1. Introduction
The capability to identify and discriminate between odorous substances is largely limited to methods and instruments based on the sense of smell. Progress in science and technology has increased interest in devices designed and operating analogously to human senses. The last 30 years have witnessed the elaboration of analytical systems capable of replacing odour evaluation by humans, at least to certain extent [
1,
2,
3,
4,
5,
6,
7,
8,
9]. The instruments have that attracted attention due to the utilization of a wide range of chemical sensors are electronic noses. These devices, being analogues of the human sense of smell, can be applied in many fields of science and industry such as medical diagnostics [
10,
11,
12,
13], environmental protection [
14,
15,
16,
17,
18,
19,
20,
21], food and agricultural industries [
22,
23,
24,
25], and forensic science and criminology [
26,
27,
28,
29]. Electronic nose instruments allow holistic analysis of gas mixture compositions without the need to separate and identify of particular components [
30,
31], moreover, they have gained increasing popularity as far as measurement of the gas mixtures with very low concentration levels of their particular components is concerned [
32,
33]. These devices are characterized by additional advantages as compared to other techniques for odour analysis such as olfactometry or gas chromatography. As opposed to olfactometric techniques there is no need for olfactory adaptation and trained personnel for defined odour perception. Electronic noses offer rapid analysis and lower costs than chromatographic techniques. With all their advantages and certain limitations electronic nose instruments are complementary with respect to the aforementioned odour analysis measurement techniques.
The sensation of odour is relatively difficult to describe quantitatively. There are four basic features of odour determined during investigations and description of odorous compounds: odour concentration, intensity, hedonic tone and olfactory threshold. In the case of gas mixtures containing odorous compounds there is a discrepancy between sensed odour and summary odour (being a sum of the odours of particular components) [
34,
35,
36].
Knowledge about olfactory thresholds of pure chemical compounds does not allow anticipation of the odour of their mixtures with other compounds. The sensible range of the mixtures is not an additive quantity. This is a result of so-called odour interactions consisting in mutual masking, amplification or attenuation of odours. The investigations on odour interaction types have been carried out for many years, however no satisfactory explanation for the mechanism of these processes has been provided so far. Most frequently the objects under investigation are air samples containing no more than two or three odorant types [
37,
38,
39]. Dependence between physical stimuli acting on senses and psychical feelings is described by a field of knowledge called psychophysics. In the case of interactions models of odour interaction are created, which describe the dependence between odour intensity of air containing pollution mixtures and:
odour intensity, which would be caused by the mixture components if they were present separately (perception models),
concentrations of mixture components and their psychophysical characteristics (psychophysical models).
None of numerous elaborated models possesses a general character. The perception models of interaction combine odour intensity of a mixture (
IAB) with intensity of its components when present separately (
IA,
IB). The best-known empirical equation combining odour intensity of a mixture with odour intensity of the individual components is Equation (1), the Zwaardemaker equation (from 1908) often called vector summation of intensities:
The interaction coefficient (a) present in the equation is approximately constant for one pair of mixture components. Recently the investigations on determination of this coefficient for the compounds from aldehyde, ester or aromatic hydrocarbons groups were conducted by Yan et al. who determined its values at the level of: −0.22 for aldehydes, −0.25 for esters, −0.129 for aromatic hydrocarbons [
40,
41]. Generally, the odour interaction coefficients falls between −0.326 and −0.423, although there were also cases when it took lower values from −0.156 to −0.208. A couple of other perception models have been elaborated, including the Berglund, Patte and Laffort or U models [
42,
43,
44,
45].
Odour intensity depends on the number of odorous substance molecules, which are in contact with olfactory receptors, namely on odorous substance concentration in the inhaled air. Intensity is defined as “strength of odour perception”, which is triggered by particular olfactory stimulus. Most frequently verbal point scales or reference scales are used in order to determine odour intensity. An example of the verbal scale can be a 6-grade scale recommended in the German guidelines VDI 3940, where 0 value means no odour and 6 value describes an extremely strong odour.
