Formaldehyde and Total VOC (TVOC) Commercial Low-Cost Monitoring Devices: From an Evaluation in Controlled Conditions to a Use Case Application in a Real Building

Abstract

Formaldehyde and volatile organic compounds (VOCs) are major indoor pollutants with multiple origins. Standard methods exist to measure them that require analytical expertise and provide, at best, an average value of their concentrations. There is a need to monitor them continuously during periods of several days, weeks, or even months. Recently, portable devices have become available. Two categories of portable devices are considered in this research paper: connected objects for the general public (price <500 €) and monitoring portable devices for professional users (price in the range >500 to 5000 €). The ISO method (ISO 16000-29) describes the standard for VOC detector qualification. It is quite complex and is not well adapted for a first qualitative evaluation of these low-cost devices. In this paper, we present an experimental methodology used to evaluate commercial devices that monitor formaldehyde and/or total volatile organic compounds (TVOC) under controlled conditions (23 °C, 50–65% relative humidity (RH)). We conclude that none of the connected objects dedicated to the general public can provide reliable data in the conditions tested, not even for a qualitative evaluation. For formaldehyde monitoring, we obtained some promising results with a portable device dedicated to professional users. In this paper, we illustrate, with a real test case in an office building, how this device was used for a comparative analysis.

1. Introduction

Indoor air quality (IAQ) has been a topic of concern for the scientific community for many years, but these days it is also becoming of interest to the general public, who are more aware of the impact of IAQ on comfort and health. As we spend more time indoors than outdoors (80–90% of our time) in new and retrofitted airtight buildings, at home, school, or the office, there is a need to better assess human exposure in these various environments through the determination of major indoor pollutant concentrations, with the main objective being to anticipate any IAQ issues (default of ventilation, punctual pollution source, etc.).

Indoor pollutants are of various natures (chemical, physical, and biological). In this article, we focus on gaseous chemical pollutants, which already represent a large family of components (aldehydes, ketones, aliphatic and aromatic and hydrogenated hydrocarbons, terpenes, etc.). They are classified as very volatile organic compounds (VVOCs), volatile organic compounds (VOCs), semi-volatile organic compounds (SVOCs), and total volatile organic compounds (TVOC). We refer to the definitions given by the World Health Organization [1] and the ISO 16000-6 [2] in this paper. Indoor VOCs are of multiple origins, originating mainly from building materials and furniture but also human activities (cleaning, cooking, personal care, smoking, etc.), and outdoor pollution infiltration, etc.

International standards (e.g., the ISO 16000 series [2,3,4]) describe reference methods used to sample and identify/quantify pollutants at the ppb level (or µg/m³ level). These methods require analytical expertise and expensive equipment (>50 k€). However, they only represent an average concentration for a short period of time (the time of sampling) which varies from less than one hour for active sampling to several days for passive sampling. Those resulting reference concentrations can then be compared to standard and/or guideline indoor levels (see an example of guideline values for indoor air in

Table 1. Selected guideline values for formaldehyde (HCOH) [5,6,7] and total volatile organic compounds (TVOC) according to the literature [7,8,9]; for formaldehyde, 1.23 µg/m³ = 1 ppb.

	Formaldehyde (HCOH)	TVOC (Expressed as Toluene Equivalent)
Residential	100 µg/m3 (short term, 30 min)	Five levels from <300 µg/m3 to >10–25 mg/m3

).

Continuous monitoring with portable devices integrating miniaturized sensors that provide direct reading of the concentration, easy collection of data, and/or comprehensive feedback for a non-expert person would be very helpful in this context. They would not substitute reference methods; they would be complementary.

As our company develops and sells building materials, such devices would be helpful to assess the long-term in situ performance of our materials. As we do not manufacture sensors, we rely on existing commercial solutions.

Portable and connected devices are now commercially available. They either target the general public or professionals (contractors, craftsmen, marketing, sales force, etc.). Their price is generally in the range <500 € to 5000 €. For devices targeting professional users, who are not IAQ experts, a small amount of training is needed. In any case, the reliability of the devices is to be checked prior to their use. Are they sensible enough and selective enough for indoor air monitoring? What is/are the use case(s) for which they are the most appropriate?

The ISO 16000-29 standard [10] describes how to evaluate VOC detectors. This methodology is not applicable for formaldehyde detectors and is too demanding for connected objects. An international standard is in preparation by the ASTM committee but it will take a couple of years before a first draft is submitted. Meanwhile, many studies are using monitoring devices for scientific purposes. Most of the time no information is given on why such a sensor is chosen. There is not much information on the claimed performance and how it was evaluated. Bartosz Szulczynski et al. [11] presented a comparative overview of the current available sensing technologies for VOC monitoring, covering a large range of prices, including analytical equipment. A list of commercial portable devices with their technical characteristics is provided. However, no detail is given on how the manufacturers evaluate these and/or if any independent evaluation was realized. A few research papers have focused on sensor evaluation. In a review paper by Laurent Spinelle et al. [12], the focus was on benzene detection. They reviewed evaluation studies and drew conclusions regarding the lack of sensitivity and/or selectivity of the tested VOC sensors. In another review paper by Deepak Kukkar et al. [13] the focus was on nanomaterial-based sensors for formaldehyde detection. They reached the same conclusion. For the most selective ones, the sensitivity often remains at the ppm level. This is one order of magnitude above the indoor building level. Selectivity and sensitivity are key criteria which are still difficult to meet for miniaturized sensors. Many research projects sponsored at the European level have been aimed at developing selective sensors for specific VOCs, such as the Volatile Organic Compound Indoor Discrimination Sensor (VOC-IDS) project [14] and the SENSIndoor project [15].

