Benefits of virtual sensors for air quality monitoring in humid conditions

The gas sensing mechanisms, response, and behaviour of a real and a virtual solid-state chemical gas sensor operating either in static or in dynamic mode have been compared. The analysis was done by exposing simultaneously both sensors to different concentrations of various volatile organic compounds diluted in dry, as well as humid, synthetic air. The results revealed similar responses and behaviours for both types of measurement modes when the sensors were exposed towards single gas compounds, but a sensitivity enhancement in measurements comprising mixtures of gases when the sensors were operated in dynamic mode. The method used is able to overcome surface saturation problems and is beneficial for applications where mixtures of gases diluted in relative humidity are present.


Introduction
The concept of using an array of sensors to improve the selectivity of chemical gas sensors by emulating the human olfactory system was developed by W.F. Wilkens and J.D. Hartman during the 1960s [1].This strategy, called electronic nose [2][3][4], has been implemented in several different ways until nowadays, either by using an array of real sensors based on different sensing materials operating in static mode [5][6][7], that means constant temperature or bias, or by using a single sensor in dynamic operation mode [8][9][10], e.g., working in repeated cycles that include a customized range of temperature variations, which is equivalent to an array of virtual sensors [11].The choice of the temperature range depends on the gas type and its concentration to achieve good sensitivity [12].
The benefits of using a virtual sensor array in comparison to an array of real sensors are notable in terms of cost, robustness, ease of fabrication and integration, as this method allows to use one real sensor instead of several.In contrast, the data treatment and the interpretation of results in dynamic operation mode are more complex when compared to static mode.However, despite this added complexity, the dynamic operation strategy provides a selectivity enhancement that cannot be achieved with static operation.For this reason, if the purpose is to improve the selectivity of a chemical gas sensor avoiding modifications on the sensing material or on the gas sensing system, the choice is clear.
In this work, we aim to quantify further advantages and differences in sensitivity and gas sensing mechanisms between the two operating modes through a direct comparison of two chemical gas sensors fabricated with the same characteristics.One of the sensors was characterized at a constant temperature of 300 • C, in static mode, and the other via customized temperature cycled operation (TCO) between 240 and 360 • C, in dynamic mode.
The gas sensors utilized to perform this study are iridium-gated silicon-carbide-based field effect transistors (SiCFETs).This sensor technology has been demonstrated as a suitable technology for highly sensitive detection of a variety of different gas molecules such as H 2 , O 2 , CO, NH 3 , H 2 S, and a long list of hydrogen-containing gases including volatile organic compounds (VOCs) [13].In particular, the iridium (Ir) gate has been demonstrated to promote high sensitivity towards VOCs, good stability and repeatability as well as excellent detection limits down to parts per trillion (ppt) levels [14].A detailed study of the surface morphology and potential of the nanostructured porous Ir before and after two-week exposure to VOCs at high temperatures can be found in the literature [14].However, these sensors, as many other chemical gas sensors, lack selectivity.This drawback can be overcome with the implementation of TCO and multivariate statistics.
VOCs can be found in a large variety of everyday products such as solvent-based paints or coatings, varnishes, wax, printing inks, household cleaning, disinfecting, degreasing, and petroleum products [15].The use of VOC-containing products causes the release of air pollutants that may harm human health, have a negative impact on the environment, or accelerate the degradation process of sensitive materials.In the context of conservation of museum collections in storage or on display, the main internally generated VOCs, which are known to accelerate material change, are acetic acid (CH 3 COOH), formic acid (CH 2 O 2 ), and formaldehyde (CH 2 O) [16].In many cases, the relative humidity (RH) values establish a threshold limit above which the degradation occurs or become much more evident [17].It is therefore important to develop a method that can measure accurately the abovementioned VOCs while controlling the variations of RH, overcoming the influence of RH on the gas measurements.To reach this goal, we tested our sensors towards individual and mixtures of the abovementioned VOCs diluted in different RH percentages, to mimic a museum environment.In a more holistic view, the results here presented provide a possible path to overcome problems of saturation in chemical gas sensors when several gases are present in the surroundings.

