RU2456590C1 - Method for test-identification of multi-component gaseous mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia - Google Patents

Abstract

FIELD: chemistry.

SUBSTANCE: method for test-identification of multi-component gaseous mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia involves forming an array of piezo sensors with different selectivity towards the analysed components. The method also involves preparing and collecting samples of equilibrium gaseous phases of sorbates and then putting them into a detection cell. The method also involves picking up and processing sorption analytical signals, calibrating sensors and constructing 'visual fingerprints' of standard and analysed mixtures. To identify benzene, toluene, phenol, formaldehyde, acetone and ammonia, an array of six different sensors is formed. To this end, electrodes of piezo crystal resonators from sorbent solutions are coated with films of Apiezon-N, a mixture of Triton X-100 with an extract of higher mycelial fungus Pleurotus Ostreatus, polyethylene glycol adipate, polyethylene glycol sebacate, polyvinyl pyrrolidone and dinonylphthalate with mass of 15-25 mcg. After injecting samples into the detection cell, the sensors are polled in the following order: the sensor with the Apiezon-N film is polled after 5 and 10 s, sensors based on the mixture of Triton X-100 with an extract of higher mycelial fungus Pleurotus Ostreatus after 20, 25, 30 s, polyethylene glycol adipate after 55 s, polyethylene glycol sebacate after 60 s, polyvinyl pyrrolidone after 70, 75 s and dinonylphthalate after 80, 85, 90 s. To predict 'visual fingerprints' of standard mixtures, a unidirectional three-layer neural network is used, which is trained based on sorption results on each sensor and physical-chemical properties of components of the mixture: molar mass, molar refraction coefficients, permittivity, density of sorbates, boiling point, pressure of saturated analyte vapour, presence and number of oxygen and nitrogen atoms and CH₃ groups or other substitutes in the molecule. During analysis of real gas systems containing benzene, toluene, phenol, formaldehyde, acetone and ammonia, the obtained 'visual fingerprints' are compared with those in the 'visual fingerprint' data base of standard mixtures. The degree of similarity and qualitative composition of the analysed mixture are determined from the shape of the fingerprints, and the area of the diagram is a quantitative criterion.

EFFECT: method for test-identification of multi-component gaseous mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia, which enables to identify multi-component mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia in different combinations, simple procedures for obtaining standard visual fingerprints of mixtures and short duration of analysis.

3 tbl, 3 dwg

Description

The invention relates to analytical chemistry, to methods for test identification of gaseous media using polysensor systems of the "electronic nose" type and can be used to analyze air containing heterogeneous combinations of mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia.

The closest in technical essence and the achieved result is a method for multivariate calibration of a multisensor sorption sensor for the analysis of vapors of ethyl acetate, acetone and toluene in air. Regression analysis models [Ya. I. Korenman, T. A. Kuchmenko, I. V. Aristov, D. A. Kudinov // Sorption and chromatographic processes. - 2004. - T.4. Electronic resource. Special issue. - S.326-351], allowing the analysis of a three-component mixture.

The disadvantage of the prototype is the inability to identify gas systems with the number of components more than 3, as well as the prediction of "visual prints" of standard mixtures when changing its composition or concentrations of individual components.

An object of the invention is to develop a method for the test identification of multicomponent gas mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia, which allows to identify multicomponent mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia in different combinations by forming a matrix of sensors with cross-selectivity to defined compounds and optimization of the conditions of its work; simplify the procedure for obtaining standard "visual prints" of mixtures; reduce the duration of the analysis.