The thorough evaluation of sensors requires expertise. F. Thevenet et al. [16] have described the real-scale evaluation of commercial miniaturized VOC sensors mounted on a multi-sensor device, under controlled conditions. They monitored the device response to transient pollution events consisting of one or multiple VOC compounds. They compared it with a selected ion flow tube mass spectrometer (SIFTS-MS) analyzer (price >100 k€) which provides accurate data, similarly to the standard method. The selected sensors were found to respond properly to pollution peaks but their response to a mixture was more complex than just the additive response to individual components. The impact of temperature and humidity on sensor response has also been studied. This is actually crucial for certain applications, such as if the sensor is to be used in a hot and humid climate. Three similar devices in the same room were compared to check repeatability. This paper illustrates how complex it is to properly assess sensor performances and that a common methodology would be helpful. Research projects funded by governmental bodies or private consortia have recently started evaluation programs on a large number of devices, giving full access to their methods and results [17,18]. This is already valuable information for researchers.

In this paper, we are interested in commercial TVOC and formaldehyde portable devices for two types of uses:

• Usage 1: Formaldehyde and TVOC sensors to follow IAQ trends (=IAQ indicator)
• Usage 2: Formaldehyde and TVOC sensors to monitor IAQ with accurate values of pollutants (comparable to standard methods)

Commercial devices were purchased. It was necessary to set up our own experiment to evaluate them as we could not obtain any documented evaluation, either from the supplier or from the scientific literature. This work is described in this paper. Based on our results, we conclude that none of the connected objects dedicated to the mass market were able to meet any of our objectives. We obtained promising results with devices dedicated to professional users, in particular, for formaldehyde monitoring. More experimental work is needed to conclude on TVOC sensors. This works highlights the necessity, in the near future, of improving the selectivity and sensitivity of miniaturized sensors targeting indoor air pollutants for commercial applications. Developing a standard methodology to evaluate them will be of great help to achieving this. It will certainly clarify things in the ever-growing IoT world of IAQ sensors.

2. Materials and Methods

2.1. Materials: Sensors and Sensing Technologies

The selection of the sensors was the result of an internal benchmark, including interviews with the manufacturers and/or an extensive search through their websites and technical brochures. Around 100 different devices were reviewed. The most promising devices were kept for the experimental evaluation (

Table 2. Summary of selected sensor technologies and their corresponding price rane. Legend: MOS, metal oxide semi-conductor; PID, photo ionization detection.

	Technology and Price Range
	Formaldehyde		TVOC
	Colorimetry (C)	Electrochemical (EC)	MOS	PID
	3	3	8	2
Connected objects for the general public (<500 €)		2	4
Monitoring devices for professional users (>500–5000 €)	3	1	4	2

). The technology of the selected sensor is briefly described in this table. More details can be found in review papers [11,13,19].

• Formaldehyde detection: electrochemistry (EC) or optical (colorimetric) (C) technology

  ⚬

  EC detection is one of the oldest known technologies used for chemical detection. It is low-cost, low-power, and compact. The principle relies on an oxydo-reduction reaction that takes place at the surface of the sensing electrode (the element is composed of two electrodes and one electrolyte). This generates a current or a voltage difference that can be measured which is proportional to the gas concentration. Selective membranes are most often added to improve selectivity. The sensitivity is generally at the ppm level and not the ppb level. The three EC sensors integrated in the tested devices are from the same manufacturer.

  ⚬

  Colorimetric detection of formaldehyde is also well known and documented [20]. It is based on a chemical reaction between formaldehyde and a probe molecule immobilized on a substrate (solid state reaction) or in solution (liquid state reaction). The newly formed molecule is detected by an optical measurement, either in transmission by measuring the optical density variation or by emission (fluorescence) of the new molecule when submitted to UV. This method is highly selective compared to EC as the probe molecule selectively reacts with formaldehyde. It is generally more sensitive and more expensive. For the three devices tested, the probe molecules were impregnated in a solid support and the reaction happened in the solid state. The probe molecules were different for the three detectors.

Sampling on DNPH cartridges was used as the reference method for formaldehyde sensor evaluation.

• TVOC detection [21]: metal oxide semi-conductor (MOS) or photo ionization detection (PID) technology

  ⚬

  MOS: The sensing element is a semi-conductor (a common one is SnO₂). In clean air, oxygen is adsorbed at the surface of tin dioxide grains, forming an insulating barrier. When the sensor is exposed to combustible gas or reducing gas, an oxidation reaction with the adsorbed oxygen occurs at the surface of tin dioxide. The potential barrier is reduced. The sensor resistance decreases. The affinity of the pollutant depends on the working temperature of the sensor. Usually, these sensors are combined with a heating system. Doping the semi-conductor is often used to improve selectivity. Seven out of the eight selected sensors tested integrate MOS from two different suppliers. The four connected objects integrate a MOS sensor from Supplier 1 and the three devices for professionals integrate a MOS sensor from Supplier 2. For the remaining sensor, no information on the sensitive element could be obtained.

  ⚬

  PID: Photo ionization detectors use high-energy photons, typically in the UV range, to excite the molecules, resulting in the ionization of the gas. The energy is quite low so only gases with low ionization energy, i.e., organic vapor, can be ionized. The resulting ions produce a current that is proportional to the gas concentration. Intrinsically, PIDs are not selective, as they ionize everything with ionization energy which is less or equal to the lamp output. There are usually more sensitive to humidity than MOS sensors and require recalibration more often. They are also more expensive. The two PID sensors tested came from different suppliers.

