Humidity is one of the main interferents often discussed in literature because it significantly affects sensors’ responses, altering their sensitivity to the target gases [53,54]. The moisture content of the air analyzed by e-noses is generally expressed in terms of relative humidity or, less frequently, absolute humidity. Relative humidity (RH) is the ratio between the vapor pressure of air and its saturation vapor pressure, usually expressed in terms of a percentage value indicating the amount of moisture stored in air under those conditions (i.e., temperature and pressure) with respect to the maximum capacity of air to contain water under the same conditions. On the other hand, absolute humidity represents the amount of water present in a volume of dry air, generally expressed in g/m³.

Several studies have dealt with e-nose applications in many different fields, such as healthcare and diagnostics, environmental odor monitoring, or industrial process control, discussing the effect of the moisture levels recorded by the humidity sensor installed in the e-nose chamber on the sensors’ responses.

As a general rule, the presence of humidity results in a shift of the whole baseline of the e-nose sensors, reducing the amplitude of sensors’ responses and, consequently, decreasing the sensors’ sensitivity.

The effect of humidity on MOS sensors is particularly studied since those sensor types are the most used in e-nose technology. Yan et al. reported ethanol analyses at different absolute humidity values (namely, 0.05, 0.25, and 0.35 g/m³) employing MOS sensors including commercial ones, such as TGS-2602 and MP135, and home-made ones based on WO₃ and SnO₂, resulting in a decrease of the sensors’ responses with the increase of water vapor content regardless of concentration [25]. A decline of gas sensors’ responses was also observed by High et al. when looking at the ratio between the resistance in air and the resistance of pure gases analyzed under laboratory conditions by means of In₂O₃ sensors [55] and by Yoon et al. with respect to the corresponding air-dry analysis [56]. By plotting sensors’ responses, temperature, and relative humidity onto a 3D surface plot, Abdullah et al. observed the lowest sensor responses when temperature and humidity were the highest, anticipating the effect of temperature on sensors that will be discussed in the next paragraph [57]. Similar trends can be observed in other types of sensors, such as SAW, optical, electrochemical, and QCM sensors, always showing a competitive mechanism of interaction between volatile chemical compounds and water molecules intervening in the working mechanism on which the outputs of sensors are based [53,58,59,60,61,62,63,64,65,66,67].

Clearly, the application of e-noses for environmental odor monitoring in the field is particularly critical in this regard because of the well-known variability of humidity in ambient air. Indeed, the variation of humidity levels could interfere with the working mechanisms of low-cost sensors usually employed in air quality measurements, such as MOS, electrochemical, and nondispersive infrared sensors [68,69]. As an example, Romain et al. observed a decrease in sensors’ responses with the increase of the relative humidity level during field experiments [70,71]. The influence of such parameters, such as seasonal meteorological effects typical of long-term applications, clearly emerged in an in-field e-nose application developed by De Vito et al. for the estimation of benzene by means of a device equipped with seven MOS sensors installed in an urban area [72]. With the purpose of developing a sensor array dedicated to environmental odor monitoring, the behavior of several types of MOS sensors according to different humidity rates was observed by Helli et al., underlining a decrease in sensitivity of some sensors (some of them showed responses independent from target gases’ water content), towards two environmental pollutants, i.e., H₂S, NO₂, analyzed at different humidity levels with dry atmosphere as reference air [73]. Sohn et al. demonstrated the humidity cross-sensitivity of conducting polymer sensors observing a shift in the plot of sensors’ response PCA on data regarding the monitoring of odor abatement performance of a biofilter [74].

It should be further highlighted that, besides the humidity content of the sample gas to be analyzed, the humidity level of the reference air should be considered. Suppose the reference air and the sample have different humidity contents. In that case, it will result in a variation in the sensor’s resistance when the input gas is changed from the reference air to the sample. This variation is not correlated with a different concentration of volatile chemical compounds. Thus, a proper setting of comparable water content between the reference air and the target gases is required. The two contributions to the sensor’s response would need to be separated [67,73,75].

When speaking about the effect of temperature on e-nose sensors’ responses, two different types of temperatures should be considered: the sensor’s operating temperature and the gas sample’s temperature.

The sensor’s operating temperature, i.e., the temperature at which the active, sensitive layer is heated, is a crucial parameter, especially for MOS or graphene-based sensors, as the surface reactions highly depend on the sensing element temperature [40,76]. CP sensors work at ambient temperature, but it has been demonstrated that an increase in working temperature can increase sensitivity and decrease sensor poisoning up to approximately 70 °C, when the polymers start degrading [40,76].