The technical task of the invention is achieved by the fact that in the test identification method of multicomponent gas mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia, which includes the formation of a matrix of piezosensors with different selectivity for the analyzed components, the preparation and sampling of the equilibrium gas phases of the sorbates with their subsequent input to the detection cell, registration and processing of analytical sorption signals, calibration of sensors, construction of “visual prints” of standard and studied mixtures, are new It is that for the identification of benzene, toluene, phenol, formaldehyde, acetone and ammonia, a matrix of six different sensors is formed, for which Apiezon-N films, Triton X-100 mixtures with Pleurotus higher mycelium fungus are applied to the electrodes of piezoelectric quartz resonators Ostreatus, polyethylene glycol adipate, polyethylene glycol sebacinate, polyvinylpyrrolidone and dinonyl phthalate with masses of 15-25 μg; after injecting the samples into the detection cell, the sensors are polled in the following sequence: a sensor with an Apiezon-N film is interrogated after 5 and 10 s, sensors based on a Triton X-100 mixture with Pleurotus Ostreatus higher mycelial fungus extract after 20, 25, 30 s, polyethylene glycol adipate after 55 s, polyethylene glycol sebacinate after 60 s, polyvinylpyrrolidone after 70, 75 s and dinonyl phthalate after 80, 85, 90 s; to predict “visual imprints” of standard mixtures, a unidirectional three-layer neural network is used, trained according to the results of sorption on each sensor and the physicochemical properties of the mixture components: molar masses, molar refraction coefficients, dielectric constant, sorbate density, boiling point, analyte saturated vapor pressure, the presence and number of oxygen, nitrogen, and CH ₃ atoms in the molecule — a group or other substituents; when analyzing real gas systems containing benzene, toluene, phenol, formaldehyde, acetone and ammonia, the obtained “visual prints” are compared with the “visual prints” of standard mixtures available in the database, and the geometry of the prints makes a conclusion about the degree of their identity and the qualitative composition the studied mixture, and the area of the diagrams is a quantitative criterion.

The technical result of the invention is to increase the expressivity of the analysis of mixtures of complex composition, to increase the number of identifiable components, to simplify the process of obtaining standard “visual prints”, to recognize components of the systems under study and to predict the geometry of the prints even when the concentration of individual compounds is changed.

Figure 1 presents the model of the artificial neural network (ANN).

(11); аммиак: присутствие атома азота (12); формальдегид: Mr (13); ρ (14).Figure 2 shows the significance of the learning criteria (ANN): acetone: P _us (1); the presence of an oxygen atom (2); CH ₃ groups (3); phenol: Mr (4); R (5); T _bale (6); the presence of an oxygen atom (7); benzene: ε (8); T _bale (9); toluene: Mr (10);

(eleven); ammonia: the presence of a nitrogen atom (12); formaldehyde: Mr (13); ρ (14).

Figure 3 shows the theoretical (a) and experimental (b) "visual prints" of multicomponent gas mixtures, the concentration of components at the MPC level.

The method of test identification of multicomponent gas mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia is as follows.

1) Preparation of the sensor matrix. The electrodes of 6 piezoelectric crystals with a natural oscillation frequency of 8-10 MHz are modified by uniformly applying sorbent solutions of a certain volume and concentration so that after removal of the solvents the film mass is 15-25 μg. A sensor with Apiezon-N film is interrogated after 5 and 10 seconds after injection of the sample, a sensor based on a mixture of Triton X-100 with extract of the higher mycelial fungus Pleurotus Ostreatus - after 20, 25, 30 seconds, polyethylene glycol adipate - after 55 seconds, polyethylene glycol sebacinate - after 60 s, polyvinylpyrrolidone after 70, 75 s and dinonyl phthalate after 80, 85, 90 s. The choice of sorbents is related to their cross selectivity to pairs of the analyzed compounds.

2) Formation of standard gas mixtures and sampling. Standard mixtures of sorbates are formed by "sequential mixing of samples in a container" within the daily average maximum permissible concentrations (MPC _s.s. ) of each of the components of the mixture. Sampling of standard mixtures from the "container" is carried out by the method of "discrete gas extraction" followed by injection into the cell detection.

3) Receiving analytical signals. After injecting the samples into the detection cell, the resonance frequency of each of the 6 sensors is measured and the analytical sorption signals are calculated:

where F _o - the initial frequency of the oscillations of the sensor, Hz; F is the oscillation frequency of the sensor after sorption, Hz.

