysis to classify the isolates into different groups. The centroid method uses the notion of the cluster center, defined as the mean vector of the variables for all cases within the cluster. We investigated whether individual sensor outputs had similar patterns between model types and between individual sensors. We drew tree diagrams of the individual sensors by analyzing the outputs obtained at temperatures of 10 °C, 25 °C, and 40 °C on 1 day and 37 days after the start of the use.

2.1.4. Calibration Method of Sensors

Multipoint observation needs many sensors for simultaneous use. All the sensors should be calibrated, but calibrating all the sensors is a time-consuming process. A simple calibration method was examined not only for individual sensors of the same model, but also for each sensor model. In this study, we adopted a combination of zero offset and correction calculation.

First, we measured zero gas and obtained the offset value offsetobs. The corrected value (CO2_offset) was calculated by subtracting offsetobs from the sensor output (CO2_obs):

CO 2 _ offset = CO 2 _ obs offset obs

Temperature dependency of offsetobs of the sensors was investigated in advance and we confirmed that offsetobs was independent of temperature (10–40 °C) for all the sensor models. Dependency of offsetobs on length of use (1–306 days) was also checked and no clear dependency was observed.

Then CO2_offset was corrected for temperature, length of use, atmospheric pressure, and mixing ratio of water vapor (Xw) using Equation (2):

CO 2 _ correct = K std     C T     C day     C p     CO 2 _ offset / ( 1 X W )
where CO2_correct is the final concentration on a dry air base, and CT, Cday, and Cp are correction factors for temperature, length of use, and atmospheric pressure, respectively. Kstd is the averaged ratio of the three individual sensor outputs measured under standard conditions (25 °C, length of use: 1 day, and 1,013 hPa) to the true CO2 concentration of the standard gases.

These factors are linearly expressed by Equations (35):

C T = 1 + α T ( T 25 )
C day = 1 + α day ( day 1 )
and:
C p = 1 + α P ( P 1013 )
where T, day, and P are temperature, duration of use, and atmospheric pressure, respectively. αT, αday, and αP are linear coefficients for CT, Cday, and Cp, respectively. The terms αT and αday were experimentally determined, but αP was obtained from the sensor instruction manuals.

To determine αT and αday, the relative root mean squared error (RRMS error in %) of CO2_correct was calculated using:

RRMS error = Σ ( CO 2 _ correct CO 2 _ ref CO 2 _ ref ) 2 N × 100   ( % )
αT and αday were determined to minimize the RRMS error of CO2_correct.

2.1.5. Response Characteristics of Sensor Models

All the CO2 sensors used in this experiment were diffusion-type sensors. Because of natural convection in the cells and resistance against air movement caused by a dustproof filter, the output response of the sensors may be delayed after an actual change in CO2 concentration. To obtain the response data of the individual sensors, the transient change in sensor outputs was recorded when the CO2 concentration was changed from 0 ppm to 407 ppm.

First, standard gas with CO2 concentration of 0 ppm was supplied to the sensor box at a flow rate of 1.0 L/min for 10 min to equilibrate the individual sensor outputs. Thereafter, standard gas with CO2 concentration of 407 ppm was supplied at the same flow rate for 10 min. We defined t = 0 when CO2 concentration was changed from 0 ppm to 407 ppm. The individual CO2 sensor outputs were recorded every second, and the individual sensor responses against this drastic change in CO2 concentration were investigated.

The sensor output can be expressed in Equation (7) using an offset of time τ and a time constant for the response to equilibrium α. The values are empirically determined for sensors:

C t = C eq { 1 e α ( t τ ) }
where Ceq and Ct are the CO2 concentrations in equilibrium and at time t (s), respectively.

2.2. Experiments Using a Portable CO2 Measurement Device

2.2.1. Fabrication of a Portable CO2 Measurement Device

On the basis of the performance test results described in the Results section, we chose the K30 for developing a portable CO2 measurement device (Figure 2). A humidity and temperature sensor unit (SHT-71, Sensirion), a GPS sensor (GPS 18×, Garmin), and an atmospheric pressure sensor (SCP1000-D01, Akitsuki) were incorporated. All data can be automatically recorded on a Secure Digital (SD) memory card (Transcend, 2 GB) under the control of a microcomputer (ATmega2560, Atmel Corporation). The device is powered by 6 lithium-ion rechargeable batteries (eneloop AA cell battery, Sanyo).