Material and methods
SiCFETs were fabricated on top of an n-type 4H-SiC substrate by growing an epitaxial p-type active layer of about 1 μm of thickness (doping level of 5•10 16 cm − 3 ), and an n-type drain and source.To create a gate passivation layer, low pressure chemical vapor deposition (LPCVD) was used to deposit SiO 2 /Si 3 N 4 /SiO 2 of 50/25/ ∼ 5 nm.The ohmic contacts were fabricated by removing the passivation layer from the top of the source and drain regions, and the subsequent deposition and rapid thermal annealing at 950 • C during 5 min in argon (Ar) atmosphere of 50 nm nickel (Ni) layer.Metal contacts to source and drain were performed via titanium (Ti) and platinum (Pt) sputter deposition of 10 and 400 nm thickness.Gate contacts, separated by 5 μm from the drain/source contacts, were deposited via DC magnetron sputtering with nanostructured porous Ir with a total thickness of about 30 nm.The porous Ir on the gate region acts as a catalytic metal and allows gas sensing mechanisms.A more detailed explanation of the fabrication procedure can be found in the literature [18].
To perform the electrical and gas sensing characterization, the chips, containing two SiCFETs each, of ∼ 2 mm × 2 mm size were glued on top of a ceramic heater (Heraeus Sensor-Nite GmbH, Germany), that provides the needed temperature to promote the gas sensing reactions of interest, and next to a Pt-100 temperature sensor, to ensure the temperature monitoring.The heater and the Pt-100 were mounted on top of a 16-pin TO8 header, and spot welded to allow electrical access.The gate, drain, and source of both SiCFETs of each chip were electrically contacted via gold wire bonding.
The electrical and gas sensing characterizations were performed with the sensors mounted in custom-made stainless-steel chambers of about 3 mL volume connected to a gas mixing system with five Bronkhorst Mass-Flow Controllers.Electrical measurements were done with a Python software specially developed to control an electronic board (3S GmbH, Germany), that allow to simultaneously measure the current of the SiCFETs, control the voltage applied to the heaters, and read the temperature reported by the Pt-100 installed next to each chip.
The gas measurements were performed with four sensors working at the same time: two SiCFETs, one working at constant temperature and one in TCO mode, a metal-oxide VOC sensor, SGPC3 (Sensirion), to measure the total VOC concentrations, and a digital humidity sensor, SHTC1 (Sensirion), to control the humidity level.A cheap and disposable top for the stainless-steel gas test chamber was designed and fabricated using a Formlabs Form 3 SLA 3D Printer (Fig. 1).The part of the gas chamber where the sensors were located, and in contact with the studied VOCs, was sealed with polytetrafluoroethylene (PTFE, trade name Teflon®), which is known to be exceptionally chemically inert, highly hydrophobic, and a good insulator, to ensure airtightness and minimize any possible chemical interactions with the disposable lid material [19].The SHTC1 sensor covers a measurement range of 0-100 % RH with an accuracy of ±3 %, whereas the SGPC3 sensor can measure a variety of VOCs from 0.3-30 ppm with an accuracy of ±15 %.Additionally, the contribution from RH in the SGPC3 sensor signal can be subtracted from the VOC measures when both sensors are operated simultaneously.Both sensors were used as reference for the VOC tests as well as to calibrate the gas measurement system before starting the tests.
The measurements were performed with all sensors in gas chambers connected in series to the gas stream.
The sensors were characterized by exposing them towards different concentrations of CH 2 O, CH 2 O 2 and CH 3 COOH, ranging from 0.25 to 3 ppm, diluted in synthetic air (SA) under different percentages of RH, ranging from dry to 55 %.The sensor behaviour was studied as well, exposing them to the mentioned concentrations of CH 2 O 2 and CH 3 COOH diluted in dry air and with backgrounds of different concentrations of CH 2 O, ranging from 0.25 to 1 ppm.The exposures to the different concentrations studied were always done alternating values to avoid an effect of accumulated adsorbents or poisoning, that could lead to false conclusions.The RH was added by deviating a part of the nitrogen flow through a bubbler.The gas flow corresponding to each RH concentration was calibrated with the SHTC1 sensor at 20 • C before testing the sensors.For all the gas measurements, a total constant flow of 100 mL/min was kept.The gas measurements include: (i) allowing the device to stabilize the baseline at the operation temperature for 240 min; (ii) 20 min of exposure towards the studied gas; and (iii) 40 min in SA to recover the baseline after the gas exposure.
During the gas measurements, the SiCFETs were maintained in active or saturation regime, which is used with field effect transistors when amplification is needed and was demonstrated to promote the largest responses to gas exposures [20], by applying a drain-source voltage (V DS ), equal to 4 V.No voltage was applied to the gate-source voltage (V GS ) and the current flow was measured to assess the response of the sensors in the different studied conditions.The relative response of the sensors towards adsorbing species is defined as the relative change in the sensor's current in absolute value, which is expressed as: where I air corresponds to the sensor's current value when exposed to SA, including RH or CH 2 O backgrounds when present, and I gas corresponds to the sensor's current value when the mentioned gases of interest are diluted in the gas flow.
The SiCFET working at a constant temperature (real sensor) was maintained at 300 • C during all measurements.This temperature has been demonstrated to be, with a small offset due to use of different electronic systems, sensitive to concentrations down to ppt for different VOCs in the case of Ir-gated SiCFETs [14].The SiCFET working in TCO mode, was cycled through five different temperatures: 360 • C, 330 • C, 300 • C, 270 • C, and 240 • C, maintaining for 22 s each temperature step.This step time was chosen to maintain the cycles as short as possible, and long enough steps to allow the sensor signal to stabilize after each temperature change.Short cycles will allow, in the daily-life applications mentioned, to implement TCO and multivariate statistics with relatively short times of performance.In total, one cycle lasted 110 s, and the cycles were continuously repeated during all measurements.The signal of the virtual sensor was deduced from the signal acquired in TCO mode by gathering the values corresponding to 300 • C. For each cycle, at this temperature, the data points of the last seconds of the sensor signal were concatenated to produce an equivalent virtual sensor signal.