The effect of the sensor’s temperature is widely studied in the scientific literature [77,78,79,80,81]; it affects sensor surface reactions, which can lead to different responses. In the literature, three different approaches for this effect are presented. On one hand, the temperature modulation approach results in operating sensors at different temperatures in the same array. This can lead to an increase of information that can be used for better classification [77,78]. Secondly, the operating temperature of the sensors can be optimized to increase mutual information [79], and finally, it has been proved that stabilization of the operating temperature with closed-loop control systems can increase sensors’ performance [80]. Even if the proposed approaches can now enhance the quality of the signal, some limitations to these solutions still need to be closely investigated, starting from the measurement of the actual operating temperature on the sensor surface in the most-used gas sensors or the effect of temperature variations on the sensor’s surface reaction. For this reason, this work will focus only on the effects of the gas temperature on the sensors’ responses [40,76].

Indeed, some studies have evaluated the impact of the gas temperature on MOS sensor responses: Romain et al. [70] observed that when e-noses based on MOS sensors are used in the field for environmental odor monitoring, a decrease in the ambient air temperature results in an increase in the sensor resistance, which in turn can alter the MOS sensor responses. A similar behavior was observed by Kashwan et al. [22], who applied an e-nose for the analysis of tea flavor, and Abidin et al. [81], who also reported a decrease in the sensors’ response to different levels of toluene in the range between 25 °C and 40 °C. Huerta et al. [81] came to the same conclusion in their work when they monitored the air quality in a toilet, as did Peterson et al. [82] in the measurement of nitrogen dioxide and ozone in urban environments. One work by Knobloch et al. [83] presents the dependence of target gas temperature on conductive polymer response. The authors conclude that the change in the response could be related to the fact that changes in temperature lead to a greater or lower concentration of volatiles in the headspace, which will generate different sensor responses, but further studies need to be carried out to evaluate the exact effects, as this temperature effect can mask sensors’ responses that can affect classification.

One of the main factors affecting gas sensors’ response is the sampled gas flow, as proven by several studies focusing their attention on the impact of this factor.

When speaking about the variations of gas flow over e-nose sensors, it should be considered that, in some applications, sampling and analysis are performed at different times. In contrast, in others, sampling and analysis occur at the same time. Considering the first kind of applications, the storage of the target gas via different techniques (e.g., Tedlar^® bags [84], Nalophan^TM bags [85], adsorption materials [86], etc.) is required, and a later analysis is performed. In these cases, the central aspect that should be taken into account is the chamber geometry and the stability of the gas flow rate during the analysis phase. On the other hand, where there is no storage of the samples, the sensors’ response can be influenced by the flow regime of the target gas. Considering, for example, the biomedical field, e-noses are widely applied to exhaled breath analysis [87,88], which is intermittently generated and characterized by specific flow waveforms [89] that, if not adequately controlled, can result in artifacts in the sensors’ responses. Furthermore, in environmental monitoring applications where e-noses are typically installed outdoors in the open air, wind can cause alterations to the flow regime inside the sensor chamber. Gas flow variations can also occur in home or car air quality monitoring applications and home appliances, where internal fans typically affect the gas flow conditions.

However, to be more specific, the effect of gas flow should be considered by distinguishing three different flow-related aspects.

The standard approach for e-noses requires that the target gas flows continuously in the sensor chamber, and the flow rate is the first flow-related aspect affecting sensor responses discussed in this work.

In a study by Madhavi et al. [28], the authors demonstrated the relationship between flow rate and sensors’ response. This study first presents a mathematical simulation of an MOS sensor’s (TGS2620, Figaro Engineering, Osaka, Japan) responses to changing flow rate and position of the sensor in the chamber. Then, measurements were performed in a setup identical to the simulated one. Five different flow rates (0.3, 0.6, 1.2, 1.8, and 2.4 L/min) and three different test gases (methanol, ethanol, and propanol) were considered. The authors conclude that response time decreases with flow increase while the amplitude of sensors’ response increases with increasing flow, thus demonstrating that increasing the gas flow rate improves the sensors’ responses in both terms of response rate and intensity. The authors also simulated the temperature on the sensor plate, which obviously decreases with the flow. As mentioned in the previous paragraph, up to now, it has not been possible to directly measure the working temperature in this kind of sensor, as they work in an open-loop control configuration, so the considerations of this effect are based only on the simulations, and no direct measurement is reported.

Despite no direct measurements of the relationship between gas flow rate and MOS sensor operating temperature being reported in the literature, this aspect is crucial because the sensor sensitivity is highly dependent on the operating temperature [40]. Similar considerations are presented by Sedlak et al. [90] regarding polymeric homemade sensors. This work tests the sensor with a concentration of 3ppm of NO₂ with flow equal to 0.1, 0.5, 0.8, and 1 L/min. Sensitivity, response time, recovery time, limit of detection (LOD), and repeatability are considered important parameters to be evaluated. Experimental results showed an improvement of all the parameters with increasing flow, except repeatability.

