Session 3: Current possibilities and pitfalls in data processing

Reported by Tom Artursson, Chairman: Martin Holmberg

Jan Mitrovics, AppliedSensor, Germany

”Hands-on improvements using data processing in electronic noses”.

The speaker gave a practical view of PCA and LDA and compared the two methods. Another very interesting subject he came across was how to improve the robustness of the models. The very first thing to do is to find the error sources, from which the error comes; is it from the sample, the sampling system or the sensor itself? If the error is unknown it is hard or impossible to pre-treat the data in a way that really would improve the robustness. On the other hand, if the error sources are known the data preprocessing could be directed and the error itself could be reduced. He gave an example of how the temperature dependency of QMB sensors could be reduced with the help of normalization.

Ricardo Guiterrez-Osuna, Texas A&M University, USA

”Signal processing methods for drift compensation”.

The speaker gave a thorough presentation of different methods for drift counteraction, and possible reasons for drift were discussed. The drift reduction methods try to compensate for the changes in sensor performance using mathematical models, thus maintaining the gas identification capability of the electronic nose. The different drift counteraction methods were divided into two classes, univariate and multivariate drift counteraction. In the class univariate frequency analysis, baseline correction, differential measurements, and multiplicative correction were discussed. In the multivariate class adaptive clustering, system identification, non-linear regression, component correction, and deflation were discussed. From each method an example was given.

General discussion

In the discussion afterwards the question ”what is the most common problem in data analysis” was raised. Two subjects were discussed according to this questions, drift and miss of information. Since drift already was discussed by the second speaker, this discussion was quite short. The other problem, miss of information, is very serious since a lot of data analysis problems could be solved if more information was available. This highlights the importance of a close collaboration between the sensor scientists and data analysis community. Another topic that was discussed was the importance of validation of the models used to extract information from the sensor data. A common understanding was that there has been an improvement in this area if we look into the history of the electronic nose, but we could still make improvements in this area.

Session 4: New concepts for chemical sensing

Reported by Anita Lloyd-Spetz, Chairman: Ingemar Lundström

In the session new concepts for chemical sensing was presented, nanobelts as potentially both sensitive and stable sensor material, e.g. phthalocyanines and conducting polymers as electrode material for electronic tongues, theoretical modelling of basic sensor parameters and the use of the computer screen as an analytical tool in e.g. medical diagnosis.

Giorgio Sberveglieri, Brescia University, Italy

“Gas sensing based on seminconducting nanobelts: a new breakthrough”.

The gas introduced resistivity changes in SnO2 films is commonly used for gas sensing. The polycrystalline SnO2 material, which is normally used, suffers from grain growth causing drift. Single crystals, on the other hand, suffer from low sensitivity due to low surface area.

A method to prepare SnO2 single crystal nanobelts was presented. In a furnace SnO2 powder is thermally evaporated and transported by an argon flow to a substrate with interdigital electrodes, onto which the nanobelts are deposited. The surface area is larger for the belt structure, which increases the sensitivity even though this is single crystalline material, which in turn increases the stability. The nanobelts are stable up to 800°C and up to 2 mm long belts are achieved. Modelling of the depletion area in three types of nanostructures was shown. Examples of the gas-response were given and SEM pictures of nanobelts shown. To increase sensitivity even further catalysts will be included in the nanobelt material in the future, and new binary and ternary oxides will be used to produce nanobelt materials.

There were several questions after the presentation:

It was pointed out that in SnO2 sensor material normally the contact points between the grains are active in the gas sensing process. In the nanobelt material there are very few contact points. It was discussed whether this is detrimental for the gas-sensitivity.

What about the conductivity in one belt? Could it be used to understand the change in the sensitivity as compared to polycrystalline SnO2 material? Sberveglieri thinks the response time might increase for a single nanobelt used e.g. in a transistor.

Single crystals have low catalytic activity, which should be a drawback here. Well, the low catalytic activity is due to a low surface area, which is improved by the belt structure.