2.2.2. Measurement of Indoor CO2 Concentration with a Portable CO2 Measurement Device

To evaluate the system performance, the portable CO2 monitoring device was placed in a room at the Shizuoka Prefectural Science and Technology High School (34.99W, 138.41E) in Shizuoka City, Japan. A widely used and highly accurate CO2 analyzer (LI-6262, Licor Co. Ltd.) was also used for comparison. The portable monitoring device and LI-6262 were operated concurrently for a day (2 February 2011) and the data were recorded every 30 s. The CO2 analyzer LI-6262 was calibrated with pure N2 and CO2 standard (407 ppm) gases following the instructions, before the measurement.

3. Results

3.1. Raw Data Obtained by CO2 Sensors

The raw outputs from the CO2 sensors for the four different CO2 concentrations of the standard gases were, of course, not the same as the standard gas concentrations. The output differences between the sensor individuals of an identical sensor model were less than 54% in K30, but larger (160%) in T6615, suggesting different intra-model variation for different models. In the case of K30 and AN100, their raw outputs increased linearly with CO2 concentrations (0.50 < r < 0.99 for K30), but the raw outputs of the other sensors increased nonlinearly (0.103 < r < 0.99). Moreover, the raw sensor outputs for all the models used in this study did not register 0 ppm for pure N2 gas, suggesting that offset correction is required.

3.2. Cluster Analysis

In our study, cluster analysis was conducted for the output of each sensor (Figure 3). Output patterns differed between sensor models. However, the same tendency was seen in the output patterns of individual sensors in identical models, suggesting that the same calibration method is applicable for the individuals of an identical model, while different calibration coefficients should be determined for different sensor models.

3.3. The Calibration Results

A two-point calibration method (combination of offset and span calibration) has been widely used for CO2 sensors, including the highly accurate CO2 analyzer LI-6262. Since the result of cluster analysis showed that the clustering patterns of individual sensor outputs of the same model were similar, it was proposed that the same method can be applied to calibrate individual sensors of the same model.

Since the CO2 standard gas was a mixture of dry air and CO2, XW can be assumed to be zero and Equation (2) can be modified as follows:

C T C day = CO 2 _ std / K std C p CO 2 _ offset
where CO2_std is CO2 standard gas concentration.

CTCday was calculated using Equation (8) for each measurement at a different temperature and for different length of use. CTCday was plotted against temperature and length of use, and the relationships between them were investigated. Figure 4 shows the relationship between CTCday for K30 and T. CTCday increased linearly with increasing T, and the values of the slope were not greatly different among data obtained for different lengths of use. Linear relationships were not observed in the case of AN100 and S100.

The relationship between CTCday and length of use (Figure 5) shows the linearity in the case of K30 and AN100. αT and αday were determined to minimize the RRMS error between CO2_std and CO2_correct obtained from Equation (2). RRMS errors calculated using αT and αday determined for individual sensors and for sensor models are listed in Table 2. RRMS errors determined for individual sensors were lower than those determined for sensor models, and the difference ranged from 4.8% to 23.6%. RRMS errors decreased in both cases with an increase in the number of correction factors considered in the K30 and AN100 sensor models, suggesting that these corrections successfully improve the accuracy of the sensors. In particular, K30 had the highest accuracy, comparable to the reference sensor GMM222C, even when the same coefficients were applied to three sensors of the K30 model (Figure 6). This result suggests that the coefficients determined for sensor models using all the data from three individual sensors can be applied for sensor correction for the K30 model.

RRMS errors determined for S100 and T6615 were increased with increasing calibration parameters (Table 2). Since the raw outputs of these sensors did not increase linearly with temperature and length of use, the calibration method might not be adequate for these sensor models.

