The electronic nose (E-nose) has been in continuous development since its conception in the early 1980s by Persaud and Dodd [1
]. The term E-nose describes an instrument consisting of an array of cross-sensitive gas sensors, coupled with a pattern recognition approach. This operating principle attempts to mimic the function of biological olfactory receptors by analysing samples as a whole (so-called “odour fingerprints” [2
]), rather than individual chemicals. In the past, E-noses were mainly used in the food safety or food quality, for example, for the determination of tea quality, adulteration of olive oil, fruit ripening and the rancidity of meat [3
]. More recently, two other major application areas for E-noses have emerged; specifically, environmental monitoring (e.g., detecting pollutants, hazardous chemicals and/or explosives) [4
] and biomedical purposes (e.g., monitoring and diagnosing diseases) [5
]. These applications rely on the ability of E-noses to generate a holistic analysis of a gas-phase mixture, which are made up of volatile organic compounds (VOCs) and some non-VOCs, such as inorganic gases. VOCs can be broadly defined as carbon-based compounds (C2
), which include a diverse group of compounds such as hydrocarbons, esters, alcohols, ketones and aldehydes, with high vapour pressures and low boiling points (50–260 °C) [6
In gas-phase chemical sensing, the gold standard is generally considered to be gas chromatography mass spectrometry (GC-MS) [7
]. While this method is highly reproducible and accurate, it is also very expensive, requires highly trained staff and lacks portability. The key advantage of E-noses is that their relatively simple operating principle and construction enables far more flexibility in the tailoring of the design and technical specification (e.g. ease of use, battery life, durability, number of sensors, sensing technology and sampling technique). There are several types of sensor technologies suitable for constructing E-nose arrays. An in-depth summary can be found in the review by Dospinescu et al. [8
]. The most commonly used sensors in E-noses are metal oxide (MOX) and conducting polymers (CP). Both types of sensors operate through a change in conductivity when exposed to a target gas [9
]. Examples of commercial E-noses, based on MOX and CP technology, include PEN3 [11
] (AIRSENSE Analytics, Schwerin, Germany), FOX4000 [12
] (AlphaMOS, Toulouse, France) and Cyranose 320 [13
] (Sensigent, Baldwin Park, CA, USA). While these E-noses have been successfully deployed in a number of studies [14
], a limitation is that they can suffer from drift and poor repeatability. MOX sensor responses can drift significantly, due to ambient temperature, pressure changes and material aging [15
], and CP sensors often demonstrate poor repeatability and reproducibility, due to the random nature of the polymer [16
]. The effects of sensor drift can be mitigated by utilising multi-sensor arrays (e.g., 6–18 sensors), to reduce the relative drift of individual sensors within the array, and by using pattern recognition techniques to account for material drift and environmental changes. Repeatability issues can be addressed by calibrating the E-nose to a number of standardised gas exposures. While these methods are relatively well-established, other sensing technologies are available that are fundamentally less susceptible to these effects, such as optical gas sensors.
Optical gas sensors measure the modulation of light properties or characteristics, such as changes in absorbance, polarisation, fluorescence or other optical properties [17
]. Since this method does not relate to chemical reactions, such as those for MOX and CP sensors, environmental changes have less effect on the response of optical sensors, with the exception of pressure. In general, optical gas sensors therefore require less frequent calibration. Moreover, with regard to sensor drift, single, or much smaller numbers of optical gas sensors, could perform similarly to larger MOX or CP sensor arrays and would also be less reliant on pattern recognition techniques to account for environmental changes. Further advantages of this sensing mechanism are that it is highly sensitive to a wide range of VOCs, often have good selectivity (especially to gases such as CO2
), good response/recovery times and longer sensor life than other gas sensor types [18
]. The disadvantages of optical sensors are that they are typically more expensive and can have lower portability, due to the delicate optics and electronics [4
]. Moreover, they are difficult to miniaturise and may require complex electrical circuits, since they do not rely on measuring a simple transduction mechanism, such as electrical resistance [19
]. Despite some of these limitations, it has been argued that optical sensors are fundamentally among those best suited for E-nose applications [5
Compared to the other gas sensor technologies, optical sensors have had less commercial success, likely due to cost and level of sensitivity, with the exception of targeted CO2
]. Review papers evaluating the application of E-noses in food safety focus entirely on MOX and CP sensor-based system and merely mention optical technology in a long list of potential sensing methods [21
]. Similar trends are observed in review articles that focus on the application of E-noses for environmental monitoring [4
] and medical diagnostics [24
]. These reviews indicate that E-noses are currently under-utilised in research and that there is still significant scope to evaluate the potential of this technology. In the commercial domain, there is a small selection of gas analysers, which utilise optical technology, such as the IRIS 4100 mid-IR and MIRAN SapphIRe (Thermo Fisher Scientific, Waltham, MA, USA). The former uses laser absorption spectroscopy to measure CO2
, while the latter uses a single beam infrared (IR) spectrophotometer for the monitoring of ambient air samples. The MIRAN SapphIRe unit is portable and can detect the presence of up to 121 gases, but is also expensive (upwards of $
30,000). Further examples of portable gas analysers include the GT5000 Terra, DX4015 and DX4000 (Gasmet Technologies, Vantaa, Finland). These operate based on Fourier Transform IR (FTIR) spectroscopy and can measure up to 50 gases simultaneously, but they cost upwards of $
40,000. These instruments exemplify the current state of commercial optical sensors, in that the devices are too costly or specialised.