Are the nanobelt films easy to handle or brittle and is it possible to stretch a film to align the belts? The films stick very well to the surface of the IDT s, which is part of the reason why this kind of investigations are not yet performed in this very new research area.

Maria Luz Rodríguez-Méndez, University of Valladolid, Spain

“Novel voltammetric sensors based on phthalocyanines and conducting polymers for the evaluation of taste”.

Metallic voltammetric electrodes can be used as the sensing units of an electronic tongue. The group at the University of Valladolid has proposed a different approach that consists in the development of voltammetric electrodes modified with a range of materials with different electroactive properties that would allow preparing non-selective sensors endowed with sufficient cross-selectivities. Phthalocyanines or conducting polymers (CP) have been used as electrode materials.

In the case of phthalocyanines, electrodes are prepared using carbon paste electrode technique (CPE) or by depositing films (by the Langmuir-Blodgett films technology or ultra high vacuum evaporation) onto conducting substrates like indium tin oxide (ITO). A large variety of electrodes may be processed from different derivates of phthalocyanines with different central metal atoms. CPE showed the best long-term stability.

The conducting polymers were deposited by electropolymerisation onto a platinum wire. The electrochemical properties of the sensors strongly depend upon the synthesis conditions allowing a simple modification of the sensor response. Several polymers have been tested and polypyrrole (Ppy) gave the best result. However, the dopants of the Ppy tend to diffuse out into the solution and redox anions and big doping agents are recommended.

An array of phthalocyanines and CPs were used to discriminate between solutions of foods with bitter taste. The array was also tested for the response to red wines. For cleaning of electrodes KCl cycling can be used in between measurements

Questions raised after the presentation:

• Is there anything to gain from increasing the voltage?
  In the case of phthalocyanines, the voltage range can be as wide as the window of the solvent. Nevertheless, in the case of conducting polymers, voltages beyond 0.6 V disturb the behaviour of the electrode, probably because a dedoping process occurs.
• Is the catalytic activity increased by using redox and ionic properties?
  Phthalocyanines have a well-known catalytic activity that can be the reason for the apparition peaks that can be attributed to the oxidation of species such as glucose, which is not ionic.
• How large is the selectivity for different metals in the phthalocyanines?
  The oxidation potentials of the phthalocyanines are affected both by the central atom and by the presence of substituents in the phthalocyanine ring as follows:

Table 1 Peak potentials of LnPc2 LB films immersed in 0.1mol.L-1 KCl

• What about the lifetime of these sensors when measuring wine?
  The lifetime of the electrodes depends on the solution tested. In the case of simple solutions (NaCl, Quinine, etc), the lifetime can be 2-3 months. Nevertheless, when analysing complex solutions such as wines, a careful use of the electrodes is needed and frequent cleaning steps are required to avoid the poisoning of the electrodes. Even if these cares are taken, the lifetime is lower than for simple solutions (i.e. 20 -30 days)

Boris Snopok, Institute of semiconductor physics, Ukraine

“Non-exponential relaxations”.

Snopok had a theoretical approach to chemical sensors. The surface can be regarded as a collection of effective areas. Topography slows down the response. Strong absorbers are bulk controlled, while for weak adsorbers the bulk is not disturbed. The Langmuir approach was applied to the adsorption process. Surface motion is a long journey of chaotic motion. The adsorption is controlled by a distribution of rates. Adsorbers were divided into weak adsorbers which are Langmuir like, strong adsorbers which are diffusion like, and those in between which are fractal like. For example different organic materials, like phthalocyanines and pentacene etc. could be classified as Langmuir like or diffusion like.

For the solid gas interaction gas molecules may interact with receptor centers on the surface or in the bulk or sorption by the gas molecules in the lattice cavities of the solid may take place. This may lead to changes in the film structure, e.g. swelling. The driving forces are van der Waals forces giving different polarizability, induction forces of electrostatic origin showing up as different dipole moments, hydrogen bonding or donor acceptor bonding by nucleophilic groups.

