Application of an artificial neural network (ANN) and piezoelectric chemical sensor array for identification of volatile organic compounds

A piezoelectric chemical sensor array was developed using four quartz crystals. Gas chromatographic stationary phases were used as sensing materials and the array was connected to an artificial neural network (ANN). The application of the ANN method proved to be particularly advantageous if the measured property (mass, concentration, etc.) should not be connected exactly to the signal of the transducers of the piezoelectric sensor. The optimum structure of neural network was determined by a trial and error method. Different structures were tried with several neurons in the hidden layer and the total error was calculated. The optimum values of primary weight factors, learning rate (η=0.15), momentum term (μ=0.9), and the sigmoid parameter (β=1) were determined. Finally, three hidden neurons and 900 training cycles were applied. After the teaching process the network was used for identification of taught analytes (acetone, benzene, chloroform, pentane). Mixtures of organic compounds were also analysed and the ANN method proved to be a reliable way of differentiating the sensing materials and identifying the volatile compounds.     

Quartz crystal microbalance sensor array for the detection of volatile organic compounds

A sensor array system consisting of five quartz crystal microbalance (QCM) sensors (four for measuring and one for reference) and an artificial neural network (ANN) method is presented for on-line detection of volatile organic compounds. Three ionic liquids, 1-butyl-3-methylimidazolium chloride (C₄mimCl), 1-butyl-3-methylimidazolium hexafluorophosphate (C₄mimPF₆), 1-dedocyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide (C₄mimNTf₂), and silicone oil II, which is widely used as gas chromatographic stationary phase, have been selected as sensitive coatings on the quartz surface allowing the sensor array effective to identify chemical vapors, such as toluene, ethanol, acetone and dichloromethane. The success rate for the qualitative recognition reached 100%. Quantitative analysis has also been investigated, within the concentration range of 0.6-6.1 mg/L for toluene, 0.9-7.5 mg/L for ethanol, 2.8-117 mg/L for dichloromethane, and 0.7-38 mg/L for acetone, with a prediction error lower than 8%.     

A piezoelectric quartz crystal sensor array self assembled calixarene bilayers for detection of volatile organic amine in gas

P-tert-butylcalix[n]arenes (n=4, 6, 8, abbreviated as CA[4], CA[6], CA[8], respectively) were immobilized on the Au surface of the piezoelectric quartz crystal by the reaction between CA[n] and the acid chloride terminated mercaptoacetic acid (MAA) self assembled monolayers to form MAA/CA[n] bilayers. The sensing films were not only immobilized easily and reproducibly, but also used to improve the reversibility of the sensor signal. The response characteristics show the response of CA to organic amine attributes to specific interaction between CA[n] (host) and organic amines (guest). The frequency shifts of n-butylamine and iso-butylamine are much larger than tert-butylamine and diethylamine because of shape-selection and hydrogen bonding. Compared to CA[6] and CA[4], CA[8] has highest sensitivity to organic amine due to having more flexibility to accommodate guest molecules. A sensor array with three-layer back-propagation neural network was applied to detect the binary mixture of n-butylamine in the range of 7.14–142 μl l⁻¹ and iso-butylamine in the range of 7.14–57 μl l⁻¹. The optimum values of learning rate (0.15) and momentum term (0.8) were determined by experiment. The best epoch of training was 1098. The root mean square error of prediction was 1.69 (μl l⁻¹) for n-butylamine, and 1.42 (μl l⁻¹) for iso-butylamine.     

