Jan Mitrovics

Odours and flavours identified with hybrid modular sensor systems

Abstract Hybrid sensor systems contain different types of chemical sensors whereby each type (transducer principle) contains an array of individual sensors. This leads to a large flexibility in the choice of transducers and sensor materials with the general aim of optimising the analytical performance of the total system. This concept makes it possible to optimise the quantitative analysis of mixtures of known gases as it will be demonstrated for mixtures of volatile organic compounds (VOCs). Alternatively this makes it possible to optimise the system for characterising odours and flavours. This will be demonstrated for different plastic as well as textile materials used in car industries and for different products of food industries, i.e. coffees, tobaccos, whiskeys, and olive oils. In our modular sensor systems we used arrays of different semiconductor gas sensors (based on metal oxides), of polymer coated quartz microbalance (QMB) sensors, of calorimetric sensors and of electrochemical sensors, with an option to add metal oxide semiconductor field effect transistor (MOSFET) sensors. These arrays are arranged as separate components in a modular sensor system ‘MOSES’. For the qualitative discrimination of different odour samples a headspace-autosampler was added and transient sensor signals were monitored. The use of different transducer principles is shown to be essential for an unequivocal identification of odours and flavours.

Development of Modular Ozone Sensor System for application in practical use

Abstract A computerized Modular Ozone Sensor System (MOSS) for evaluating the sensitivity and reliability of different sensor/transducer combinations is presented. The system consists of a compact and portable apparatus that can be used to compare various sensors. The sensor types bench-marked using MOSS are electrochemical cells, various semiconducting gas sensors based on metal oxides (In2O3, SnO2), and Thickness Shear Mode Resonator (TSMR) sensors, which were mounted on separate modules. The gas mixtures containing known amounts of ozone were analyzed using this modular array of different sensors. All electronic sensor modules, including standardized connections are home-built and optimized as a part of this investigation. Computerized data acquisition and evaluation was performed using a standard hardware and an analysis software developed during this work. In order to test the setup, a set of different sensors was employed. Their sensitivity to ozone and their cross-sensitivity to other gases in ambient condition and to humidity were evaluated. Sensors, based on indium oxides, performed best in detecting ozone. They had the largest sensor sensitivity as well as the smallest cross-sensitivities.

Application of neural-network systems to the dynamic response of polymer-based sensor arrays

The conventional calibration method for sensor arrays uses steady-state signals that depend on the gas concentration. This method can be time consuming if many concentrations and compositions of a multicomponent mixture are required. Good experimental design may reduce the necessary effort so that the number of calibration experiments is minimized. Dynamic measurements may significantly reduce the time of each calibration experiment. In the present approach a random walk through the domain of the gas concentrations is chosen with each step of the walk adjusted for a short time only. The sensor array consists of six polymer (polysiloxanes with functional groups)-coated bulk acoustic wave (BAW) devices. The concentration domain is defined by a binary mixture of n-octane and toluene (150 to 800 ppm). Neural networks evaluate both qualitative and quantitative information from the sensor response. In particular, the extensions of feed-forward nets towards recurrent or time-delay structures can be used to solve problems related to dynamic evaluations (e.g., no steady-state signal, parameter drift). These network architectures with different numbers of hidden neurons are applied to evaluate the data from the BAW device array. The networks are trained with back-propagation-like training algorithms and are validated with arbitrary gas mixtures.

Different strategies for the identification of gas sensing systems

Abstract We have shown that it is possible to extract dynamical models for non-linear gas sensors from experimental input-output data. Seven different methods (two linear and five non-linear) have been evaluated in terms of their prediction performance on the sensor response to white gaussian inputs. Two methods, artificial neural networks and modified Wiener kernels estimated by least squares, show very low prediction errors with models extracted from only 300 input-output data pairs. A detailed discussion on the advantages and disadvantages of very method is presented.

