Verification of CO 2 Sensor Performance

The performance of the SprintIR CO2 sensor did not meet the specification declared

on the datasheet provided by the manufacturer (GSS). The response time of the sensor was not sufficient for breath measurement (14.1 s measured compared to the 1.2 s specified [21]) and the accuracy was poor (70 % CI barely sufficient to meet 1.20 % target accuracy). To verify whether the design of the system (e.g. chamber, gas mixing profile) affected the performance, the affordable commercial device was tested against a relatively expensive reference sensor (Hummingbird IR3107 CO2 sensor [22]).

The CO2 sensors were exposed to CO2 in the range of 0 to 4 %, generated using a gas

testing rig and 5 % cylinder of CO2, diluted with synthetic air where required. The

total flow rate was kept constant at 0.5 SLPM to avoid damage to the expensive commercial sensor. The test pattern had 1 % steps of CO2, for an interval of 1 minute

each, returning to a baseline of synthetic air. The Hummingbird commercial sensor was connected downstream of the SprintIR sensor to ensure the response time comparison was altered only by the operation of the sensors and not by the gas mixing time. Fig. 4.14 shows the output from both the commercial devices, where a median filter was applied to each device at 5 sample points (data logged at 100 Hz).

The IR3107 sensor demonstrates that the level of CO2 generated by the gas rig is likely

to be accurate and only a minor delay in the response time shown by the SprintIR sensor can be attributed to the design of the sensor chamber. The spurious peaks shown at the switching points from synthetic air to low percentage concentrations of CO2 for

the Hummingbird output are due to the sudden switching of the MFCs, which cause a momentary increase in CO2 concentration in the chamber (< 2 s). This effect could be

reduced by reconfiguring the PID control setup in the MFCs, but was left constant so as not to disrupt the performance of the gas testing system for higher gas concentration mixtures.

Fig. 4.14 – Comparison between the SprintIR and the IR3107 gas sensors for 1 minute pulses of CO2 in the range of 0 to 4 %.

The SprintIR sensor underestimates the concentration of CO2 present, by approx. 7.5

% in some measurements (e.g. peak values on average of 3.71 % were found when exposed to 4.0 % CO2 concentration). In the parallel measurements, the Hummingbird

sensor measures approximately 4.0 %, although some level of noise is present in the output (~0.05 % variance). The level of noise present in the SprintIR sensor output is ~0.1 % CO2. The device was calibrated prior to the experiment, to 0 % CO2 in air and

nitrogen. The device produces consistent measurements, if the excessive level of noise is filtered. For example, the average output after 1 minute of exposure to 2 % CO2 was

1.60 % for the first step and 1.59 % for the second step.

The main factor preventing the use of the SprintIR in a breath-by-breath analyser is the inadequate response time. To verify if any sampling techniques could be implemented to capture the end of an exhalation and hold it for a short period (e.g. sample alternate exhalations) the 𝑡90response time was measured for both the GSS and Hummingbird sensors. To avoid the level of noise falsifying the 90 % of final value measurement point, bi-exponential curves, Figs. 4.11, were fitted to the output response from the instant the step change in gas concentration change occurred. A 20 s window was taken to enable the best exponential curve fitting, where it was assumed that the sensor would be likely reach its 𝑡90 value within 15 s. Fig. 4.15 a) and b) show

the output from the models for the GSS SprintIR and Hummingbird IR3107 sensors respectively, for the fifth pulse of CO2 (as shown in Fig. 4.14), to a 4 % CO2 input

concentration. The calculated 𝑡₁₀ to 𝑡₉₀ times for each step (from Fig. 4.14) are shown in Table 4.3. This period indicates the time taken for the sensor output to alter from 10 % of the final output value to 90 % of the final output value.

Fig. 4.15 – Bi-Exponential models fitted to 20 s output window of a) Affordable commercial SprintIR sensor and b) Relatively expensive commercial Hummingbird

sensor.

Table 4.3 – Response times calculated for the turn-on phase for both commercial CO2

devices. Step 1 % 2 % 3 % 4 % 4 % 3 % 2 % 1 % SprintIR 𝒕_𝟏𝟎 to 𝒕_𝟗𝟎 [s] 17.95 38.25 56.69 58.45 57.57 61.52 42.71 41.04 IR3107 𝒕_𝟏𝟎 to 𝒕_𝟗𝟎 [s] 0.55 1.00 1.11 1.33 1.84 2.43 0.96 1.33

The 𝑡₁₀to 𝑡₉₀response times were calculated to investigate the response time of the device, without considering the initial mixing time of the chamber and avoiding any baseline noise. The exponential fit for the SprintIR sensor required extending to a 60 s range, to allow response times above 20 s to be calculated. The disappointing performance of the SprintIR sensor was consistent throughout the measurements, where the response time improved for lower concentrations, but never approached the

quick response required for breath analysis. The Hummingbird device was slower (~1.5 s) than specified in the datasheet (0.1 s) [22], however the flow conditions were not optimised for this sensor (as previously detailed, the SprintIR sensor was connected in the chamber prior to the IR3107). Considering the flow rate (0.5 SLPM) and the volume (gas pipes and chambers) that delayed the mixing of the gas, the IR3107 responded with reasonable speed for breath-by-breath analysis. However, the relative expense of the device means that it cannot be included in a low-cost breath analyser.

The output sensor responses shown in Fig. 4.15 are both after exposure to 4 % CO2;

the SprintIR device only reaches ~3 % after the 20 s period (whereas the IR3107 reaches its final output ~4 % after < 2 s). The poor performance from the affordable sensor demonstrates the need for development of a reliable, affordable and fast CO2

sensor for use in projects such as breath analysers, safety sensors and portable equipment. The design and construction of a prototype CO2 sensor designed for

breath-by-breath analysis is detailed in the chapter IV.