The total analytical signal of the sensor matrix is represented in the form of “visual prints” of gas mixtures, which allows you to graphically characterize the state of the gas system. Taking into account possible combinations of benzene, toluene, phenol, acetone, formaldehyde and ammonia, to obtain standard “visual imprints” of their mixtures, it is necessary to compile and analyze 1134 model gas systems (with a minimum of three times the experiment). The complexity and duration of the process increases significantly with changes in the concentration of sorbates in the mixtures. The use of regression analysis methods greatly simplifies the task of obtaining “visual imprints” of standard mixtures, however, for the studied sorbates with their simultaneous presence in the systems, competitive sorption is characteristic. Therefore, to predict the total signal of the sensor matrix and the further construction of “visual prints” of standard mixtures, the nonparametric data processing method is used - artificial neural networks.

4) Processing the sorption signals of the polysensor matrix. The total signal of all 6 sensors is processed using the standard Neuro Network program (NeuroPro version 0.25). The network operation algorithm implements the black box model (Fig. 1). The neural network is trained to solve the problem on the basis of some training sample (“problem book”) and then is able to solve examples that are not included in it (test sample). All neurons of an artificial neural network (ANN) are grouped into several layers. In this case, the neurons of each subsequent layer receive the output signals of the neurons of the previous layers; in turn, their output signals arrive at the neurons of the next layer, forming a specific network architecture, Table 1. The described approach was used to predict "visual imprints" of standard mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia.

)] и качественные параметры, учитывающие строение молекулы сорбата, присутствие атомов кислорода, азота, СН₃-группы [учитывали с помощью коэффициентов заместителей]); на выход - прогнозируемый аналитический сигнал матрицы сенсоров в парах смеси (

). Давление насыщенного пара позволяет учесть концентрацию вещества в газовой фазе, температура кипения является индивидуальной физической характеристикой и не зависит от концентрации вещества. Выбор молярной рефракций в качестве обучающего параметра сети обоснован наибольшей информативностью по сравнению с плотностью, поляризуемостью и показателем преломления. Молярная рефракция учитывает способность молекулы поляризоваться и зависит только от природы соединения.Analytical signals from each of the six sensors obtained from the sorption of analytes from the training sample ΔF _c and standard physicochemical properties of the individual components of the mixture [molar mass (Mr), molar refraction (R), dielectric constant (ε), saturated steam pressure (P _n ), boiling point (T _k ), sorbate density (ρ), refractive index (

)] and qualitative parameters, taking into account the structure of the sorbate molecule, the presence of oxygen, nitrogen, and CH ₃ atoms [were taken into account using substitution factors]); output - the predicted analytical signal of the sensor matrix in the mixture vapor (

) The saturated vapor pressure allows you to take into account the concentration of the substance in the gas phase, the boiling point is an individual physical characteristic and does not depend on the concentration of the substance. The choice of molar refraction as the training parameter of the network is justified by the highest information content in comparison with the density, polarizability, and refractive index. Molar refraction takes into account the ability of the molecule to polarize and depends only on the nature of the compound.

Before applying to the network, each component of the input data vector is replaced with a normalized value calculated according to equation (2):

- нормированное значение входного параметра; x_i - компонента входного вектора; max x_i и min x_i - максимальное и минимальное значения компоненты, вычисленные по всей обучающей выборке.Where

- normalized value of the input parameter; x _i is the component of the input vector; max x _i and min x _i are the maximum and minimum values of the component calculated over the entire training sample.

The output signals of the network are normalized according to equation (3):

For any change in the training sample, the data normalization rule is applied.

An empirical approach is used to establish the optimal network topology and select the most significant learning parameters. After choosing the most informative training parameters, ANN testing is carried out on the training sample of mixtures. Upon reaching the specified forecast accuracy between the experimental results and the data predicted by the network for the training sample, the optimal ANN topology and the parameters of its training are established, which are used in the future to make a forecast of the sorption signal for all standard mixtures. The significance of the input parameters of the neural network is shown in figure 2. The most informative input parameters for training the network are the molar masses of the compounds, the molar refraction coefficients and dielectric constant, boiling points, as well as qualitative criteria - the presence of oxygen, nitrogen, and CH ₃ atoms. The saturated vapor pressure of the compounds, except for acetone, is the most insignificant input criterion for network training. The significance of the vapor pressure of acetone is justified by its anomalously high concentration in the gas phase (two orders of magnitude higher than for other compounds from the sample), therefore, for acetone, this parameter is perceived by the network as the most important.