Sampling on Tenax cartridges was used as the reference method for the TVOC sensor evaluation.

• Multi VOCs: No commercial portable device able to detect individual indoor VOCs was identified.

2.2. Methods

2.2.1. Sensor Evaluation Under Controlled Conditions

The evaluation was performed in one of the two following facilities:

• Laboratory-Scale Evaluation Chamber

The chamber system used is located at CertainTeed Insulation R&D in Malvern, PA. It includes a full set of two chambers, gas cylinders, pumps, and mass flow controllers. This setting was used for the evaluation of formaldehyde sensors. The schematic principle of the setting is shown in Figure 1. Details of the equipment are listed below (corresponding pictures are shown in Figure 2).

⚬

ThermoElectron Forma Environmental Cabinet (Model 3920 or equivalent) capable of maintaining 23 °C ± 1 °C.

⚬

Sixty-seven-liter stainless steel climatic chambers with a temperature and humidity sensor and plumbing to supply high-purity air and sampling outlets to allow for simultaneous formaldehyde sampling. As shown in Figure 2, there are three ports for the gas inlet; these were connected to dry air, wet air, and the formaldehyde cylinder. The gas inlet goes to the bottom of the chamber and there is no fan or splitter inside the chamber to agitate the gas flow. There are four sampling ports and one exhaust port for the gas outlet on the outside cover lid.

⚬

Ultra-zero-grade air cylinders (<0.1 ppm total hydrocarbons, <1 ppm CO₂, and <1 ppm CO) or air from a zero-air generator that was tested for background aldehydes and VOCs.

⚬

Two formaldehyde cylinders (500 ppb in N₂ and 1 ppm in N₂) were purchased from Air Liquide America Specialty Gases, LLC.

⚬

Mass flow controllers, a sampling pump, and proper electronic control (e.g., Build Time software) to maintain stainless steel chambers at 50% relative humidity (RH) with a total inlet flow of 1.12 L/min. RH deviation was <±3% for the 50% RH set up.

Desired formaldehyde concentrations in the chamber were achieved by adjusting the flow from the formaldehyde cylinder and the dry air. The chambers were allowed to purge for 4 h prior to the sampling of formaldehyde.

For these tests, concentrations were varied between 0 (clean air) to 100 ppb in formaldehyde to test the sensor response at 23 °C, 50% RH. The actual temperature and RH profiles inside the chamber were recorded by temperature and RH probes.

During this experiment, active sampling with DNPH cartridges and HPLC-UV analysis was used as the reference method for formaldehyde measurement. DNPH cartridges from two major suppliers, Sigma-Aldrich Supelco LpDNPH S10L (St. Louis, MI, USA) and Waters Sep-Pak XPoSure Plus short cartridges (WAT047205) (Milford, MA, USA), were used. When sampling, air at a rate of 0.5 L/min was collected through the DNPH cartridge for a period of 2 h. The DNPH analysis was run in our own laboratory in the US. Uncertainty in the measurements was 20%, including in all the steps of measurements. Those tests were conducted in 2015.

• Full-Scale Evaluation Chamber

The chamber system used is located in the Insulation Research and Development Center of Saint-Gobain Isover, in Rantigny, France (CRIR). This chamber was used to evaluate formaldehyde and TVOC sensors. A schematic of the chamber settings and a picture are shown in Figure 3 and Figure 4, respectively.

The air from outside went into a preparation room in which the air could be purified or polluted. A hole was drilled close to the door for sampling. This treated air was then equally distributed (controlled by flowmeters) in two identical test rooms using one hole for each. Each room followed ISO 16000 instructions and represented a reference room in Europe.

–

The walls were 2.5 m high.

–

The floor and ceiling both measured 3 m × 4 m, resulting in surfaces of 12 m² each.

–

There was one door of 0.8 m (width) × 2 m (height) (1.6 m²).

–

There was one window of 2 m².

–

Sealants and other very small surfaces accounted for up to 0.2 m².

The total wall area (minus door and window) was 31.4 m². The total air volume was 30 m³.

Two holes were drilled in each room for air extraction and sampling (Figure 3). The temperature and humidity were monitored and controlled by sensors. During the trials, the temperature was between 20 °C and 22 °C. For the RH, two conditions were tested: 50–55% and 65%.

For the purpose of the evaluation, the usual setting of the full-scale chamber (preparation room and two test rooms) was slightly modified. Indeed, the three rooms of the full scale were used to represent three different levels of pollution so that the sensor evaluation could be performed under different conditions at the same time, with each one being representative of the indoor air pollution level in a real building. The three different environments were called low-, medium- and high-pollution. To create those different levels of pollution, wood panels were used as the major pollution source. In the preparation room, wood panels were stored in order to generate a moderate concentration. This medium polluted air was then sent to the two test rooms. In one of these, extra wood was deposited to achieve higher pollutant concentrations (high-level). In the other, depolluting plaster boards acting both on formaldehyde and VOCs were installed to reduce the pollution and reach lower pollutant concentrations (low-level). To obtain a homogeneous mix of pollutants in each room, additional mechanical ventilators were introduced.