An electronic nose for pharmaceutics and a QCM based e-nose as an artificial sniffer were given as examples. Finally spatially and temporally resolved analyte mapping was proposed in order to model artificial architectures for realization of nanoscale structures with self-adaptability self-sensing memory and multiple functionality…

Discussion after the presentation:

It was pointed out that cross-section is important.

It was questioned that the bulk really is involved for strong adsorbers, and that adsorption and absorption should be differentiated. Henri isotherm or the Langmuir will probably apply. The Langmuir like adsorption has OH group on the surface involved.

Daniel Filippini, Linköping university, Sweden

“Computer screen assisted technique”.

The computer screen is used as a light source for a chemical sensor system based on the scanned light pulse technique (SLPT) in for example medical diagnosis for simplicity and reduced cost. A web camera is used for readout. The nanoscale frequencies of the red, blue and green light can be used and expensive parts like chopper and mirrors can be replaced. Software has been developed for the configuration of a certain computer screen, including both SED and SRD screens. The files may be accessed from a web page.

Cell viabilities and response to antibiotics may be studied. For example pigment-analysis of aggregation of melatonin in fish-flake cells can be performed by this method. The pigment aggregation is sensitive to poisons and hormones, while the aggregation reflects the presence of these substances.

The use of the computer screen when running SLPT was also demonstrated. In this application also a lock-in amplifier is needed.

It was explained during the questioning that every computer screen needs calibration.

General discussion

The chairman listed, New concepts for Application Specific Chemical Sensor Systems, ASSS: New materials, new principles, new theoretical models and new infrastructure.

The four speakers were asked to tell us about the biggest challenge (problem) of their technology

Maria Luz Rodriguez-Mendez mentioned reproducibility and poisoning of the taste sensors as the biggest problem

Giorgio Sberveglieri needed models to explain the different behaviour of the nanobelts as compared to the polycrystalline SnO2 material. Questions to answer are for example, why are the cross section of the nanobelts rectangular? The growth method as well as the interpretation of the results when using nanobelts as chemical sensors needs modelling. What happens when one parameter is changed?

Daniel Filippini: To use the computer screen and a web camera as your equipment gives constraints to what you can do. The question is how far you can go, what measurements are possible to perform. The audience commented that these are commercial aspects. Fillippini concluded that this is an instrument to be used when there are constraints to the availability of equipment.

Boris Snopok: The sensor surfaces should be carefully defined in order to distinguish problems arising from the sensors and from the transducer.

The discussion now took a more general turn:

The importance of sampling was emphasized. A short course on this topic is soon coming up in the Nose II network. More specific sensors designed for a certain problem were suggested, but Sberveglieri commented that the electronic nose concept was designed as a solution to that.

It was also suggested that it is important to also spend some time to understand the basic mechanisms of the chemical sensors. It was commented that we should learn from the nature, use small size and multiply!

Udo Weimar concluded the discussion by reporting about the opinion of the EU commission. They want to know what we will do in the future regarding both new concepts but also academically in terms of training and education of researchers.

Session 5: Distributed sensor systems

Reported by Fredrik Winquist, Chairman: Udo Weimar

In this session the concept of integrated sensor system was presented. Thus, both general concepts as well as specific applications were dealt with. Could distributed sensor systems be a new concept for electronic noses or electronic tongues?

Martin Holmberg, FOI and Linköping University, Sweden

The starting point for this lecture was that sensor systems produce a lot of data.

Sometimes it is also necessary to distribute sensors over a large area. This includes applications such as for environmental monitoring, surveillance, or traffic control. When the area is large or the bandwidth is limited it is, however, not possible to send all data to a central point. Thus, information systems have to be designed for data fusion in a distributed and autonomous way.