For training and testing, ANNs composed model gas mixtures (Table 2). In the absence of a component in the mixture, zero was put in the corresponding cell (taking into account the normalization of -1). When modeling the composition of mixtures, the content of all components corresponds to their MPC _s.s .

) для смесей из тестируемой выборки.After establishing the optimal structure of the ANN (determined when the specified error level of 0.002% is reached) in the training sample, the difference between ΔF _5c experimentally obtained and predicted by the network was used to predict the sorption signal of the mixture components on each modifier (

) for mixtures from the test sample.

5) Identification of the components of the mixture. When analyzing real gas systems containing benzene, toluene, phenol, formaldehyde, acetone and ammonia, practical “visual prints” obtained in the polysensor matrix are compared with the “visual prints” of standard mixtures available in the database (Fig. 3 - predicted (a) and experimental (b) “visual imprints” of the sensor matrix in the analysis of multicomponent gas mixtures, component concentrations at the MPC level _s.c. ). Based on the geometry of the prints and the degree of their identity, they conclude about the qualitative composition of the mixture under study; chart area is a quantitative criterion.

The test identification method of multicomponent gas mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia is illustrated by the following example.

Example 1 (prototype)

A method for multivariate calibration of a multisensor sorption sensor for analyzing ethyl acetate, acetone, and toluene vapors in air involves sampling aialites using a discrete gas extraction method, forming a matrix of three sensors based on films of polyethylene glycol and its zfirs, injecting three-component mixtures into the detection cell, and then registering the analytical signal , graduation of the matrix in a wide range of concentrations. After registration of analytical signals, “visual prints” of both individual components and their mixtures are built. The method is feasible for analysis of a three-component gas mixture of ethyl acetate, acetone and toluene, however, it does not allow predicting the sorption signal when the composition of the mixture or the concentrations of its individual components change.

Example 2 (the claimed method)

Mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia are identified using a polysensor matrix. To do this, the electrodes of 6 piezoelectric crystals with an eigenfrequency of 8-10 MHz are modified by uniformly applying sorbent solutions with a chromatographic microsyringe so that after removal of solvents the mass of sorbent films is 15-25 μg, Apieson-N (Ap-N) films were used as sorbents, biomodifier (a mixture of Triton X-100 and extracts of the mycelial fungus Pleurotus Ostreatus - SmBio), polyethylene glycol adipate (PEGA), sebacinate (PEGS), polyvinylpyrrolidone (PVP) and dinonylphthalate (DNF).

Taking into account the sorption kinetics of toluene, phenol, benzene, acetone, formaldehyde, and ammonia, we compiled an algorithm for interrogating the sensor matrix when analyzing a real mixture: a sensor with a SmBio film should be interrogated to detect ammonia 20, 25, and 30 s after injection of the gas mixture into the cell (ammonia in mixtures inhibits phenol sorption); sensor with Ap-N film - after 5 and 10 s (toluene), PEGA - after 55 s (ammonia), PEGS - after 60 s (toluene), PVP - after 70 and 75 s (benzene, acetone, formaldehyde, phenol) , a sensor with a DNF film during phenol detection should be interrogated after 80, 85 and 90 s after injection of the mixture into the cell.

Gaseous mixtures of sorbates were formed by the method of “mixing in a container” and were selected by the method of “discrete gas extraction”.

After injection of the samples into the cell, the resonance frequency of the oscillations of each of the 6 sensors was measured and the analytical sorption signals (ΔF _c , Hz) were calculated.

The total analytical signal of the sensor matrix was presented in the form of “visual prints” of gas mixtures. Processing the total signal of all 6 sensors was carried out using the standard Neuro Network program (NeuroPro version 0.25). The network operation algorithm implements the black box model (Fig. 1). The characteristics of the ANN used are given in table 1.