All the sensors were installed on a trolley which was moved from one room to the other two during the trials. Such a procedure implied that we regularly opened the door of the room, disrupting the homogeneity and pollution levels inside the room. However, all the sensors were connected to the general power supply without being turned off when being moved from one room to another. This way, we believe that the sensor stabilization period was shorter. Enough time (generally more than an hour) was left in order for the initial concentrations to be recovered and all the sensors stabilized.

The concentrations of pollutants were regularly measured with the standard methods in each room to check the stability and homogeneity of the pollution levels during the trials. Active Tenax^® sampling was used for TVOC and active DNPH sampling was used for aldehydes according to the respective ISO 16000 standards. VOC and aldehyde analyses were performed in our internal chemical laboratories in France (SGR Paris and CRIR).

The tests were conducted from 19 to 22 April 2016; three days of measurements were performed at 50% RH and a fourth day at 65% RH.

2.2.2. Test Case in a Real Building

A test case was performed in our office building, i.e., Saint Gobain Research India, located in Chennai. The final objective was to demonstrate, using in situ and real conditions, the performance of a depolluting product that was being developed by one of our R&D teams. The principle was to select two meeting rooms in the building. One of the rooms would remain unchanged and would be considered as a reference room, while the second one would be retrofitted with the depolluting product. The impact of the retrofitting could be monitored and compared to the reference room. This required a preliminary investigation step consisting of monitoring the air quality in the different meeting rooms across a few weeks (weekdays and weekends). Only formaldehyde was monitored for this test case. Three pairs of meeting rooms were considered. They were paired according to collected information on their location as well as the ventilation system (room pictures have been provided in Supplementary Figure S1).

Pair 1:

• Takshasila (ground floor): near the entrance, can be accessed by outsiders and employees, and two ACs are available
• Sydney opera (first floor): can only be accessed by employees and two ACs are available

Pair 2:

• Lothal (ground floor): can be accessed by outsiders and employees, one AC is available, and construction work was occurring outside the room
• Empire state (first floor): can only be accessed by employees, one AC is available, and construction work was occurring outside the room

Pair 3:

• Chandrasekhar (second floor): can only be accessed by employees, and is connected to the central HVAC
• Visvesvaraya (second floor): can only be accessed by employees and is connected to the central HVAC

The investigation was started on January 2017. Continuous measurements were combined with some passive sampling on DNPH cartridges. DNPH analyses were performed by an external laboratory. The rooms were monitored during three different periods: weekdays (with occupancy and with HVAC on), weeknights (without occupancy and with HVAC off) and during weekend days and nights (without occupancy and with HVAC off).

3. Results

3.1. Evaluation of Formaldehyde Sensors

3.1.1. Evaluation at Lab Scale

Two devices based on colorimetric detection were tested in this facility. They are named Sensor C1 and Sensor C2 in the following section. They incorporated different sensing probe molecules. The concentration was measured by the two sensors inside the chamber at 23 °C and 50% relative humidity and was compared with the reference measurement given by DNPH active sampling. The results are presented in Figure 5, Figure 6 and Figure 7. For Sensor C1, the experimental points represent the cumulated values of two different devices of the same supplier. For Sensor C2, only one device was used for the testing.

As shown in Figure 5, Sensor C1 was able to pick up formaldehyde signals down to 10 ppb. The supplier actually claims a sensitivity down to 20 ppb. There is a linear relationship which can be observed between the sensor value and the reference formaldehyde concentration. The regression line in the range of 10–100 ppb has a slope of 0.57 with an R² of 0.89. The slope is much lower than unity, indicating that the sensor is underestimating the real formaldehyde concentration. These results suggest that Sensor C1 could be a good formaldehyde sensor down to 20 ppb if the readings could be corrected by a factor. The correction factor was calculated to be 1.75 at 23 °C and 50% RH from the reciprocal of the regression slope (Figure 6). One explanation of the correction factor should be a bias in the calibration step of the cartridges at the supplier level. In this chamber study, we tested two devices and 10 sensing parts from different production batches (14 April 2015 and 28 July 2015). As shown in Figure 5, all data fit well to the same regression line, indicating the good repeatability of Sensor C1.

As for Sensor C2 (Figure 7), the experimental results show that it was not sensitive for an HCHO concentration lower than 30 ppb, which differed from its claimed sensitivity of 10 ppb. When the HCHO concentration reached 20–30 ppb, Sensor C2 started to pick up signals; however, 43% (three out of seven measurements) were false negatives. At a concentration of 30–100 ppb, Sensor C2 reported concentrations of HCHO close to that obtained with the active DNPH method. The regression line in the range of 30–100 ppb can be seen to have a slope of 0.83, which is lower than unity, indicating that the sensor underestimated the real formaldehyde concentration. These results suggest that Sensor C2 could be a good indicator at 23 °C and 50% RH when the readings are higher than 30 ppb.

3.1.2. Evaluation in Full-Scale Chamber

Five devices based on two different technologies were evaluated:

• Colorimetric: two commercial references, namely, Sensor C1, previously evaluated at lab scale, and Sensor C3, were tested. Sensor C3 used a different probe molecule from Sensors C1 and C2.
• Electrochemical: three commercial references, namely, Sensors EC1, EC2, and EC3, were tested. Out of the three, two presented dysfunctions during the trials and had to be excluded from the data analysis (battery problems, no sensitivity for concentration variations, poor working time <15 min before standby mode, etc.). This was the case for the connected objects that fall into this category.

The three remaining devices (C1, C3, and EC1) belonged to the category of devices for professionals (price >500 to 5000 €).