Studies on the applicability of artificial neural network (ANN) in emulsion liquid membranes

An artificial neural network (ANN) model of emulsion liquid membrane (ELM) process is proposed in the present study which is able to predict solute concentration in feed during extraction operation and ultimate % extraction at different initial solute concentration in feed phase, internal reagent concentration, treat ratio, volume fraction of internal aqueous phase in emulsion and time. Because of the complexity in generalization of the phenomenon of ELM process by any mathematical model, the neural network proves to be a very promising method for the purpose of process simulation. The network uses the back-propagation algorithm (BPA) for evaluating the connection strengths representing the correlations between inputs (initial solute concentration in feed phase, internal reagent concentration, treat ratio, volume fraction of internal aqueous phase in emulsion and time) and outputs (solute concentration in feed during extraction operation and % extraction). The network employed in the present study uses five input nodes corresponding to the operating variables and two output nodes corresponding to the measurement of the performance of the network (solute concentration in feed during extraction and % extraction). Batch experiments are performed for separation of nickel(II) from aqueous sulphate solution of initial concentration in the 200–100 mg/l ranges. The network employed in the present study uses two hidden layers of optimum number of nodes being thirty and twenty. A leaning rate of 0.3 and momentum factor of 0.4 is used. The model predicted results in good agreement with the experimental data and the average deviations for all the cases are found to be well within ±10%.     

Prediction of extraction efficiency in supported liquid membrane with a stagnant acceptor phase by means of artificial neural network

An artificial neural network model of supported liquid membrane extraction process with a stagnant acceptor phase is proposed. Triazine herbicides and phenolic compounds were used as model compounds. The model is able to predict the compound extraction efficiency within the same family based on the octanol–water partition coefficient, water solubility, molecular mass and ionisation constant of the compound. The network uses the back‐propagation algorithm for evaluating the connection strengths representing the correlations between inputs (octanol–water partition coefficients logP, acid dissociation constant pK_a, water solubility and molecular weight) and outputs (extraction efficiency in dihexyl ether and undecane as organic solvents). The model predicted results in good agreement with the experimental data and the average deviations for all the cases are found to be smaller than ±3%. Moreover, standard statistical methods were applied for exploration of relationships between studied parameters.     

Selection of quasi-optimal inputs in chemometrics modeling by artificial neural network analysis

Instrumentation spectra used for chemometrics analysis are often too unwieldy to model, as many of the inputs do not contain important information. Several mathematical methods are used for reducing the number of inputs to the significant ones only. Artificial neural networks (ANN) modeling suffers from difficulties in training models with a large number of inputs. However, using a non-random initial connection weight algorithm and local minima avoidance and escape techniques can overcome these difficulties. Once the ANN model is trained, the analysis of its connection weights can easily identify the more relevant inputs. Repeating the process of training the ANN model with the reduced input set and the selection of the more relevant inputs can proceed until a quasi-optimal, small, set of inputs is identified. Two examples are presented—finding the minimal set of wavelengths in benchmark diesel fuel NIR spectra, and in spectra generated in a recent work, modeling of “artificial nose” sensor array. In the last example, 1260 inputs were reduced to optimal sets of <10 inputs. Causal index calculation can analyze the influence of each of selected wavelengths on the predicted property. Some of the resulting minimal sets are not unique, depending on the ANN architecture used in the training. The accuracy of the resulting ANN models is usually better, and more robust, than the original large ANN model.     

Electrochemical inhibition bacterial sensor array for detection of water pollutants: artificial neural network (ANN) approach

This work reports on further development of an inhibition electrochemical sensor array based on immobilized bacteria for the preliminary detection of a wide range of organic and inorganic pollutants, such as heavy metal salts (HgCl₂, PbCl₂, CdCl₂), pesticides (atrazine, simazine, DDVP), and petrochemicals (hexane, octane, pentane, toluene, pyrene, and ethanol) in water. A series of DC and AC electrochemical measurements, e.g., cyclic voltammograms and impedance spectroscopy, were carried out on screen-printed gold electrodes with three types of bacteria, namely Escherichia coli, Shewanella oneidensis, and Methylococcus capsulatus, immobilized via poly l-lysine. The results obtained showed a possibility of pattern recognition of the above pollutants by their inhibition effect on the three bacteria used. The analysis of a large amount of experimental data was carried out using an artificial neural network (ANN) programme for more accurate identification of pollutants as well as the estimation of their concentration. The results are encouraging for the development of a simple and cost-effective biosensing technology for preliminary in-field analysis (screening) of water samples for the presence of environmental pollutants.