), и качественные параметры, учитывающие строение молекулы сорбата, присутствие атомов кислорода, азота, СН₃-группы [учитывали с помощью коэффициентов заместителей]); на выход - прогнозируемый аналитический сигнал матрицы сенсоров в парах смеси (

).Analytical signals from each of the six sensors, obtained from the sorption of analytes from the training sample (ΔF _c ), and standard physicochemical properties of the individual components of the mixture (molar mass (Mr), molar refraction (R), and dielectric constant (ε) ), saturated vapor pressure (P _n ), boiling point (T _k ), sorbate density in the liquid phase (ρ), refractive index (

), and qualitative parameters that take into account the structure of the sorbate molecule, the presence of oxygen, nitrogen, and CH ₃ atoms [were taken into account using substitution factors]); output - the predicted analytical signal of the sensor matrix in the mixture vapor (

)

Before applying to the network, each component of the input data vector was replaced with a normalized value calculated according to equation (2). The output signals of the network were normalized according to equation (3).

An empirical approach was used to establish the optimal network topology and select the most significant learning parameters. Made up a network:

- trained only on the basis of the physicochemical properties of sorbates;

- trained on the basis of data on the structure of the molecules of the compounds (number and nature of substituents);

- trained on the basis of data on the structure of compounds and their physico-chemical characteristics (universal neural network).

For each sensor, taking into account the polling time, an individual network was created with a similar set of input and output parameters. The number of network training cycles for each sensor (taking into account the polling time was an individual network) is 11273, the training step is 0.0084, and the forecast accuracy is ± 0.002. Then, a uniform simplification of the network was carried out, the most significant input training parameters were determined, and interfering (noise) training criteria were excluded. The significance of the input network parameters is shown in figure 2.

After selecting the most significant training parameters, an ANN consisting of model gas mixtures was tested (Table 2).

строили «визуальные отпечатки» сигналов сенсоров (фиг.3а), которые сопоставляли с экспериментально полученными по результатам суммарной сорбции смесей ΔF_c (фиг.3б). При достижении заданной точности прогноза

между прогнозируемыми сигналами сорбции для смесей из обучающей выборки и экспериментальными данными устанавливали оптимальную топологию сети и параметры ее обучения, которые применяли в дальнейшем для прогнозирования суммарного сигнала сорбции газовых смесей из тестируемой выборки.According to the results of testing standard gas mixtures from the training sample by values

built "visual fingerprints" of the sensor signals (figa), which were compared with experimentally obtained by the results of the total sorption of mixtures ΔF _c (fig.3b). Upon reaching the specified forecast accuracy

between the predicted sorption signals for mixtures from the training sample and the experimental data, the optimal network topology and the parameters of its training were established, which were used later to predict the total sorption signal of gas mixtures from the tested sample.

не более 3%. Исключение составляет сенсор с пленкой ПВП, средняя погрешность по 7 модельным смесям 15,7%. Несоответствие прогноза экспериментальным результатам на этой пленке объясняется высокой полярностью модификатора и его высокой чувствительностью практически ко всем соединениям тестируемых смесей (конкурентная сорбция). С увеличением числа компонентов в смеси закономерно возрастает погрешность прогнозирования на пленке ПВП (максимальная для смесей 5 компонентов).For the studied mixtures from the test sample on almost all sensors, the prediction error of the total analytical sorption signal

no more than 3%. An exception is a sensor with a PVP film; the average error in 7 model mixtures is 15.7%. The forecast mismatch with the experimental results on this film is explained by the high polarity of the modifier and its high sensitivity to almost all compounds of the tested mixtures (competitive sorption). With an increase in the number of components in the mixture, the prediction error on the PVP film naturally increases (maximum for mixtures of 5 components).

Comparative characteristics of the proposed method and prototype are given in table.3.