The full experiment covered concentrations of formaldehyde between 5 to 140 µg/m³ based on the reference method. This concentration range is representative of real concentrations of a very clean and highly polluted indoor environment, respectively.

The results obtained at 50% RH and 65% RH are plotted in Figure 8 and Figure 9. The dotted line represents the perfect correlation between the two methods (the slope is equal to 1). The gray cone represents the uncertainty of 15%. For Sensor C1, measurements from previous tests in full scale performed in 2015 (called past data) are also included.

To compare reference measurements performed at specific times with continuous monitoring by the device, we plotted the average concentration of the sensor responses during the full duration of the cartridge acquisition (i.e., for one hour) versus the reference concentration. Thus, each cartridge analysis corresponds to a single point per device. The error bars shown in the plot correspond to the standard deviation of the mean values during the measurement interval.

• Concerning trends:

At 50% RH (Figure 8), Sensor C1 and Sensor EC1 globally show a good trend (which is better for Sensor C1), namely, when formaldehyde concentration increased, the sensor detected an increase. This was not the case with Sensor C3. We noticed that the signal was unstable for the three highest concentrations.

At 65% RH (Figure 9), an impact on the linear behavior can be observed: for concentrations below 40 µg/m³ the sensor response is linear, but at higher concentrations the sensors underestimated value (in particular Sensor C1). For Sensor C3, it was difficult to make a conclusion, as we did not have enough experimental values.

• Concerning accuracy:

At 50% RH, the linearity of Sensor C1 with the reference method can be observed to be acceptable, except for very low concentrations (<20 ppb, consistent with the manufacturer’s claims). This conclusion is also in agreement with the results obtained in the laboratory-scale evaluation in the previous section. On the contrary, Sensor EC1 shows a right trend, and the accuracy of the measurements was low, with a high error present, especially at low concentrations. At best, Sensor EC1 can be used to obtain a trend in a comparative analysis where formaldehyde pollution is suspected. Sensor C1 should be preferred when more accuracy is needed. Another disadvantage is that electrochemical technology suffers from interferences, in particular when alcohol is present (not investigated here). When selectivity is critical, this is not the best solution. For Sensor C3, the results were quite good at very low concentrations; however, we observed an unstable signal at higher concentrations. The supplier claims that a 30 min sampling time is enough to obtain stabilized results. We waited for this duration in our experimental protocol but still faced issues. Additional testing is required to conclude on Sensor C3. This work is ongoing.

3.1.3. Application in a Test Case in Meeting Rooms in a Real Building

The formaldehyde concentration in the meeting rooms was recorded and compared over a few days. Sensor C1 was chosen based on the results detailed in the previous paragraphs. Sensor C1 is a formaldehyde sensor based on colorimetric technology. This device also records temperature and relative humidity at the same time. This test case is also an opportunity to check if the device is easy to use for professionals who are using them for the first time and who are not IAQ experts.

Preliminary recordings were carried out in January 2017 during between two and five consecutive days. Temperature, relative humidity, and formaldehyde concentration are plotted as a function of time in Figure 10, Figure 11 and Figure 12 for the six meeting rooms.

As shown in Figure 10, Visvesvaraya and Chandrasekhar were observed to have very similar variations recorded throughout the day and night, whereas the other two pairs had quite different relative humidity and formaldehyde concentrations (Figure 11 and Figure 12).

In Visvesvaraya and Chandrasekhar, the temperature was below 25 °C and the relative humidity was below 65%, which is lower than in the four other rooms. It is known that a higher temperature and/or relative humidity results in higher formaldehyde emission from sources; as a consequence, we expected a lower formaldehyde concentration in Visvesvaraya and Chandrasekhar, which was actually the case during the recorded period.

The formaldehyde concentration slightly increased overnight. This was correlated with an increase in the relative humidity and was related to the shut-down of the HVAC system at the end of the working day (it is turned off after 20:00). When the system was switched on again the next morning at 8:00, the relative humidity, as well as the formaldehyde concentration, decreased and reached the same values as the day before. We noticed, however, a formaldehyde spike at 35 ppb which then rapidly decreased. This is likely an artifact linked to the switching on of the HVAC system. The sensor took a measurement every 30 min. At 7:40, the recorded value was 28 ppb. At 8:10, 10 min after HVAC was switched on, it was 37 ppb. Thirty minutes later, it was 31 ppb.

For the two other pairs of rooms the temperature was very similar, but relative humidity showed around a 10% difference between the rooms. The difference in relative humidity had an impact on the formaldehyde concentration. For the Lothal and Empire meeting rooms (Figure 11), the room with the higher humidity had the higher formaldehyde concentration. For the other pair (Figure 12) we also observed a difference, but one which was the opposite, meaning that the higher humidity had the lowest formaldehyde concentration. A possible explanation for this is that the Takshasila meeting room close to the building entrance on the ground floor may be subjected to a higher air renewal due to repeated door opening.

Based on these preliminary results, the recommendation was to continue with the Visvesvaraya and Chandrasekhar meeting rooms. Longer recordings were carried out. Those recordings took place from the end of January 2017 to the beginning of March 2017. Reference measurements on DNPH cartridges were done in parallel between 16 and 27 February. The cartridges were analyzed by an external laboratory. The corresponding results are displayed in Figure 13.

Those longer periods of recording confirm that the two rooms display very similar behavior in terms of temperature, relative humidity, and formaldehyde concentration. Once again, the impact of the switching on or off of the HVAC system during weekdays and weekends can be systematically observed.