The growing demand for Internet-of-things (IoT) enabled environmental monitors [26
], point-of-care diagnostic devices [24
] and real-time food quality sampling equipment [27
] calls for novel E-nose designs and concepts. These should enable E-noses to operate in environmentally challenging conditions, with continuous temperature and humidity fluctuations. While current MOX and CP-based gas sensors may struggle in these situations, optical sensing technology has the potential to thrive. The aim of this work is therefore to develop a novel optical-based E-nose, which is portable and can be deployed in a wide variety of typical E-noses applications.
2. Materials and Methods
The proposed optical E-nose is comprised of four emitters, paired with tuneable detectors. Each emitter–detector pair is individually encapsulated in a heated sensor chamber. The sample inlet is controlled using a valve and the system operates a negative-pressure system, using a pump at the exhaust. The system is controlled by the user using a laptop, via wired USB connection. A system diagram of the optical E-nose is shown in Figure 1
The following section discusses the individual sub-systems of the device in more detail.
2.1. Optical Gas Sensors
A nondispersive infrared (NDIR) sensor system consists of three main parts: emitter (IR source), gas flow path and IR detector (with filter) (Figure 2
). The operating principle is based on molecular absorption spectrometry. When a gas is inside the chamber, molecules absorb the radiation from the IR source. The filter on the optical sensor will only let the radiation wavelength related to the target gas pass. Molecular absorption of IR is measured to detect a fall in signal, which can produce a unique odour fingerprint [28
]. Several VOCs relevant for food safety and biomedical monitoring applications are known to have absorption frequencies in the range of 2.8–5.2 μm, such as acetone, 2-butanone and isopropanol [29
]. It is worth noting that the number of photons absorbed is directly proportional to the power of the photon beam from the emitter and thus the amount of the gas/vapour detected. It is therefore possible to determine the concentration of the measured molecule. There are two advantages associated with IR radiation. Firstly, IR radiation scatters less than visible radiation in the presence of steam, mist or smoke, due to its longer wavelength [31
]. Secondly, the miniaturisation of IR sources and detectors is possible using the latest MEMS technology.
The optical E-nose utilises commercially available emitters and detectors. The emitters are high emissivity thermal IR emitters (EMIRS200, Axetris, Kägiswil, Switzerland). The tuneable detectors were developed by InfraTec (Dresden, Germany) and are summarised in Table 1
These detectors have the ability to scan the mid- and long-wave IR range (3.1–10.5 μm) [32
], in steps of 20 nm. This should be considered the wavelength resolution of the optical E-nose. Due to this feature, any fixed frequency measurements can be considered as “virtual sensors”. Conventional IR detectors use expensive precision filters to select specific frequencies [33
]. Modern IR detectors use a filtering technique based on micromachined Fabry–Pérot interferometer (FPI), which involves an optical cavity with two parallel reflecting surfaces, which act as a half-wave resonator [34
]. By changing the plate separation, the central wavelength can be tuned [35
]. This can be achieved by simply adjusting the voltage across a control pin. Since gases/VOCs have a unique infrared absorption frequency [36
], this optical E-nose can be used to identify individual gases/VOCs in a complex sample, as well as analysing the odour as a whole without identification of each chemical [37
2.2. Sensor Chamber Design and Heating
The emitter–detector pairs are encapsulated in individual, heated chambers. Gas chambers for IR sensors can be designed in two different ways, intended to either maximise resolution or signal-to-noise ratio (SNR). Since most applications of E-noses require the detection of low concentrations of VOCs (ppm to ppb range), the high-resolution design was chosen. The length of the chamber is a critical feature of this design. According to Beer–Lambert law, increasing the optical path will increase sensitivity, but this decreases the SNR from emitter to detector. Different sensor chamber lengths were evaluated to determine the optimal length of the selected sensors. Lengths under consideration were 10, 20, 30, 40 and 50 cm. The results from these experiments demonstrate that the 30 cm chamber was associated with the highest differential voltage sensor response, using CO2
as the test gas. The term “differential voltage sensor response” refers to the difference between the baseline (ambient air) sensor response vs. signal when exposed to an analyte—as measured in Volts (or milli-Volts). The voltage drops in the presence of 1000 ppm CO2
(absorption frequency set to 4.2 μm) are summarised in Table 2
The sensor chamber length was manufactured to a length of 30 cm. The four chambers were arranged in a 2 × 2 formation, with mounting plates on either end, as shown in Figure 3
. An alternative approach could involve fitting all four emitters and detectors into a single wider chamber; however, this is likely to result in poor SNR.