Table 1 Number Learning algorithm Optimization method Rate of accumulation of significance The average forecast error for the samples,% input parameters output parameters layers neurons 61 one 3 10 back propagation of error conjugate gradients sum of modules of input parameters 3

table 2 Sorbate Training Testing presence in the mixture presence in the mixture Acetone + + + + + + + + + + Phenol + + + + + + + + + + + + + Benzene + + + + + + + + + + Ammonia + + + + + + + Toluene + + + + + + + + Formaldehyde + + + + + + + + + +

As can be seen from example 1 and 2, tables 1-3 and figure 1-3, a positive effect in the test identification of multicomponent gas mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia is achieved by using a six-sensor system based on Apiezon films N (the sensor is interrogated 5 and 10 seconds after injection of the samples), a mixture of Triton X-100 with extract of the higher mycelial fungus Pleurotus Ostreatus (20, 25, 30 s), polyethylene glycol adipinate (after 55 s), polyethylene glycol sebacinate (60 s), polyvinylpyrrolidone (70, 75 s) and dinonyl phthalate (80, 85, 90 s) with masses of 15-25 μg. To predict the “visual imprints” of standard mixtures, a unidirectional three-layer neural network is used, trained according to the results of sorption and the physicochemical properties of the mixture components: molar masses, molar refraction coefficients, dielectric constant, sorbate density, boiling point, saturated vapor pressure, presence and number in a molecule of oxygen, nitrogen and CH ₃ atoms or other substituents.

Changing the nature, the number of sensors in the matrix, the algorithm for their survey makes it impossible to identify gas mixtures. An increase in the stratification of ANNs, changes in data on the structure of compounds or their physicochemical criteria during network training does not allow predicting the “visual imprints” of standard mixtures. The total duration of the analysis of mixtures, taking into account the comparison of “visual prints” of real samples with standard samples, is not more than 5 minutes.

The proposed method for the test identification of multicomponent gas mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia makes it possible to recognize the components of the systems under study and to predict the geometry of the prints even when changing the concentration of individual compounds, to increase the expressivity of the analysis of mixtures of complex composition, to increase the number of identifiable components, to simplify the process of obtaining standard "visual prints."

Claims (1)

The method of test identification of multicomponent gas mixtures of benzene, toluene, phenol, formaldehyde, acetone and ammonia, including the formation of a matrix of piezoelectric separators with different selectivity for the analyzed components, the preparation and sampling of the equilibrium gas phases of the sorbates, followed by their entry into the detection cell, registration and processing of analytical sorption signals, calibration of sensors, construction of “visual imprints” of standard and test mixtures, the new is that for the identification of benzene, toluene, fe of nol, formaldehyde, acetone and ammonia form a matrix of six different sensors, for which Apiezon-N films, Triton X-100 mixtures with Pleurotus Ostreatus higher mycelial fungus extract, polyethylene glycol adipate, polyethylene glycol polyacin dinonyl phthalate with masses of 15-25 μg; after injecting the samples into the detection cell, the sensors are polled in the following sequence: a sensor with an Apiezon-N film is interrogated after 5 and 10 s, sensors based on a Triton X-100 mixture with Pleurotus Ostreatus higher mycelial fungus extract after 20, 25, 30 s, polyethylene glycol adipate after 55 s, polyethylene glycol sebacinate after 60 s, polyvinylpyrrolidone 70, 75 s and dinonyl phthalate after 80, 85, 90 s; to predict “visual imprints” of standard mixtures, a unidirectional three-layer neural network is used, trained according to the results of sorption on each sensor and the physicochemical properties of the mixture components: molar masses, molar refraction coefficients, dielectric constant, sorbate density, boiling point, analyte saturated vapor pressure, the presence and number of oxygen, nitrogen and CH ₃ atoms or other substituents in the molecule; when analyzing real gas systems containing benzene, toluene, phenol, formaldehyde, acetone and ammonia, the obtained “visual prints” are compared with the “visual prints” of standard mixtures available in the database, and the geometry of the prints makes a conclusion about the degree of their identity and the qualitative composition the studied mixture, and the area of the diagrams is a quantitative criterion.

Images