⚬

Temperature was maintained between 20 and 24.5 °C in both rooms with a minimum temperature difference of 0.5–1 °C always observed between both rooms, which may have been due to instruments or because of the HVAC system.

⚬

Relative humidity was in the range 55%–75% and sometimes reached 80%–85%, the RH difference between both the rooms was minimal, and the relative humidity increased overnight when HVAC was switched off.

⚬

Once the HVAC was switched on in the morning, we detected formaldehyde spikes.

Figure 14 presents a ‘zoom-in’ of February recordings from 16 to 27 February when DNPH samplings were performed. During this period, the rooms were kept unoccupied to make a comparison without the impact of occupancy. DNPH sampling periods are indicated in Figure 14 (black circle and arrow) and the corresponding results are listed in

Table 3. Comparison of formaldehyde concentrations as indicated by Sensor C1 and reference DNPH passive sampling in Chandrasekhar and Visvesvaraya meeting rooms.

Date	Time	Condition	HCHO (Chandrasekhar)		CorrectIon Factor	HCHO (Visvesvaraya)		CorrectIon Factor
			DNPH (ppb)	Sensor (ppb)		DNPH (ppb)	Sensor (ppb)
16 February 2017	10:00 a.m.	HVAC on, with occupancy	22.7	25.6	0.89	32	21.4	1,50
16 February 2017	6:20 p.m.	Daytime, during week
16 February 2017	8:15 p.m.	HVAC off, no occupancy	31.8	25.6	1.24	32	23	1.39
17 February 2017	8:30 a.m.	Overnight, during week
17 February 2017	8:15 p.m.	HVAC off, no occupancy	59.3	35.6	1.67	44.5	30	1.48
20 February 2017	9:15 a.m.	Over weekend
23 February 2017	10:15 a.m.	HVAC on, room locked	45	40.4	1.11	50.5	38	1.33
23 February 2017	6:40 p.m.	Daytime, during week
23 February 2017	8:00 p.m.	HVAC off, room locked	29	26.2	1.11	32.8	20	1.64
24 February 2017	8:20 a.m.	Overnight, during week
24 February 2017	6:30 p.m.	HVAC off, room locked	40.6	33.6	1.21	42	33.7	1.25
27 February 2017	9:00 a.m.	Over weekend

. Temperature is not shown in this graph for ease of visualization, but as can be seen in Figure 13, the temperature was very similar between the rooms during this period.

The sensor always underestimated the formaldehyde concentration, except for one data point, on 16 February, which was actually the lowest concentration measured by DNPH and was very close to the detection limit of the sensor indicated by the manufacturer. As explained in the previous paragraph, relative to the laboratory-scale testing, we determined a correction factor to correlate sensor values and DNPH values. This correction factor was 1.75 at 23 °C and 50% RH. This correction was different here. As indicated in Table 3, the calculated correction factor (concentration DNHP/concentration sensor) was lower than 1.75. Indeed, the factor was even different between the two rooms. Our hypothesis is that this correction factor could be humidity-dependent. As shown in Figure 14, the RH level was slightly different between the two rooms (showing an around 5% difference) and the humidity level was always higher than the 50% RH used in the laboratory-scale experiment.

Whatever this correction factor, Sensor C1 is appropriate as our objective was to make a comparative analysis between the rooms. Sensor C1 gives us the right trend. When DNPH indicated a higher level, this was also the case for the sensor. The next step consisted of retrofitting one of the two meeting rooms with the active product and following the evolution of the concentration between the rooms after retrofitting, using the same monitoring device.

3.1.4. Overall Conclusion Regarding the Evaluation of Formaldehyde Sensors

For formaldehyde detection, a total of six portable devices, based on two different technologies (colorimetric and electrochemical), were purchased and tested under controlled conditions. Two out of the six target the mass market. Our experimental results suggest that none of the connected objects can be used whatever the use case, not even as rough air quality indicators. The results for devices targeting professionals were more promising:

• Colorimetric technology is preferred, especially when sensitivity is needed. With Sensor C1, we could detect concentrations as low as 20 ppb. We even obtained sensor responses for concentrations down to 10 ppb and below (data not shown) but we prefer to make no recommendations for this range as the supplier does not guarantee accuracy for <20 ppb. However, to obtain absolute values comparable to the reference method, a correction factor has to be applied. Based on the different tested conditions, this correction factor could depend on temperature and humidity. This device was used in a practical real building to compare meeting rooms, confirming the interest of the device for such a use case. This in situ test also showed that Sensor C1 is quite easy to use for non-experts, based on the feedback collected.
• Electrochemical technology may still retain some interest for its ability to obtain a very basic trend, but not at concentrations <30 ppb, which are often met in buildings today. Although they were not presented in this study, electrochemical sensors have interference issues, in particular in the presence of alcohol. When sensitivity and selectivity are key criteria, colorimetric technology is preferred.

3.2. Evaluation of TVOC Sensors

From our internal benchmark, a total of 10 TVOC devices were selected: out of the 10, three (integrating a MOS sensor) had to be removed from the study as they presented dysfunctions during the trials. Information regarding the seven remaining devices is given in

Table 4. Summary of technology/price range of tested TVOC monitoring devices.

	Technology and Price Range
	MOS	PID
Connected objects for the general public (<500 €)	MOS from supplier 1 MOS1_S1, MOS2_S1, MOS3_S1
Monitoring devices for professional users (>500–5000 €)	MOS from supplier 2 MOS1_S2, MOS2_S2	Unknown PID supplier PID1, PID2

. We only had one device of each type, except for PID1, for which we had two types of equipment that were co-located in the same room.