As far as odour hedonic tone is concerned, hedonic interaction of odour is an interaction of an odorous substance, which, upon evaluation of a given olfactory sensation, is attributed to a certain feature located between two extreme situations described as extremely pleasant and extremely unpleasant, respectively. In practice odour hedonic tone is evaluated in the way similar to odour intensity using single-dimensional scales (verbal, graphical, point ones). Negative values of the verbal scale describe unpleasant olfactory sensations, positive values of that scale correspond to pleasant sensations, a value of 0 is associated with a neutral olfactory sensation.
Determination of odour intensity or odour hedonic tone using an electronic nose requires application of a “teaching under supervision” approach. Depending on the research problem these techniques (teaching under supervision) are employed to construct calibration, discrimination or classification models. Construction of the above models utilizes a set of explanatory variables (signals from the sensors comprising an electronic nose) and a set of dependent variables (values of odour intensity or hedonic tone expressed on the verbal scale). A calibration method task is construction of a model, which allows quantitative evaluation of particular property or the properties based on the set of explanatory variables. The most popular calibration techniques include multiple linear regression (MLR), principal component regression (PCR) and partial least squares (PLS). Operation on a big number of correlated variables results in limited applicability of the MLR models. Hence, the main methods used for construction of linear calibration models are PCR and PLS, as they are capable of managing correlated variables. These methods have found successful application for monitoring of odour concentration changes in processes such as biofiltration or sewage treatment as well as a support to dynamic olfactory measurements [
46,
47,
48,
49,
50,
51,
52,
53,
54]. Literature in the field [
55] reveals that for binary mixtures (
n-butanol, acetone) of odorous compounds, a linear regression between odour intensity and averaged sensor response is appropriate to represent the relationship between odour intensity and electronic nose measurement when conducting polymer sensors are used, but it is inadequate for metal oxide sensors. Moreover, for binary mixtures of odorous compounds (
n-butanol, acetone), a neural network can be trained to accurately predict odour intensity from commercial electronic nose sensor responses.
This paper describes an attempt to apply the PCR method and an electronic nose instrument to determine odour interactions of three-component gas mixtures characterized by different types of odour, which are components of the odorous mixtures typically present in municipal landfills or sewage treatment plants. The PCR method was used to determine if there occurred amplification of odour intensity with respect to the theoretical value estimated with Patte and Laffort model. The same method was utilized to check whether the electronic nose could be used to predict odour hedonic tone and how this value differed from the theoretical value calculated as algebraic sum of particular odorous components. Proposed PCR models were verified via coefficient of determination (R2) and root mean square error of prediction (RMSEP).
2. Materials and Methods
2.1. Types of Three-Component Mixtures and Their Preparation
The investigation employed pure substances and three-component mixtures of the following compounds: toluene, acetone, triethylamine (set A and B) and formaldehyde, butyric acid, α-pinene (set C and D).
Table 1 presents characteristics of the investigated odorous compounds including odour type, vapour pressure, and olfactory threshold in the gas phase [
56,
57,
58].
Five aqueous solutions characterized by a 2-step dilution were prepared for each of the investigated substances. For acetone and formaldehyde the concentrations were as follows: 200, 400, 800, 1600, 3200 ppm v/v in deionised water. For toluene, triethylamine, butyric acid the concentrations formed a series: 5, 10, 20, 40, 80 ppm v/v in deionised water. In case of pinene the concentrations were as follows: 1, 2, 4, 8, 16 ppm v/v in deionised water. Prepared solutions were used for determination of odour intensity and hedonic tone of each sample, which was performed by a team of assessors. Obtained results were utilized to plot odour intensity versus logarithm of concentration of particular odorous substance in water as well as to depict hedonic tone versus logarithm of concentration of particular odorous substance in water. These plots allowed estimation of the olfactory thresholds of given substances in aqueous solution in order to confront them with the theoretical values. Additionally, the plots were used to determine concentration of particular substances in deionised water corresponding to odour intensities equal to 1 and 2 according to the scale proposed in the VDI 3940 guidelines.