3.4. Response Time of the Sensors

It is necessary to consider the response delay of the sensor output values against actual CO2 concentration change. Figure 7 shows the response of the output of the S100 sensor model. A time constant for the response to equilibrium α and an offset of time τ in response of each sensor model were calculated from Equation (7) (Table 3). Each sensor had a good agreement between the calculated curve fit and actual data when τ in response was employed. The 90% response time of all the sensors, excluding T6615, was <3 min, and these values were all larger than the catalog values. The responses of small CO2 sensors used in this study were slower than that of the reference sensor GMM222C. No clear relationship between the length of use and the sensor response was observed. No temperature dependency of α and τ was also confirmed (data are not shown).

3.5. Development of a Portable CO2 Measurement Device and Validation Test

The developed measurement device was compact (100 × 100 × 150 mm) and lightweight (900 g). Measured data can be displayed on a graphic 2.5 inch liquid crystal display in real-time. The device is powered by six lithium-ion AA batteries and can be operated for 5 h continuously. An observation interval and the coefficients required for the CO2 calibration can be input into an on-board microcomputer. The cost of production was about 60,000 yen or USD 770 per device.

The CO2 concentration in a lecture room of the Shizuoka Prefectural Science and Technology high school was measured every 30 s throughout a day (7 February 2011) using a high-precision CO2 analyzer (LI-6262) and the developed portable CO2 measurement device (Figure 2). Before the measurement, the portable CO2 monitoring device was corrected for offset value using the pure N2 gas. The sensor output of the portable CO2 measurement device was corrected for temperature, length of use, atmospheric pressure, and water vapor partial pressure.

Owing to the response delay of the sensor in the developed device, the measured CO2 data had to be time-shifted to the CO2 data obtained with the LI-6262. Subsequently, the RRMS difference of the portable CO2 measurement device against the LI-6262 was calculated. The RRMS difference was 3.5%, indicating a good agreement of the outputs between the high-precision CO2 analyzer and the developed device (Figure 8).

3.6. Discussion

A simple 2-point calibration method is widely used for many sensors, including CO2 analyzers. Other calibration methods such as a second-order polynomial equation [20] and a cubic spline function [25] have been applied in the case of CO2 sensors. In this study, the K30 and AN100 showed a good linear relationship between the sensor outputs and standard CO2 gas concentrations. These clear linear relationships were observed under most measurement conditions for different temperatures and lengths of use. The other two sensor models showed no clear relationship between the standard CO2 gas concentrations and their outputs.

When many CO2 sensors are used for a spatial distribution measurement of CO2 concentration within a limited time, the calibration of each sensor is a time-consuming process. A simplified calibration method is required. Pandey et al. compared CO2 data obtained with all sensor units manufactured by the same company, and found an excellent compatibility between them throughout the entire side-by-side analysis [3]. In our measurements, we found similar tendencies for individual sensor outputs within identical sensor models by cluster analysis. We also obtained acceptable RRMS errors, even when the coefficients for temperature and length of use were determined using all the data obtained from the three sensors of the same model. We have chosen the K30 sensor model to minimize RRMS errors and propose that, in practical measurements, only the offset calibration shown in Equation (1) should be conducted and then CO2_correct is calculated using αT (=0.00141) and αday (=0.000478) determined here. Our result suggests that, once the coefficients for the sensor model are determined, the coefficients can be applied for other individual sensors of the same model. This greatly reduces the effort required for calibration compared with conventional calibration.

The 90% responses of the CO2 sensors used in this study were slower than that of the reference sensor GMM222C (Table 3). This is probably caused by gas diffusion from one side only in the box-type IR cells of all the sensors compared with diffusion from both sides in the cylinder-type IR cell of GMM222C. The thickness and material of the dustproof filter of the sensors might also provide a resistance for gas diffusion.

Time drift [18] and temperature dependency [21] of the IR detector in CO2 sensors were reported to be important factors to be considered. Sega et al. [20] confirmed that there was little difference in the secondary polynomial curve for CO2 outputs of an analyzer observed for lengths of use of 1 month and 3 years. They proposed that the calibration of the analyzer should be conducted once a year. Apart from the above study, there are no reports investigating effect of the length of use on the output of CO2 analyzers. In this study, we found that K30 has a linear relationship with length of use and showed that the slope can be used to correct the sensor output for length of use. However, it should be noted that the coefficients determined in this experiment may not be valid if the sensor is used for over 1 year.