Conversely, some works [91,92] propose another approach, called stop-flow operation, where the chamber is filled up with the target gas, and then the inlet and outlet ports of the chamber are closed, and the flow is stopped. In all the works considered, the authors show that the sensors’ response highly depends on the flow rate, which can be regarded as equal to zero in stop-and-wait mode. When flow is stopped, the sensors’ response changes due to the change in the flow rate, generating a transient reading before reaching a new plateau. Since the transient is related mainly to the flow rate change, this part of the signal does not give any information on the sample volatile chemical compound’s content and therefore has to be discarded from the analysis. This leads to a loss of data that can be extracted from the sensors’ response curve since in other works, it has been proven that the transient phase is also an important source of information [12,93]. Despite the abovementioned limitations of the stop-flow operation mode, this mode can be combined with the normal operation mode to increase measurement information [91]. Indeed, considering the different sensor responses when varying the flow rates during the analysis of one target gas, the quantity of information provided by each sensor is increased, thereby introducing the concept of “flow modulation” to the sensors’ response. This operation mode may, in principle, improve e-nose classification performance, but further studies are required to analyze the effects in more detail.

The direction of the target gas flow concerning the sensors’ active surface layer also affects sensor response. Since the beginning of the century, this aspect has been widely investigated via simulations and experiments. Lezzi et al. [94] proposed a simulation model considering different orientations and sensor distances with respect to the chamber inlet. The authors concluded that having the sensor near the chamber input and perpendicular to the flow can reduce the time needed to have a constant concentration on the sensors’ surface, reducing rising time and increasing the sensors’ response. Shyla et al. [95] confirmed this result, simulating how orientation affects gas speed on the sensor’s surface. They concluded that the lower the gas speed on the sensor’s surface, obtained with the perpendicular configuration between the flow and the sensors, the larger the time window in which a single molecule in the sampled gas can react with the active layer, resulting in higher and more complete response rate. Moreover, in a work by Sedlack et al. [90], which was already cited in the previous paragraph, besides the flow rate, the authors studied the effect of the orientation of the gas sensor on its response. In this experiment, the flow was set equal to 1 L/min and the concentration equal to 3 ppm of NO₂. Four different orientations were considered (0°, 45°, 90°, and 270°), with 0° identified as parallel to the flow direction. Sensitivity, response time, recovery time, limit of detection (LOD) repeatability, and signal-to-noise ratio were evaluated. All the parameters, excluding repeatability, improved by increasing the angle from 0° to 90°. Also, when the angle was set equal to 270°, repeatability had an opposite trend compared to the other parameters, i.e., improving when all the others worsened. Another work by Ryu et al. [96] came to the same conclusion by testing an MOS sensor, changing the gas impact angle and the distance between the gas inlet and the sensor active layer. Three different angles were tested (0°, 45° and 90°), considering 0° the condition when the flow direction was parallel to the active layer. NO₂ was considered the target gas, and its concentration was 5 ppm. Sensitivity was considered the target parameter for evaluating best performance. Results show that the best configuration to enhance sensitivity was at 90°. Finally, there is another work by Scott et al. [97] on quartz crystal microbalance gas sensors (QCM), presenting results that seem to contrast those previously mentioned [97]. In this study, the performances of an array of three gas sensors are compared. Three different experimental setups were considered: (a) sensors parallel to the flow direction, (b) sensors perpendicular to the flow direction, and (c) sensors parallel to the flow direction, placing a baffle between the gas inlet and the sensors. Response time (i.e., the time to reach a stationary response) was considered the target parameter, given that configuration (c) showed a faster response than configuration (b), which in turn showed better results than configuration (a). These results can be interpreted in different ways. First, QCM sensors rely on a transduction mechanism based on mechanical oscillations of the active part, which can be more sensitive to flow impact as it can directly modify the QCM vibration. Furthermore, in this work, a new element is introduced with respect to all the works mentioned until now, which is the consideration of how the flow regime condition in the chamber affects the sensors’ response. This is indeed the third flow-related aspect affecting sensors’ behavior investigated in the next paragraph.

In 2004, Scott et al. [97] introduced the observation that e-nose chamber fluid dynamics affect sensors’ responses. In this study, the presence of a baffle increases the performance of the sensors in terms of response time, resulting in a faster response compared to the perpendicular configuration, which on the other hand was proven to perform best in other works [90,94,95,96].

All the authors [27,94,98,99,100,101,102,103,104,105,106,107] focusing on the sensors’ chamber design agree that, inside the chamber, stable and uniform conditions should be reached as soon as possible. To ensure stable and uniform conditions, laminar flow is required in the chamber, with no stagnant or recirculating zones, which may result in a different exposure of the sensors to the target gas. Moreover, to ensure a reproducible test, all the sensors should come into contact with the gas simultaneously. Finally, it is important to remember that in ideal conditions, the concentration in the e-nose chamber can be described as [108]: where C(t) is the function of the concentration, C_in is the input concentration, f_in is the input flow rate, and V is the volume. According to Equation (1), the time to reach the steady state increases with the chamber volume and decreases with the flow rate. A possible approach to ensure a fast response would be to reduce the chamber volume to a minimum, which typically depends on the number and size of sensors used.