Graphical abstract

     

Instrument for generation of volatile organic compounds for the development of chemical sensors and evaluation of tube-type diffusive sampling

An automatic calibration apparatus for the dynamic generation of organic vapours was developed. The accurately controlled stream of nitrogen was drawn at a low flow-rate through a thermostated container filled with the standard substance, thus generating a continuous stream of saturated vapour of the compound. The compound holder vessel was thermostated at −16°C. A large stream of pure carrier gas was mixed with a low stream of substance in a mixing chamber for dilution. The fittings were manufactured from PTFE, and tubes were made of special PTFE with an inert inner surface to eliminate the wall adsorption and to decrease the cross-diffusion. Moisture interferences were reduced using a Nafion membrane filter. The vapour generator was validated by diffusive sampling and gas chromatographic methods. Standard mixtures have been prepared containing toluene at concentrations ranging from 3 to 3000 ppm. The combined uncertainty of preparative and analytical error components associated with the concentration of the analytes at the 95% confidence level typically ranges from 2 to 5% relative, depending upon the concentration. The measured and the calculated values were compared and good correlation (r²>0.99) was found.     

Evaluation of doped and undoped poly (o‐anisidine) as sensing materials for a sensor array for volatile organic compounds

Poly (o‐anisidine) (PoANI) and PoANI doped with nickel oxide and zinc oxide were evaluated as sensing materials for four gas analytes (methanol, ethanol, acetone, and benzene). The sensing materials had high sensitivity (showing an affinity towards the target analytes even at low concentrations, in the range of 1‐5 ppm), but rather poor selectivity, especially when the gas analytes were in a mixture. To exploit the poor selectivity, the three sensing materials were combined into a sensor array using principal component analysis (PCA) as a sensing algorithm. It was found that using a sensor array, the four individual gases could be separated. However, when all four gases were present (in analyte mixtures), there was too much overlap in the responses to distinguish between individual gas analytes and their related mixtures.     

Recognition of adulteration of Italian wines by thin-film multisensor array and artificial neural networks

The paper reports the use of a chemoresistive multisensor array for recognition of some adulterated Italian wines (two white, four red, two rosè) added with methanol, ethanol or other same-colour wine. A multisensor array constituted by four thin-film semiconducting metal oxide sensors, surface-activated by Pt, Au, Pd, Bi metal catalysts, has been used to generate the chemical pattern of the volatile compounds present in the wine samples. The responses of the multisensor array towards wines tested by headspace sampling have been evaluated. Multivariate analysis including principal component analysis (PCA) as well as back-propagation method trained artificial neural networks (ANNs) have been applied to analytical data generated from the multisensor array to identify both the adulteration of wines and to determine the added content of adulterant agent of methanol or ethanol up to 10 vol.%. The cross-validated ANNs provide the highest achieved percentage of correct classification of 93% and the highest achieved correlation coefficient of 0.997 and 0.921 for predicted-versus-true concentration of methanol and ethanol adulterant agent, respectively.     

2D bitmapping approach for identification and quantitation of common base flavor adulterants using surface acoustic wave arrays and artificial neural network data analysis

Arrays of polymer-coated surface acoustic wave microsensors are used in conjunction with a variety of signal-processing algorithms known as artificial neural networks (ANN). This format of data analysis has the capability to characterize complex mixtures of volatile and semi-volatile organic compounds commonly found in base flavors. The approach described, which minimizes the number of training sets while retaining the robustness of an ANN, utilizes a 2D bitmap matrix. The matrix is obtained by converting the time domain kinetics of sensor response into a bitmap. The high data throughput of this approach enables quantitation on the order of ppm of common base flavor adulterants.     

Application of pattern recognition and piezoelectric sensor array for the detection of organic compounds

Organic vapours were measured by an array of piezoelectric crystal detectors. Quartz crystals were coated by different GC stationary phases. Four coated crystals were placed in an array and pattern recognition was used for identification of the compounds including acetone, benzene, chloroform and pentane. A computer program was developed for the measurement of the frequency changes and data processing. Pattern recognition method using feature extraction was applied for identification of analytes.     