Our results indicate that temperature, pressure, and length of use independently affect the output of the sensors, and therefore, the calibration equation can be expressed as a product of correction terms for the three parameters. The RRMS error was lowered by using the three independent correction terms. When using several sensors simultaneously, the difference in the sensor outputs might result in significant measurement errors. We found that the sensor output tendency against environmental factors was similar between sensor individuals of the same sensor model, but different between sensor models. On the basis of this result, we could propose a method that corrects the outputs of several sensors of same model by using one coefficient each for temperature, air pressure, and length of use. This can eliminate the time-consuming process of calibration and reduce the sensor output error caused by using several sensors simultaneously for CO2 distribution measurement.

In recent years, CO2 measurement techniques based on on-road mobile laboratories have been applied in urban areas to determine the typical inhomogeneous spatial distribution of CO2 concentration and emission sources within urban areas [1,7,2630]. Most of the mobile laboratories were mounted on a car, and therefore, the measurement area is limited to roads. However, one can easily carry the portable CO2 measurement device developed in this study in a hand and measure CO2 concentration not only by using a vehicle but also by walking around with this device since it is lightweight and battery-operated. CO2 concentration, temperature, humidity, atmospheric pressure, and measurement time and location can be measured and recorded onto an SD memory card. The portable CO2 measurement device can be used to measure the CO2 concentration in urban forests, public greens, and industrial areas for scientific research and environmental education purposes.

4. Conclusions

When four samples of standard gas with different CO2 concentrations (0, 407, 1110, and 1810 ppm) were measured with four models of small commercial CO2 sensors, the outputs of the K30 and AN100 systems showed linear relationships with temperature and length of use. With an increase in the number of environmental factors considered for calibration, the accuracy of K30 and AN100 was improved. In particular, the accuracy of the K30 improved significantly, even when the same correction coefficients were used for three individual sensors of this model. Using the K30, we have developed a portable CO2 measurement device. Good agreement was obtained for the outputs between a high-precision CO2 analyzer and the developed device. This portable device allows measurements to be made while walking and cycling. The measurement device can be used for measuring the heterogeneity of CO2 distribution in urban areas and for environmental education.

Acknowledgments

This research was partially supported by the Ministry of Education, Science, Sports, and Culture, Grant-in-Aid for Scientific Research (B), No. 21310026 and Grant-in-Aid for Scientific Research on Innovative Areas No. 20120005. It was also supported by the Japan Society for the Promotion of the Science, A3 Foresight Program “CarboEastAsia”. The authors wish to thank Tatsushi Shiozawa (Institute for Environmental Sciences, University of Shizuoka, Japan) for the elucidating discussions. The authors are very grateful to the reviewers for their helpful and valuable comments.