The sensor chambers are made of stainless-steel tubes. The internal surface was polished, which results in more reflections. The material is also relatively easy to clean and has good thermal characteristics to implement an effective chamber heating system. A heating system is necessary in order to avoid condensation forming inside the chambers, thereby reducing environmental effects on the sensors. By having a fixed temperature above ambient, thermal effects of different sample temperatures is reduced. However, this also reduces the SNR.
Nichrome (NiCr) wire was used as a heating element, since it is low-cost and simple to implement and control. NiCr wire was coiled around each of the chambers and wrapped in high temperature resistant Kapton tape (436-2778, RS, Corby, UK) to provide some insulation and prevent short circuits. The NiCr wire’s heat dissipation is in direct relation to the amount of voltage and current applied to it. The temperature inside the chamber is automatically regulated using a 24 V supply and a PID controller algorithm (proportional, integral and differential). The temperature set-point is selected by the user (e.g., 35 °C) and feedback from inside the chamber is provided by a SHT75 temperature and humidity sensor (Sensirion, Stäfa, Switzerland). The maximum temperature should not exceed 65 °C so as not to damage the sensors. The heating procedure is initiated during start up and takes round 5 min, as shown in Figure 4
2.3. Electronic Design and Software
This system can be controlled by a tablet or PC, via wired USB connection. A custom app was created using Universal Windows Platform (UWP) to view the sensor responses in real-time and change system properties (e.g., chamber temperature). The UWP app communicates with a central microcontroller, a Teensy 3.6 (PJRC, Portland, OR, USA) development board. This is used to interface with the emitter and detector boards (using UART), as well as the rest of the control electronics. The electronic design requirements for the optical E-nose are divided into four segments: (1) sensor drive; (2) emitter drive; (3) heater system; and (4) control electronics.
The sensor drive was designed to run all four emitter/detector pairs. The sensors have different supply voltage requirements: 3.3, ±5, and 12 V. The electronic drive board was designed to work with 24 V and regulated to lower and negative voltages using linear voltage regulators. In addition, depending on the sensor, a voltage of 30–90 V was required for controlling the central wavelength; specifically, 30 V for LFP3144C-337, 45 V for LFP-3850C-337, 60 V for LFP-8850-337 and 90 V for LFP-80105C-337. The voltage regulators used to generate 12, 5, 3 and −5 V were LM2937IMP, L78L05ABD, LD1117S33TR and L79L05ABD, respectively. To generate 90 V from 12 V, a LT1372HVCS8 high frequency switching regulator was used in combination with a flyback transformer.
The emitter drive implemented a power-regulated circuit. This approach can compensate the inherent variation in source parameters and is more efficient for DC voltages with low frequency pulses. A square-pulse wave from the Teensy 3.6 was used to turn on and off the emitters, at a frequency of 10 Hz. The heater system regulates the temperature of the sensor chambers. A simple switching circuit turns the voltage on/off across the NiCr wire. A digital control signal from the Teensy 3.6 is used to turn on/off a MOSFET, which allows current to pass through the NiCr wire, when the temperature reading falls below a predetermined threshold. The control electronics provide 24 V to the four heated sensor chambers, sensor circuitry and 12 V to the pump and valve. Two input power adaptors were used to separate the heating input power from the control circuitry. This is because of the high current required for the heater system. An overview of the optical E-nose communication/control lines is shown in Figure 5
2.4. Mechanical Design
Mechanical elements of the optical E-nose include the internal flow path configuration and mounting of components inside an electronics enclosure. The flow path within the optical E-nose can be configured in two ways, as shown in Figure 6
. The sample flow can pass through the sensor chambers either in series or in parallel. Both versions have their advantages and disadvantages. In both variants, a valve controls the sample inlet and a pump pulls the sample through the device, to create a negative pressure system. In the series version, the pump flowrate is crucial for the sensor readings. If the sample passes through the system too quickly, the sensors may not be able to produce an accurate and reproducible reading. The advantage of this configuration is that the total amount of VOCs is present across all chambers during analysis. In the parallel configuration, the sample will be split. Since the sample is divided by the number of chambers (in this case four), dilution and marginal pressure differences may occur from chamber to chamber. The advantage of this version is that all sensors will have sufficient time to detect the VOCs in the sample. The current version of the optical eNose utilises the parallel flow path configuration.