These sensors were only evaluated in the full-scale chamber as our laboratory is not equipped with a gas mixture representative of indoor pollutants that could be used as the pollution source in a small-scale evaluation. For full-scale testing, VOCs were emitted from the wood panels installed in different compartments to mimic indoor pollution. The wood panels were purchased just before the trials to guarantee high emission levels. Identification and quantification of individual components emitted by the wood panels were obtained from reference measurements. The main VOCs detected were α-pinene, hexanal, and β-pinene (see Supplementary Table S1 for the full list and concentrations). In terms of TVOC concentrations, the experiment covered a range from 40 to 170 μg/m³ toluene equivalent (when taking all individual components eluted and with a concentration value >LOD) which is representative of quite a low level of pollution.

• Sensor scaling factor:

Contrarily to formaldehyde sensors, which directly give the results in µg/m³, the TVOC concentration displayed by the devices is given in various units, namely, ppm equivalent CO₂, ppb equivalent formaldehyde, ppb equivalent isobutylene, and µg/m³ equivalent toluene. The reference method gives the TVOC concentration in µg/m³ equivalent toluene. To compare the data, all the results were converted to µg/m³ equivalent toluene. This was done using the ratio of the gases’ molar masses.

For example, 1 ppb _{eq.isobutylene} = (M_isobutylene/M_toluene) * ppb _{eq toluene} = 0.5 ppb _eq.toluene.

The scaling factor S, defined as S = S_sensor/S_tenax, is plotted in Figure 15, where S_sensor is the TVOC concentration given by the sensor in µg/m³ equivalent toluene and S_tenax is the corresponding TVOC concentration measured via the reference method. A scaling of 1 means that the sensor has been well calibrated and that the result can be seen as an absolute value. The bigger the difference, the lower the accuracy of the sensor.

We obtained significant differences in the scaling factor depending on the device.

➢

For MOS: Devices with MOSX_S2 sensors are better calibrated than MOSX_S1 devices, as their scaling factor is closer to 1. For MOSX_S1 devices, the TVOC concentration is given in equivalent CO₂. Indeed, these devices do not have a CO₂ sensor inside. It is not valid to make a direct correlation between CO₂ and TCOV concentration in a room. The first is essentially due to human breathing and the second has multiple origins, including building materials, perfumes, and human activities. Therefore, such an analogy is confusing.

➢

For PID: The two tested devices have a scaling factor close to one, which would suggest they are well calibrated.

From this first comparison, we would not recommend using the MOSx_S1 series, which displays TVOC values which are more than one order of magnitude higher than the real concentration. At best, they could be used for a comparative analysis if their response is linear.

• Sensor linearity

Because of the different units, linearity was studied by plotting the sensor response S_sensor expressed as the % of the full scale of values covered by the sensor. S_sensor was plotted as a function of the reference TVOC concentration in µg/m³ equivalent toluene.

$${\mathrm{S}}_{\mathrm{sensor}}\left(\mathrm{x}\right)\%=\left[\frac{{\mathrm{S}}_{\mathrm{sensor}}\left(\mathrm{x}\right)-{\mathrm{S}}_{\min }\mathrm{Sensor}}{{\mathrm{S}}_{\max }\mathrm{Sensor}-{\mathrm{S}}_{\min }\mathrm{Sensor}}\right]\ast 100$$

where x is the TVOC concentration measured by the reference method, S_min is the sensor response for the minimum TVOC concentration measured by the reference method, and S_max is that for the maximum TVOC concentration. At 50% RH, S_sensor was assessed for four different TVOC concentrations: 39 µg/m³, 60 µg/m³ (2 series), 103 µg/m³, and 116 µg/m³ for the PID and MOSx_S2 sensors (Figure 16 and Figure 17) and 41 µg/m³, 51 µg/m³, 108 µg/m³, and 121 µg/m³ for the MOSx_S1 sensors (Figure 18). For measurements collected at TVOC concentration of 60 µg/m³, the result on the graph is the average of the two measurements.

➢

PID sensors (Figure 16): In the concentration range tested, the sensor did not show a linear response. In addition, the two measurements performed at the same TVOC concentration of 60 µg/m³ gave very different results, suggesting a high error: for PID1-1, the average value of the two measurements was 17%, but the error bar was 100%; for one of these measurements the sensor did not detect any concentration and for the second measurement the value was 34% of the full scale. A similar observation was performed with PID1-2 co-located with PID1-1: the average result of the two measurements was 34% but again the error bar was 100%. A possible explanation could be that the devices were tested for a very low range of TVOC concentrations; these tests may have been performed for below their detection limit, and additional trials should be carried out at higher TVOC concentrations (200–1000 µg/m³).

➢

MOS sensors: At 50% RH, the MOSx_S2 devices showed linear behavior in the concentration range tested (Figure 17). However, the error bar at 60 µg/m³ was high (100%); for one of the two measurements the device hardly picked up a signal (1–2%). A possible explanation could be that the tests were performed close to their detection limit, as for the PID sensors. Additional testing should be carried out at higher TVOC concentrations. As for the MOSx_S1 devices (Figure 18), the behavior was not linear at all. Together with the scaling issue detailed before, we discarded these connected objects; they could not even be used for a qualitative analysis.