Use of sequential injection analysis to construct an electronic-tongue: application to multidetermination employing the transient response of a potentiometric sensor array

An electronic tongue based on the transient response of an array of non-specific-response potentiometric sensors was developed. A sequential injection analysis (SIA) system was used in order to automate its training and operation. The use of the transient recording entails the dynamic nature of the sensor's response, which can be of high information content, of primary ions and also of interfering ions; these may better discriminated if the kinetic resolution is added. This work presents the extraction of significant information contained in the transient response of a sensor array formed by five all-solid-state potentiometric sensors. The tool employed was the Fourier transform, from which a number of coefficients were fed into an artificial neural network (ANN) model, used to perform a quantitative multidetermination. The studied case was the analysis of mixtures of calcium, sodium and potassium. Obtained performance is compared with the more traditional automated electronic tongue using final steady-state potentials.     

Detection and classification of gaseous sulfur compounds by solid electrolyte cyclic voltammetry of cermet sensor array

Electrochemical sensors composed of a ceramic-metallic (cermet) solid electrolyte are used for the detection of gaseous sulfur compounds SO₂, H₂S, and CS₂ in a study involving 11 toxic industrial chemical (TIC) compounds. The study examines a sensor array containing four cermet sensors varying in electrode-electrolyte composition, designed to offer selectivity for multiple compounds. The sensors are driven by cyclic voltammetry to produce a current-voltage profile for each analyte. Raw voltammograms are processed by background subtraction of clean air, and the four sensor signals are concatenated to form one vector of points. The high-resolution signal is compressed by wavelet transformation and a probabilistic neural network is used for classification. In this study, training data from one sensor array was used to formulate models which were validated with data from a second sensor array. Of the 11 gases studied, 3 that contained sulfur produced the strongest responses and were successfully analyzed when the remaining compounds were treated as interferents. Analytes were measured from 10 to 200% of their threshold-limited value (TLV) according to the 8-h time weighted average (TWA) exposure limits defined by the National Institute of Occupational Safety and Health (NIOSH). True positive classification rates of 93.3, 96.7, and 76.7% for SO₂, H₂S, and CS₂, respectively, were achieved for prediction of one sensor unit when a second sensor was used for modeling. True positive rates of 83.3, 90.0, and 90.0% for SO₂, H₂S, and CS₂, respectively, were achieved for the second sensor unit when the first sensor unit was used for modeling. Most of the misclassifications were for low concentration levels (such 10-25% TLV) in which case the compound was classified as clean air. Between the two sensors, the false positive rates were 2.2% or lower for the three sulfur compounds, 0.9% or lower for the interferents (eight remaining analytes), and 5.8% or lower for clean air. The cermet sensor arrays used in this analysis are rugged, low cost, reusable, and show promise for multiple compound detection at parts-per-million (ppm) levels.     

Comparative study of artificial neural network and multivariate methods to classify Spanish DO rose wines

Classical multivariate analysis techniques such as factor analysis and stepwise linear discriminant analysis and artificial neural networks method (ANN) have been applied to the classification of Spanish denomination of origin (DO) rose wines according to their geographical origin. Seventy commercial rose wines from four different Spanish DO (Ribera del Duero, Rioja, Valdepeñas and La Mancha) and two successive vintages were studied. Nineteen different variables were measured in these wines. The stepwise linear discriminant analyses (SLDA) model selected 10 variables obtaining a global percentage of correct classification of 98.8% and of global prediction of 97.3%. The ANN model selected seven variables, five of which were also selected by the SLDA model, and it gave a 100% of correct classification for training and prediction. So, both models can be considered satisfactory and acceptable, being the selected variables useful to classify and differentiate these wines by their origin. Furthermore, the casual index analysis gave information that can be easily explained from an enological point of view.