EP1461590A1

EP1461590A1 - Measurement system for sensors

Info

Publication number: EP1461590A1
Application number: EP01988104A
Authority: EP
Inventors: Davide Fabrizio; Ingemar LUNDSTRÖM; Marcus Andersson; Martin Holmberg
Original assignee: Rome International University Srl
Current assignee: Rome International University Srl
Priority date: 2001-05-11
Filing date: 2001-12-21
Publication date: 2004-09-29
Also published as: SE0101657D0; WO2002093111A1; SE524708C2; SE0101657L

Abstract

Sensor device including at least one sensor and one self-oscillating chaotic electronic non linear dynamic circuit and one evaluation device coupled to the electronic non linear dynamic circuit. One component of the circuit can be trimmed to adjust a switching point for the response from the circuit. A small change in the response in the response of the sensor to a chemical or physical change will, due to chaotic behavior of the circuit, cause easily detectable changes in the read out.

Description

MEASUREMENT SYSTEM FOR SENSORS

DESCRIPTION

The inventors have made up the first system of measuring chemical and physical variables based on an array of suitable sensors incorporated in a non linear circuit or system.

The non linear circuit or system mentioned above may be self-oscillating and exhibit chaotic behavior.

The inventors have made up a method to read out an array of sensors that relies on considering the sensors as elements of dynamic electronic circuits that are operated in non regular regimes, preferrably on the edge-of-the-chaos.

The measurement procedure and the electronic circuits are designed such that each measure will drive the circuit in a specific regime.

These regimes are easily recognizable with standard methods because either simpler than general chaotic dynamic or chaotic on specific attractors.

Owing to this invention the measurement of chemical and physical variables is possible even in many cases in which preexisting methods used to fail.

Moreover, because of the unique structure of this invention, it is possible to avoid several problems that usually occurs wliile measuring chemical and physical variables. In fact, for many types of sensors the signal response is too small compared to the noise, so that it is not possible to have a correct measurement. In addiction, the time of reaction of those sensors is generally too long in order to allow the event detection and control of for instance a fabrication process. Thus, the detection of changes in the sensor response is generally slowly, exactly as the signal-to-noise ratio, while this invention has those improved.

TECHNICAL FIELD

One important problem for many types of sensors is that the signal response as a result of sensor changes is too small compared to the noise. Another problem frequently experienced for chemical sensors is that it may take considerable time for the sensor to react sufficiently to allow event detection and possibly consequent control of for instance a fabrication process. There thus exists a need for methods and devices that speed up the detection of changes in the sensor response and increase the signal-to-noise ratio.

The object of the invention is to solve the problems mentioned above. In accordance with the invention this is carried out by incorporating the sensor or sensors in a non-linear circuit or system that may be self-oscillating and exhibit chaotic behaviour.

This invention regards the measurement of physical and chemical variables through an array of suitable sensors. The invented method allows the measurands to be measured in real-time and both analog and discrete mode. It provides better measurement results in comparison with the same sensors used in standard read-out configuration. BACKGROUND OF THE INVENTION

Deterministic non-linear dynamic systems have been studied extensively only in the last few years but have nonetheless already found applications in real life. The basic characteristic of these systems is that they though apparently random for an external observer do have laws governing their evolution over time. The laws are in general recursive, which makes them very sensitive to the initial conditions. A small difference in the initial state of two identical systems may give after sufficiently long time a very large difference between the two systems. This is an indicator of the so-called chaotic regime.

Non-linear systems are often described and characterised by the so-called phase space. This is a graph with a state variable of the system on one axis and its delayed versions on the other axes (often only two axes are drawn). Even if only one variable is considered to build this space, a lot of information about the whole system can be obtained due to non-linearity that mixes information among the state variables. Since some variables often cannot be directly observed in a system, the analysis of phase space is therefore of utmost importance in the study of non-linear systems.

In phase space, an attractor consists of a number of measurement points that are encountered by the dynamics out of an initial transient. This operative definition can be implemented collecting states in a relatively short period of time. This time depends on the circuitry, but for the intended applications a time of the same order of the sensor reaction time (i.e. to get 10% of the final value) is considered sufficient for mapping the attractor. We call the time necessary to map the attractor the attractor mapping time (tam).

In the field of chemical and physical sensors, the sensor signals have so far mainly been used as individual parts in the system. In this invention, the sensors are used in a circuit that exhibit strongly non-linear dynamics and, under certain conditions, chaotic behaviour.

The aims of introducing a complex behaviour such as chaos into the field of physical and chemical sensing is:

1. To fabricate a system that more rapidly detects changes in the environment than would be possible using a single sensor or an a sensor array (called electronic nose in the gas sensing) with traditional pattern recognition techniques

2. To decrease the sensitivity of drift in the sensor system;

3. To improve the signal-to-noise ratio;

4. To avoid limitations of harmonic read-out techniques.

These ideas stem from the sensitivity to small changes of a chaotic system, together with a robust persistence of the attractor statistics with noise, and the co-variation of similar sensors for measurands changes, as we have utilised in previous works.