Sensor arrays with only one or several transducer principles? The advantage of hybrid modular systems

Abstract Artificial olfaction by means of “electronic noses” aims at the determination of chemical features to characterize molecules, odors, environmental conditions, process parameters, etc. In order to achieve sufficient discrimination a broad spectrum of independent features has to be determined. Since the choice of suitable materials for each individual transducer is limited, different transducers have to be combined in a sensor system to obtain a sufficiently broad discrimination power. This is illustrated for typical examples chosen from the food and textile industries which illustrate the advantage of such “hybrid modular systems”.

A composed neural network for the recognition of gas mixtures

Artificial neural networks are generally considered as the most promising tools for untangling pattern-recognition problems in chemical sensing. Different neural networks have been shown to be suitable for solving partial aspects of the pattern recognition. For instance, feed-forward networks are particularly able to find out the rules for the feature extraction, while self-organizing maps show better behaviour in classification and identification tasks. In this paper a hybrid network, which exploits the benefits of both these networks, is introduced and applied to the identification of binary mixtures of organic solvent gases using a quartz-microbalance-based sensor array.

Quality tests of electronic noses: the influence of sample dilution and sensor drifts on the pattern recognition for selected case studies

Abstract Arrays of chemical sensors have a broad spectrum of applications, e.g., in the field of process control and quality analysis. Especially for the fast and objective evaluation of food quality and off-flavour contents in plastic materials, such sensor systems, often called “Electronic Noses”, have an increasing demand. In this context, the performance and reproducibility of an array analysis is usually not checked systematically. The present paper therefore deals with the influence of sample dilution (external effect) and of changed sensor parameters with time (internal effect) on the subsequent results of pattern recognition.

Structure identification of non-linear models for QMB polymer-coated sensors

Abstract This article deals with the representation of quartz-microbalance (QMB) polymer-coated sensors by block-structured models, their structural identification and their functional identification. The block-structured modelling approach is one of the techniques available for obtaining a complete description of the dynamic behaviour of a non-linear sensing device without having detailed information about its inner structure. A QMB sensor, employed for the detection of mixtures of n-octane and toluene, is studied at the high concentration range (up to 100000 ppm), where a non-linear behaviour is expected. The sensor is exposed to time-varying concentrations of toluene and n-octane having quasi-white power spectral density. It is recognized that the sensor is very likely to admit a special model structure, called the ′Wiener model′, which is a simple non-linear cascade consisting of sequential dynamic linear, static non-linear and dynamic linear processes. The model is completely identified and gives a good approximation to the response of the sensor to any time-varying concentration. Our approach has not yet been widely applied because of the complication of the structure testing procedure, but it is demonstrated that the method is well suited for the study of most chemical transducers provided certain considerations and preliminary analyses are made, as described in this paper.

Linearisation In Multicomponent Analysis Based On A Hybrid Sensor-array With 19 Sensor Elements

We present experimental results obtained with a modular hybrid sensor-array for gas mixture analysis and for odour sensing. [1] Sensor elements are based on mass sensitive, electrochemical, field effect, and electrical transducers. Linearisation in such a complex setup is an essential feature of the subsequent multicomponent analysis. Three different methods are applied for this task. To characterize in a first simple example the different performances of these approaches, we analysed various mixtures of carbon monoxide (CO) and toluene (tol.) in air. Principal Component Regression (PCR) and Partial Least Squares (PLS) were used for the quantitative multicomponent analysis.

Detection of off-odors using a hybrid modular sensor system

Experimental results are presented of a qualitative analysis of off-odors in packaging materials, obtained with a hybrid modular sensor system. The sensing elements are combined in two modules of the hybrid modular sensor system MOSES II which contains different types of transducers. The modules chosen in the present study consist of 8 metal oxide-based semiconductor gas sensors and 8 polymer coated quartz microbalance sensors (QMB). Pattern recognition was performed by principal component analysis (PCA). For packaging materials a discrimination between unprinted and printed materials could be achieved successfully by comparing three different samples. These results characterize the performance and reproducibility of the newly designed hybrid modular sensor system for investigations of a broad spectrum of other applications.