References

  1. Idso, C.D.; Idso, S.B.; Balling, R.B. The urban CO2 dome of Phoenix, Arizona. Phys. Geogr 1998, 19, 95–108. [Google Scholar]
  2. Grimmond, C.S.B.; King, T.S.; Cropley, F.D.; Nowak, D.J.; Souch, C. Local-scale fluxes of carbon dioxide in urban environments: Methodological challenges and results from Chicago. Environ. Pollut 2002, 116, 243–254. [Google Scholar]
  3. Nemitz, E.; Hargreaves, K.J.; McDonald, A.G.; Dorsey, J.R.; Fowler, D. Micrometeorological measurements of the urban heat budget and CO2 emissions on a city scale. Environ. Sci. Technol 2002, 36, 3139–3146. [Google Scholar]
  4. Grimmond, C.S.B.; Salmond, J.A.; Oke, T.R.; Offerle, B.; Lemonsu, A. Flux and turbulence measurements at a densely built-up site in Marseille: Heat, mass (water and carbon dioxide) and momentum. J. Geophys. Res 2004, 109. [Google Scholar] [CrossRef]
  5. Soegaard, H.; Møller-Jensen, L. Towards a spatial CO2 budget of a metropolitan region based on textural image classification and flux measurements. Remote Sens. Environ 2003, 87, 283–294. [Google Scholar]
  6. Moriwaki, R.; Kanda, M. Seasonal and diurnal fluxes of radiation, heat, water vapor and CO2 over a suburban area. J. Appl. Meteorol 2004, 43, 1700–1710. [Google Scholar]
  7. Henninger, S.; Kuttler, W. Mobile Measurements of Carbon Dioxide within the Urban Canopy Layer of Essen, Germany. Proceedings of the 5th Symposium on Urban Environment, Vancouver, BC, Canada, 23–27 August 2004.
  8. Velasco, E.; Pressley, S.; Allwine, E.; Westberg, H.; Lamb, B. Measurements of CO2 fluxes from the Mexico City urban landscape. Atmos. Environ 2005, 39, 7433–7446. [Google Scholar]
  9. Vogt, R.; Christensen, A.; Rotach, M.W.; Roth, M.; Satyanarayana, A.N.V. Temporal dynamics of CO2 fluxes and profiles over a Central European city. Theor. Appl. Climatol 2006, 84, 117–126. [Google Scholar]
  10. Burri, S.; Frey, S.; Parlow, E.; Vogt, R. CO2 Fluxes and Concentrations over Urban Surface in Cairo/Egypt. Proceedings of the Seventh International Conference on Urban Climate, Yokohama, Japan, 29 June–3 July 2009.
  11. Fine, G.F.; Leon, M.; Cavanagh, L.M.; Afonja, A.; Binions, R. Metal oxide semi-conductor gas sensors in environmental monitoring. Sensors 2010, 10, 5469–5502. [Google Scholar]
  12. Mandayo, G.G.; Herrán, J.; Castro-Hurtado, I.; Castaño, E. Performance of a CO2 impedimetric sensor prototype for air quality monitoring. Sensors 2011, 11, 5047–5057. [Google Scholar]
  13. Wei, Q.; Luo, W.D.; Liao, B.; Liu, Y.; Wang, G. Study of the characteristics of Ag doped CuO-BaTiO3 CO2 sensors. Sensors 2002, 2, 166–179. [Google Scholar]
  14. Chu, C.S.; Lo, Y.L. Fiber-optic carbon dioxide sensor based on fluorinated xerogels doped with HPTS. Sens. Actuators B 2008, 129, 120–125. [Google Scholar]
  15. Takashi Asakawa, T.; Kanno, N.; Tonokura, K. Diode laser detection of greenhouse gases in the near-infrared region by wavelength modulation spectroscopy: Pressure dependence of the detection sensitivity. Sensors 2010, 10, 4686–4699. [Google Scholar]
  16. Wang, Y.; Nakayama, M.; Watanabe, K.; Yagi, M.; Nishikawa, M.; Fukunaga, M. The NDIR CO2 monitor with smart interface for global networking. IEEE Trans. Instrum. Meas 2005, 54, 1634–1639. [Google Scholar]
  17. Chen, S.; Yamaguchi, T.; Watanabe, K. A Simple, Low-Cost Non-Dispersive Infrared CO2 Monitor. Proceedings of the ISA/IEEE Sensor for Industry Conference, Houston, TX, USA, 19–21 November 2002; pp. 107–110.
  18. Wong, J.Y. Self-calibration of an NDIR gas sensor. U.S. Patent 5,347,474,. 13 September 1994. [Google Scholar]
  19. Mizoguchi, Y.; Ohtani, Y. Comparison of response characteristics of small CO2 sensors and an improved method based on the sensor response. J. Agric. Meteorol 2005, 61, 217–228. [Google Scholar]
  20. Sega, M.; Amico Di Meane, E.; Plassa, M. Stability of a NDIR analyzer for CO2 at atmospheric concentration. Ann. Chim 2002, 92, 897–902. [Google Scholar]
  21. Francey, R.J.; Steele, L.P. Measuring atmospheric carbon dioxide—The calibration challenge. Accreditation Qual. Assur 2003, 8, 200–204. [Google Scholar]
  22. Amico Di Meane, E.; Sega, M. Uncertainty estimation in chemical measurements: Carbon dioxide determination at atmospheric concentration. Ann. Chim 2003, 93, 583–587. [Google Scholar]