The optical E-nose was designed and constructed to match the form factor of traditional E-noses. The completed system weighs around 12 kg and the dimensions of the white electronics enclosure are 48 cm × 23 cm × 47 cm. The circuit boards and sensor chamber were mounted inside the enclosure using 3 mm acrylic sheet support structures (434-295, RS, Corby, UK). Internal and external views of the unit are shown in Figure 7
. A screenshot of the graphical user interface (GUI) is shown in Figure 8
Unlike classical gas sensor technologies, which are based on transduction principles (e.g., electrical resistance, potential and current, frequency), optical sensors measure changes in light properties. The key advantage of this method is that it is able to overcome some of the limitations associated with typical MOX and CP sensor-based E-noses. To the best of our knowledge, the developed optical E-nose instrument is the first device that operates based on FPI sensors with tuneable filters. This technology is known to be less susceptible to influences from environmental parameters, such as humidity and temperature. However, further gas rig testing is required to characterise these effects on the sensors deployed in the E-nose. Other sensor parameters, such as sensitivity, detection limits, repeatability, response and recovery times, are also worth further investigating. Due to the operating principle of the optical E-nose, the response and recovery times are very rapid (less than a second). Therefore, the response time is that of the system and of the sample rate, not of the sensor. Since the addition/removal of the sample is controlled using the valve-pump negative-pressure system, the effects of altering the pump flow rate could be investigated.
Traditional E-noses use an array of sensors to analyse a chemically complex sample. In our system, not only do we have an array of four optical sensors, but these sensors can scan across a range of frequencies. In use, we can use specific frequencies (or a range of frequencies) to behave as “virtual” sensors to form arrays of different sizes. Furthermore, different combinations of these virtual sensors can be used for specific applications, making the instrument more flexible than a traditional E-nose.
The unit could also be used as a traditional gas analyser, but it would be necessary to determine the detection limits for a number of potentially relevant compounds, such as CO2
, ozone, carbon monoxide, oxygen, ammonia and sulphur dioxide [46
]. However, this is not necessarily needed for E-nose applications, since these focus on analysing complex mixtures. An alternative use is to use this E-nose to produce a so-called total VOC (TVOC) reading. The concept of TVOC attempts to report the combined effect of different VOCs, which may not be otherwise captured, due to low concentration levels of some contributing VOCs [47
]. In this case, the E-nose would be set to scan the widest possible range of wavelengths and use these measurements to calculate a single value reading for air quality. To the best of our knowledge, an optical sensor has not been used to establish TVOC readings.
In the case of traditional quality assurance and quality control (QA-QC) applications, E-noses are likely to be in a controlled or possibly even laboratory environment to assure product uniformity in manufacturing is maintained. However, E-noses are being increasingly required to operate in uncontrolled conditions, such as open-air for environmental monitoring [48
] and patient wards for point-of-care testing [49
]. Food quality control protocols are also moving towards “spot-checks” which are conducted during transit or upon delivery of good [50
]. For these applications, the sensor array must have low sensitivity to a wide variety of environmental parameters, but in particular to temperature and humidity [17
]. In these situations, the benefits of utilising optical gas sensing technology is likely to be greatest. However, these environments also bring rise to new challenges.
There are some other limitations and drawbacks associated with the developed unit. In its current implementation, the sample vials need to be connected to the device inlet manually for analysis. This is both time-consuming and vulnerable to human error. An automatic injection system could be added to address this drawback. There are commercially available autosamplers, suitable for sample headspace analysis, which include agitator functionalities and allow the method parameters to be configured. In addition to these modifications, there may be some opportunities to further miniaturise the sensor configuration by implementing a more compact gas chamber design [51
]. This could reduce the overall dimensions of the system and allow the internal components to be housed in a more portable enclosure.
Further scope for improvement relates to the transmission of sample data. The system currently connects to a laptop or PC via wired USB connection. In order for the device to be more practical in an outdoor or clinical setting, and keep up with current technological trends, the optical E-nose could be fitted with a Wi-Fi and/or Bluetooth compatible microcontroller to interface with a smartphone app or transmit data directly to a cloud-based computing platform for data analysis [52
]. A previous publication has demonstrated how E-nose devices can be designed or upgraded to be Internet-of-things (IoT) enabled [53
]. The appropriate data protection measures would also have to be implemented in order to ensure secure transmission and storage of data. Lastly, further work is also needed to ensure the environmental tolerance of the unit, in harsh conditions, and the addition of data processing capabilities to remove background signals.