Results and discussion
The acquired signal corresponding to the real sensor and the calculated signal for the virtual sensor during one of the gas measurements are shown in Fig. 2. For both signals a Savitzky-Golay [21] filter was applied.This filtering technique is versatile and can be easily adapted to specific needs.It reduces noise while maintains the shape and height of peaks, thus, reporting few losses of information while minimizing non-desired contributions.This fact promotes very good outcomes in comparison to other smoothing or filtering techniques.
The summarized responses of the real and virtual sensors operating at 300 • C when exposed towards different concentrations of CH 2 O, CH 2 O 2 and CH 3 COOH diluted in dry SA are shown in Fig. 3.
It can be observed that, for both types of measurements, for increasing concentrations the response increased, as well as that CH 2 O reported the highest response values and CH 3 COOH the lowest.Therefore, the real and the virtual sensors presented the same behaviour when exposed towards individual VOCs.
The reason of the different sensitivities observed towards the different studied gases can be explained by using the kinetic theory of gases and the Hertz-Knudsen equation [22], which predicts that the number of gas molecules impinging a solid surface is directly proportional to the pressure and inversely proportional to the root square of the temperature and the molecular mass.As the temperature and pressure were maintained constant during the gas measurements, the only varying factor was the molecular mass of each gas studied.The weights of the studied gas molecules are m CH2O = 30.031g/mol, m CH2O2 = 46.030g/mol and m CH3COOH = 60.052g/mol.Thus, the heaviest molecular weight of the three gas species is m CH3COOH and the lightest is m CH2O .As the relation is inversely proportional, the difference in molecular weights may explain the highest response towards CH 2 O, and the lowest towards CH 3 COOH.The sensitivity of the SGPC3 used as a reference reported the same behaviour.Based on this result, it is most likely that this behaviour can be extrapolated to other chemical gas sensors sensitive to VOCs, either operating at a constant temperature or in TCO mode.
The summary of the responses calculated from the measurements performed with a SiCFET working at 300 • C when exposed towards different concentrations of CH 2 O diluted in SA with different percentages of RH as background is shown in Fig. 4a.The responses show a clear decreasing tendency with increasing presence of RH for the studied CH 2 O concentrations.A possible explanation for this observation is that, as the RH was introduced before the exposure toward CH 2 O pulses, the water molecules have already gone through an adsorption process in the gate surface and have occupied the adsorption sites when CH 2 O was introduced in the gas test chamber.Therefore, the adsorption sites that CH 2 O had access to decreased with increasing RH percentage, and, thus, the response toward CH 2 O decreased.Most likely, the gate surface was not completely saturated with water molecules and this may explain why the response increased with increasing CH 2 O concentration.Fig. 4b shows the results for the same measurements in case of the virtual sensor operating at 300 • C. Here, the trend is very similar to the real sensor at the beginning, but, for values higher than 10 % RH, the response increases.The relative response increases with increasing CH 2 O concentration as well.
Therefore, the general trend in this case is to increase with RH concentration, instead of the decrease observed when the sensor worked at a constant temperature.This different behaviour can be related to the continuous temperature change during the temperature cycles, that promotes the desorption and consequent surface cleaning of the gate's sensing material when the temperature is increased up to 360 • C, at the beginning of each cycle.Thus, a desorption process is promoted, and the adsorption sites are available at the beginning of each temperature cycle for both RH and CH 2 O, every 110 s.In this situation, the CH 2 O molecules have higher probability to be adsorbed when competing with RH in comparison to the constant temperature situation.Instead of finding the adsorption sites already occupied, a new competition starts after every cycle, and temperature step.In conclusion, an opposite behaviour was observed for the virtual sensor.
As a quantified measure of comparison, the percentage of response change were calculated for both sensors, regarding the exposure to 1 ppm CH 2 O diluted in dry SA or in humid SA with 55 % RH.The values obtained are a response decrease of about 59 % for the real sensor, and a response increase of about 31 % for the virtual sensor.
Regarding the measurements towards CH 2 O 2 and CH 3 COOH diluted in SA with different percentages of RH, a very similar behaviour as for CH 2 O was observed for the real sensor, as shown in Fig. 5a for CH 3 COOH.Increasing RH values decrease the response observed,  In case of the virtual sensor operating at 300 • C exposed to CH 2 O 2 and CH 3 COOH diluted in SA with different levels of RH, a similar tendency as in case of CH 2 O under different RH percentages was observed, as shown in Fig. 5b for CH 3 COOH.It can be observed that the response increased with increasing concentration of CH 3 COOH.Moreover, an initial increase at 5 %RH was observed, and, after a decrease at 10 %RH, the trend was to increase for values from 10 % to 35 %RH.Finally, a certain stabilization regime was achieved for values higher than 35 % RH.This behaviour is, again, very different when compared to the behaviour of the real sensor working at a constant temperature, and the differences observed can be attributed to the TCO mode.The higher increase in the response, when compared to the case where the sensor was exposed to CH 2 O and different percentages of RH, can be related to the differences among molecular weights of the gases studied.The water vapor molecules, with a weight of m H2O = 18.015 g/mol, are much lighter than the CH 3 COOH molecules, and, thus, promoted higher response of the sensor when the competition to occupy available adsorption sites started, after every cycle and temperature step.
In this case, the response of the real sensor decreased by 85 % when the sensor was exposed to 1 ppm CH 3 COOH diluted in dry SA in relation to the same concentration but diluted in SA with a 50 %RH, and the response of the virtual sensor increased by about 330 % in the same situation.
A similar approach can be used to interpret the results when CH 2 O was introduced as background, instead of RH, and the sensors were exposed to pulses of different concentrations of CH 2 O 2 and CH 3 COOH.As shown in Fig. 6a for different concentrations of CH 3 COOH, the response of the real sensor operating at a constant temperature decreased with increasing concentration of background CH 2 O. Conversely, the summarized response reported by the virtual sensor  when performing the same measurement was, again, considerably different.Instead of decreasing, the response increased with increasing concentration of CH 2 O.The same behaviour was observed for CH 2 O 2 , as shown in Fig. 6b for the virtual sensor.In this case, the percentile change in response, when the sensors were exposed to 1 ppm CH 2 O 2 diluted in SA with backgrounds of 0 or 1 ppm CH 2 O, is a decrease of about 70 % for the real sensor and an increase of about a 240 % for the virtual sensor.In case of CH 3 COOH with a background of CH 2 O, the change in response is a decrease of about 44 % for the real sensor and an increase of about 26 % for the virtual sensor.