In the lecture, it was shown how a network of co-operating physical sensors was used for classification and tracking of vehicles. One tool for this is to use an Unattended Ground Sensor Network (UGS). This is in principle not a new technique, but is simply based on many small sensor units spread out over a large area that have the possibility to communicate with each other. The new approach is that the sensors can be put into a network using intelligent agents, which are autonomous programs with a simple, but specific task. Use of agents offers many advantages, they can be added and subtracted from the system while it is running without requiring external intervention, they are self-configuring, and consistent with the object oriented paradigm. In a specific application, it was described how a hostile vehicle could be tracked and classified over a large area. Sensors will then be placed over the area of interest for surveillance. Two sensors, communicating with each other, will be sufficient for telling the direction and class of vehicle. The topology of the network is important, in principle it is based on that several sensors are connected with one node. Several nodes communicate with each other. The sensor data obtained the will consist of a signature and a direction. Association of data from different sensors is made autonomously before data fusion. The fusion will result in a position and classification. There are four agent types. There is one track agent for each known target, and when the assigned target is tracked the agent moves through the network and combines sensor data with its current perception of the track. The sensor agent controls activation/deactivation of the sensor, and sends data to the local dispatch agent. The dispatch agent passes sensor data to the right track agent. Finally, the node agent facilitates track agent movements in the network.

In an animation it was shown how a vehicle was followed and the track agents moved to minimize the distance to the vehicle. Different signal characteristics made it possible to distinguish between different types of vehicles.

It was concluded that distributed sensor systems are useful for treatment of rather complicated tasks and that the development in data processing is an important tool for the sensor community

Fredrik Gustavsson, Linköping University, Sweden

“Signal processing and information fusion for distributed sensors”.

In this lecture, more application areas concerning distributed sensor systems were described as well as the mathematics behind.

The principle of sensor fusion was described. It is based on basic principles of spatial and temporal averaging when all sensors measure the same parameter, which is time-invariant. Inference of a parameter is made, measured directly or indirectly using one or many sensors at one or many time instants. For temporal and spatial correlation, a spatial and dynamic model was described. Examples described were based on data from both an electronic tongue and an electronic nose.

In one example, automotive navigation was described. Distributed wheel speed sensors and a gyroscope were used, and sensor fusion with map information was performed. In another example, ship/aircraft tracking was performed using distributed passive radar warning systems. Each radar warner measured only angle, they were non-synchronized, and a dynamic model was needed for the sensor fusion. A cellular phone positioning system was also described. Other examples included car positioning using a map and the wheel speed. The system could work both with and without GPS. The principles of Bayesian filtering were discussed, as well as Kalman vs. particle filters. For the electronic tongue, a challenge is to develop a dynamic model for the collected current information due to onset of voltage pulses. In an application of measurements of bacterial growth, a dynamic model was developed, resulting in a very large data reduction. It was concluded that sensor fusion is a versatile tool with many applications. Further, the Bayesian paradigm offers a general framework for sensor fusion and the particle filter is a modern alternative and extension to classical Kalman filter approaches.

General discussion

The area of distributed sensor system is a new concept that may be very important for both electronic noses and tongues. One interesting approach is the moving chemical sensor system. The key feature is, however, to find good applications. These may include moving robots to search for gas leaks or people. How many robots should be used? Should it be one complicated or many simple, but communicating? One approach is the gas distribution surveillance, based on chemical sensor and GPS. In chemical plants, distributed chemical sensors could be very valuable for leak detection. A three dimensional problem may arise, making the system complicated and expensive. Another suggestion was to place chemical sensors in tunnels for the identification of accidents, fire etc. Using simple camera surveillance appeared to be a better idea. There is a large need for stable sensors. A large distributed system offers the possibility to keep track of pollution in Europe. These systems would be stationary and placed both in cities and industrial areas. The surveillance systems used today are only placed at special localized places to get a general overview, and more sites are needed. A suggestion was raised that the EU commission probably would be positive to the idea. The discussion was concluded with that distributed sensor system is a very useful concept, especially when developing applications for chemical sensors.