  23. Pandey, S.K.; Kim, K.H. The relative performance of NDIR-based sensors in the near real-time analysis of CO2 in air. Sensors 2007, 7, 1683–1696. [Google Scholar]
  24. Pandey, S.K.; Kim, K.H.; Lee, S.H. Use of a dynamic enclosure approach to test the accuracy of the ndir sensor: Evaluation based on the CO2 equilibration pattern. Sensors 2007, 7, 3459–3471. [Google Scholar]
  25. Francey, R.J.; Steele, L.P. Measuring atmospheric carbon dioxide--the calibration challenge. Accreditation Qual. Assur 2003, 8, 200–204. [Google Scholar]
  26. Shorter, J.H.; Herndon, S.; Zahniser, M.S.; Nelson, D.D.; Wormhoudt, J.; Demerjian, K.L.; Kolb, C.E. Real-time measurements of nitrogen oxide emissions from in-use New York City transit buses using a chase vehicle. Environ. Sci. Technol 2005, 39, 7991–8000. [Google Scholar]
  27. Bukowiecki, N.; Dommen, J.; Prevot, A.S.H.; Richter, R.; Weingartner, U.; Baltensperger, U. A mobile pollutant measurement laboratory-measuring gas phase and aerosol ambient concentrations with high spatial and temporal resolution. Atmos. Environ 2002, 36, 5569–5579. [Google Scholar]
  28. Kolb, C.E.; Herndon, S.C.; McManus, J.B.; Shorter, J.H.; Zahniser, M.S.; Nelson, D.D.; Jayne, J.T.; Canagaratna, M.R.; Worsnop, D.R. Mobile laboratory with rapid response instruments for real-time measurements of urban and regional trace gas and particulate distributions and emission source characteristics. Environ. Sci. Technol 2004, 38, 5694–5703. [Google Scholar]
  29. Park, J.S.; Cho, H.C.; Yi, S.H. NDIR CO2 gas sensor with improved temperature compensation. Proc. Eurosens 2010, 5, 303–306. [Google Scholar]
  30. Sugawara, S. “Urban CO2-island” observed in the central area of Sendai city. Bull. Miyagi Univ. Educ 2008, 43, 79–87. [Google Scholar]
Figure 1. Experimental apparatus.
Figure 1. Experimental apparatus.
Sensors 12 03641f1 1024
Figure 2. Overview of a portable CO2 measurement device.
Figure 2. Overview of a portable CO2 measurement device.
Sensors 12 03641f2 1024
Figure 3. Cluster analysis using the data obtained on 25 February 2010. Air temperatures were 10 °C, 25 °C, and 40 °C, length of use was 1 day. A durable solid electrolyte CO2 sensor CDM4160 (Figaro Engineering Inc.) was also included in the analysis.
Figure 3. Cluster analysis using the data obtained on 25 February 2010. Air temperatures were 10 °C, 25 °C, and 40 °C, length of use was 1 day. A durable solid electrolyte CO2 sensor CDM4160 (Figaro Engineering Inc.) was also included in the analysis.
Sensors 12 03641f3 1024
Figure 4. Relationship between temperature and CTCday in the case of K30 (n = 3).
Figure 4. Relationship between temperature and CTCday in the case of K30 (n = 3).
Sensors 12 03641f4 1024
Figure 5. Relationship between CTCday and length of use in the case of K30 and AN100 (n = 3).
Figure 5. Relationship between CTCday and length of use in the case of K30 and AN100 (n = 3).
Sensors 12 03641f5 1024
Figure 6. Average RRMS errors using coefficients determined for sensor models.
Figure 6. Average RRMS errors using coefficients determined for sensor models.
Sensors 12 03641f6 1024
Figure 7. Time course of the output of S100 sensor model. Solid line shows actual sensor output value. Dotted line shows CO2 concentration estimated using α and τ.
Figure 7. Time course of the output of S100 sensor model. Solid line shows actual sensor output value. Dotted line shows CO2 concentration estimated using α and τ.
Sensors 12 03641f7 1024
Figure 8. Comparison of CO2 concentration measured with the portable CO2 measurement device and LI-6262 on 7 February 2010.
Figure 8. Comparison of CO2 concentration measured with the portable CO2 measurement device and LI-6262 on 7 February 2010.
Sensors 12 03641f8 1024
Table 1. Catalog specifications of five commercial diffusion type NDIR CO2 sensors.
Table 1. Catalog specifications of five commercial diffusion type NDIR CO2 sensors.
SensorGMM222CK30S100AN100T6615
manufacturerVaisala a cylinder with 18 mm diameter and 140 mm length *SenseAirELTKCDGE sensing
L×W×D (mm)51 × 57 × 1433 × 33 × 1382 × 45 × 1857 × 35 × 15
Weight (g)22017102917
Measurement range (ppm)0–2,0000–5,0000–5,0000–5,0000–5,000
Accuracy30 ppm + 2% of reading30 ppm + 5% of reading30 ppm +5% of reading200 ppm + 3% of reading75 ppm or 10% of reading
Response Time (s)30 (63%)20 (63%)60 (90%)30 (63%)<120 (90%)
Operating voltage (V)11–20 VDC4.5–14 VDC5.0–5.5 VDC8–14 VDC5 VDC