Conclusions
According to the sensitivity enhancement observed when TCO is implemented in measurements with mixtures of two VOCs or RH and a VOC, the use of solid-state chemical gas sensors operating at a constant temperature is recommended for applications where few gases are present in the surroundings, with controlled atmospheres, such as automotive engineering or fugue detection in industrial processes.In applications where RH and other gas species such as VOCs can be present in the atmosphere simultaneously, such as dwellings, offices, schools, hospitals, churches, workshops, or museums, it is recommended to use the TCO mode, to avoid surface saturation.In these situations, where mixtures of gas species are present, high surface coverage complicates new gas-solid interactions, and the use of TCO is essential to overcome this issue.The use of temperature cycles allows to clean the surface of the sensing material periodically, thus preserving the sensitivity of these gas sensors independently of the presence of different gas compounds, such as RH or CH 2 O.

Fig. 1 .
Fig. 1.Gas test chamber with 3D-printed disposable top to allocate the commercial SGPC3 and SHTC1 sensors.

Fig. 2 .
Fig. 2. Current evolution of a virtual and a real sensor operating at 300 • C exposed towards pulses of different concentrations of formaldehyde.

Fig. 3 .
Fig. 3. Response summary as function of CH 2 O, CH 2 O 2 , and CH 3 COOH concentration for the real and the virtual sensor operating at 300 • C.

Fig. 4 .
Fig. 4. Response summary for four concentrations of CH 2 O as a function of different relative humidity percentages studied for (a) a real sensor and (b) a virtual sensor operating at 300 • C.

Fig. 5 .
Fig. 5. Response summary for four concentrations of CH 3 COOH as a function of different relative humidity percentages studied for (a) a real sensor and (b) a virtual sensor operating at 300 • C. Note a considerably increased sensor response in the case of virtual sensor.

Fig. 6 .
Fig. 6.Response summary as a function of different CH 2 O concentrations for (a) four concentrations of CH 3 COOH measured with a real sensor and (b) four concentrations of CH 2 O 2 measured with a virtual sensor operating at 300 • C.