3.3. Overall Conclusion from Evaluation of TVOC Sensors

In this work, a total of 10 devices integrating two different sensor technologies (MOS and PID) were benchmarked. Three out of the 10 had to be removed from the testing experiment as they presented dysfunctions.

Our experimental results in controlled conditions suggest that none of the connected objects targeting the mass market can be used, not even as qualitative IAQ indicators. As for the devices targeting professional users, our results, given directly below, are not fully conclusive at present.

• PID technology: The tested devices gave TVOC concentrations close to the reference method, based on the scaling results. However, in the range of tested TVOC concentrations, the results were not linear. We think that we probably tested them in a range close to their detection limit. Further tests should be done at higher TVOC concentrations (200–1000 µg/m³ equivalent toluene). The impact of humidity should be further investigated. We know that PID sensors are more sensitive to humidity, which is a problem for hot and humid climates. During these trials, we did not collect enough data at 65% RH to present conclusive results in this paper.
• MOS technology: The results were very dependent on the sensor manufacturer. One series had to be discarded completely (MOSx_S2). For MOSx_S1, the response was linear, but should be confirmed at higher TVOC concentrations. The TVOC concentrations were overestimated, as their scaling factor was higher than 1. If their linearity was to be confirmed for a larger TVOC concentration range, they could be used for comparative analysis.

The diversity of results obtained as a function of technology and sensor manufacturer confirm the necessity of being very cautious before using any of these devices for experimental purposes.

4. Discussion

Today, a lot of connected objects targeting the general public and monitoring devices for professionals are on the market. In most of cases, the consumer buys a “black box”. It is important to understand how such a device is built. We can consider that devices are structured using three main parts (Figure 19) which are described directly below.

• A sensitive element (i.e., the sensor) which is defined by a specific technology and which is the most important part is the first part. The signal delivered by this component is usually an electron or a photon. It is not always easy to obtain information on the sensor part from the manufacturer. A good sensitive element will be selective, sensitive enough, and robust. Robustness can be understood in different ways from mechanical robustness to the absence of maintenance or drift in the measured value in the long term.
• Integration of the sensitive element in an end-user product, which determines how the raw data are acquired and how to minimize noise, is the second part. This includes the hardware interface as well as possible drivers requested for the sensitive element to perform under the best conditions. It is also important to know if a calibration has been performed, how this calibration was carried out, and how often recalibration needs to happen.
• Data display, which is the way the raw values of the sensitive element are converted to pollution concentration values, accounting for possible cross-sensitivity via a compensation system, is the third part. The way data can be extracted is also important (e.g., API, or as a .csv file). Usually, for connected objects targeting the general public, the end user is not able to extract the raw data and obtain a comprehensive result through a mobile application.

When the device is not working as expected, it is not necessarily the sensing element that is responsible for this; any of the three steps can be critical.

Before choosing a device, it is also key to know exactly the targeted usage; you do not need the same level of accuracy if you want to follow trends, if you want to make a comparative analysis, or if you suspect an IAQ issue and need to detect precisely the components present and their concentrations. We found very few scientific publications detailing the evaluation of such commercial devices, although some studies have been performed that use them for research purposes. The few studies on sensor evaluations often highlight their lack of selectivity and sensitivity [12,20]. Our study presents the first evaluation of a large number of devices tested at the same time under controlled conditions.

Two types of monitoring devices for formaldehyde and TVOC detection were considered, for two main uses:

• Connected objects targeting the general public, from which we might expect to obtain a good trend, without expecting too high an accuracy.
• Monitoring devices for professional users, from which we expect better accuracy, even if we do not expect to obtain the exact concentration as the reference method would give.

For formaldehyde, we tested two mature technologies, namely, electrochemical and colorimetric, for TVOC, MOS, and PID. We know that other technologies based on nanomaterials exist but they are still in the research stage [21,22].

Our evaluation of these integrated sensors indicates the following:

➢

All connected objects targeting the general public must be excluded. Most of their manufacturers buy sensitive elements off-the-shelf; they are not experts on sensing technologies. Their added value is more in terms of their design and their mobile application interface; they may neglect some key steps in their integration. Indeed, we know that miniaturized sensors for formaldehyde detection are still difficult to find. There are new proprietary technologies under development, but in the meantime, the user should be very cautious when buying such objects.

➢

For devices targeting professionals, we achieved promising results for formaldehyde detection using colorimetric technology. It is known that electrochemical technology is not as selective and has some issues regarding the detection limit at the ppb level. Still, for colorimetric devices, the acquisition time is recommended to be at least 30 min to obtain sufficient sensitivity. There are still some issues with very low concentrations (<20 ppb) which can be met in buildings. In addition, the sensing part of the device is consumable and needs to be replaced after a few days or a few weeks, has to be kept in a refrigerated environment, and has an expiration date (usually one year starting from the manufacturing date if kept refrigerated). If we consider these devices to be used by professionals, it may be difficult for them to manage these practical aspects. For the most promising device identified in our benchmark, our results showed that the formaldehyde concentration was always underestimated. A correction factor was determined to obtain the absolute concentration. This correction factor seems to depend on temperature and humidity. As for TVOC detection, we could not find a solution for our use cases, and we may have to test them at a higher range of TVOC concentrations. Our results suggest that their evaluation is even more complex as a large variety of units can be used by the manufacturer.

Our evaluation is only partial: testing a device must include all the relevant factors. Creating a common protocol through a standard could be very beneficial. It should at least encourage scientists to question suppliers when they are buying a monitoring device and in particular when they are low-cost, even if they integrate technologies that are known to be mature.