2.2. Olfactory Triangles
A triangle presenting distribution of three-component mixture samples is shown in
Figure 1. The samples present on the vertexes of the triangle (1, 6 and 11) are comprised of pure substances in deionised water, characterized by an odour intensity equal to 1 or an odour intensity equal to 2. The points located on the sides of the triangle represent two-component mixtures of the respective compounds in deionised water. Three-component mixtures in deionised water are inside the triangle. The composition of prepared three-component mixtures (set A and B with intensity 1 and 2; set C and D with intensity 1 and 2) is shown in
Table 2. Moreover,
Table 2 contains the information about hedonic tone of particular samples calculated based on the plots described in the
Section 2.1.
2.3. Measurement of Odour Intensity and Hedonic Tone
The investigation was carried out by a group of assessors whose task was the olfactory evaluation of the prepared samples. This group consisted of five persons trained according to St. Croix Sensory 2006, a procedure elaborated by St. Croix Sensory, Inc. (Stillwater, MN, USA). The assessors were also aware of and followed the rules concerning olfactory investigations contained in the standard PN-EN 13 725 “Air quality. Determination of odour concentration by dynamic olfactometry”. The task of each assessor was determination of odour intensity and hedonic tone of prepared samples of aqueous solutions. Each sample was attributed the odour intensity within the range from 0 to 6 and the hedonic tone from −4 to 4. The assessors evaluated the total of 880 samples in case of the interactions within the olfactory triangle and 450 samples in case of the functional dependences: odour intensity versus logarithm of concentration of particular odorous substance in water as well as hedonic tone versus logarithm of concentration of particular odorous substance in water.
2.4. Description of Experimental Setup for Electronic Nose Investigations
A scheme of the experimental setup is illustrated in
Figure 2. It consisted of:
bottle with carrier gas (compressed air) with reducing valve,
system of air purification containing three filters filled successively with: active carbon (C), molecular sieve 5A and silica (SiO2),
three-way V1and cut-off V2 valves,
sample mounting system,
mass flow controller (red-y smart series GSC-B9SS-BB23, Voegtlin, Aesch, Switzerland)
prototype of electronic nose equipped with a matrix of seven sensors: six sensors of MOS-type (TGS 813,TGS 816, TGS 822, TGS 2444, TGS 2602, TGS 2620-FIGARO USA Inc., Arlington Heights, IL, USA) and one PID-type sensor (MiniPID-Ion Science Ltd., Cambridge, UK)
PC-class computer.
2.5. Methodology of Measurement Using Electronic Nose
Clean air was flowing through a measurement system. Inlet and outlet tubes were connected to the investigated sample in order to provide carrier gas flow. After mounting of a tube with the sample a three-way valve V1 was switched in order to change a direction of air flow. Air was supplied into the sample via the inlet tube while the outlet tube led aerated phase into a measurement sensors chamber of the electronic nose. Recording of signals started 25 s after the moment when air had been passed through the probes with investigated substances. The signal was recorded for 15 s. After that time the three-way valve was switched into its initial position enabling cleaning of the measurement system due to undisturbed flow of the carrier gas until e-nose signal returned to the initial level. The measurement parameters were determined via optimization method and they were as follows:
volumetric flow rate of air, determined using the rotameter, was equal 0.3 L/min,
time of carrier gas flow through the sample: 25 s,
signal recording: 15 s.
The system operated in a stop-flow mode meaning: 25 s of carrier gas flow, 15 s of carrier gas flow interruption, 5 min of carrier gas flow. A total of 880 samples were investigated in order to define odour interactions in the olfactory triangle, wheminutere 440 samples were the training ones and 440 samples constituted the tested ones. In case of utilization of the semiconductor sensors (in the investigations presented) a quotient technique of baseline correction was applied:
where:
Smeas–value of signal after baseline correction,
S(
t)–value of signal at given time instant prior to baseline correction,
S0–value of baseline signal.
This correction allowed reduction of the drift exhibiting multiplicative character. Application of this technique was found justified and purposeful due to its theoretical background, which suggests that the quotient method should provide the best effects [
59].
2.6. Data Analysis
Analysis of the data obtained with the electronic nose prototype was carried out using free R software being a part of Free Software Foundation (Free Software Foundation, Boston, MA, USA). All original variables were subjected to transformation via autoscaling, which resulted in variance of all properties equal each other and equal 1.