*Size of a probe housing.

Table 2. Average RRMS errors using coefficients for temperature, length of use, and atmospheric pressure.
Table 2. Average RRMS errors using coefficients for temperature, length of use, and atmospheric pressure.
Sensor ModelK30S100AN100T6615

Individual Coefficient*Model Coefficient **Individual Coefficient *Model Coefficient **Individual Coefficient *Model Coefficient **Individual Coefficient *Model Coefficient **
temperature7.37.79.610.513.020.420.520.8
temp + day4.55.011.112.015.316.518.323.6
temp + day + pressure4.44.811.512.914.514.617.323.3

*The average RRMS error determined using coefficients determined for individual sensors.**The average RRMS error determined using coefficients determined for sensor models. Same coefficients were used for the calibration of three sensors.

Table 3. Offset of time τ, time constant for the response to equilibrium α and 90% response time for sensors.
Table 3. Offset of time τ, time constant for the response to equilibrium α and 90% response time for sensors.
Sensor ModelGMM222CK30S100AN100T6615
Length of Use (days)373063730637306373061200
α (s−1)0.0550.0140.045 ± 0.0260.024 ± 0.0040.014 ± 0.0030.0380.015 ± 0.0040.019 ± 0.0060.0215 ± 0.0040.014 ± 0.003
τ (s)5.72.66.1 ± 1.84.8 ± 2.612.7 ± 1.216.713.5 ± 3.714.5 ± 7.414.2 ± 10.019.4 ± 18.0
90% Response Time (s)47.587.887.8 ± 52.1102.9 ± 18.9175.3 ± 38.277.8151.0 ± 28.1141.6 ± 39.2124.2 ± 26.2190.4 ± 38.8

Measured at 25 °C, n = 3.

Share and Cite

MDPI and ACS Style

Yasuda, T.; Yonemura, S.; Tani, A. Comparison of the Characteristics of Small Commercial NDIR CO2 Sensor Models and Development of a Portable CO2 Measurement Device. Sensors 2012, 12, 3641-3655. https://doi.org/10.3390/s120303641

AMA Style

Yasuda T, Yonemura S, Tani A. Comparison of the Characteristics of Small Commercial NDIR CO2 Sensor Models and Development of a Portable CO2 Measurement Device. Sensors. 2012; 12(3):3641-3655. https://doi.org/10.3390/s120303641

Chicago/Turabian Style

Yasuda, Tomomi, Seiichiro Yonemura, and Akira Tani. 2012. "Comparison of the Characteristics of Small Commercial NDIR CO2 Sensor Models and Development of a Portable CO2 Measurement Device" Sensors 12, no. 3: 3641-3655. https://doi.org/10.3390/s120303641

Article Metrics

Back to TopTop