For fulfilling the first aim, the attractor of the circuit is studied in real time. If a change in the physical or chemical surroundings causes a change in the shape of the attractor, detection can be made within one or a few tam, i.e. much shorter than the normal response time (i.e. to get 90% of the final value) of a single sensor. The assumption we have to make for the second aim to hold is that the system is stable also when the sensors drift, provided that they drift in a similar manner. When this condition is met, the invention solves one of the most difficult problems for process monitoring in the industry, namely to detect slow changes in the process that may be difficult to separate from drift phenomena in the sensor array. The sensor system would then be stable (i.e. have qualitatively the same attractor) also when the sensors drift. It would, however, change if the sensor variations were due to slow changes in the environment. We can also include in the design the knowledge of how the attractor as the process parameters change over time.

The evaluation of the state of the circuit (i.e. the type of attractor) can be made quickly and simply via either digital or analogue computation. FFT is enough to completely describe the harmonic spectrum in the signal, and this is enough when periodic vs. broadband signal should be distinguished. A simple series of filters can also solve the problem still remaining in the analogue processing field.

There are several advantages with this configuration compared to operating the sensor in the traditional way and/or using other types of use and evaluation of non-linear dynamic systems:

• The circuitry is very simple;

• Ordinary sensors can be used without modification;

• The response time is very short (order of seconds to get a stable state for chemical sensors, that are often very slower);

• Evaluation of the state of the circuit is simple and quick;

• The dynamic system "probes" the sensor response many times, in a broadband mode, in many different electric regimes and thus obtains a better signal-to-noise ratio.

DISCLOSURE OF INVENTION

The sensor array of interest in a physical and chemical measurement may consist either of a single sensor, if highly selective to a single intended measurand, or of many sensors. This is required when either the measurands are many or the sensors have partially overlapping selectivities. Involved sensors are those with conversion of the measurand into electrical properties. A method to read out an array of sensors used for measurement of physical and chemical variables has been invented. The method relies on considering the sensors as elements of dynamic electronic circuits that are operated in non-regular regimes, preferrably on the edge-of-the-chaos. The measurement procedure and the electronic circuits are designed such that each measure will drive the circuit in a specific regime. These regimes are easily recognizable with standard methods because either simpler than general chaotic dynamics or chaotic on specific attractors. Comparing with other methods involving the same sensors, the rate, the stability and the signal-. to-noise ratio of the measurement are strongly improved. In many cases in which preexisting methods fail, such as sensors affected by a strong drift, this invention makes the measurement possible.

BRIEF DESCRIPTION OF DRAWINGS

These, and other features and advantages of the present invention will be better understood by reading the following detailed description, taken together with the drawings wherein: FIG. 1 is a block diagram of the basic configuration of the measurement system, according to the present invention, with one sensor and a Non Linear Dynamic Circuit (NLDC). The

NLDC is schematically represented as a two-ports block, one used to connect the sensor, the other to connect a load, whose equivalent resistance is marked as RL. An Analogue to Digital converter provides for information on the circuit regime to an analyser that may be analogue as well as digital.

FIG. 2 is a block diagram of a multi-sensor system, composed of an NLDC connected to two sensors. More sensors are allowed as well.

FIG. 3 is a block diagram of a multi-sensor system, each sensor being connected to each own

NLDC. The outputs from the NLDC's are then connected to one or several other NLDC's with a weight g, and thereby influence their dynamic behaviour.

FIG. 4 shows a multi-sensor system where the sensors influence only the comiectivity strength g between two NLDC's.

FIG. 5 shows a generalisation of the schemes shown in FIG. 3 and 4 with an interconnected network of several NLDC's (shown as circular nodes) and several sensors included in both the interconnection weights and the nodes.

FIG. 6 shows an example of synchronization diagram with the locus of the (single) eigenvalue, for two Rδssler-like circuits coupled as in FIG. 4, with two MOS chemical sensors affecting the connections. The measurand, that is a gas concentration x, moves the eigenvalue γ along the arrow representing the locus.

FIG. 7 Example of a bifurcation diagram referred to a circuit like the one shown in FIG. 1.

The x-axis represents the measurand (e.g. the concentration of a certain gas in air), while the y-axis shows the dynamic pattern of the NLDC as accumulation points in a Poincare section.

This map shows an edge-of-the chaos regime at the lowest value of the measurand and regular periodic regime above a certain minimum detection limit (mdl). At certain values of the measurand (so-called bifurcation points) the period decreases of one half. Pi marks the range in which there is a period of i length.

FIG. 8 The Rδssler-attractor circuit used in a chemical sensing example.

FIG. 9 The chaotic attractor obtained for zero hydrogen concentration. The attractor was forced to be chaotic at zero concentration by varying the voltage V. The graph was made by plotting the voltages at X and Y in Fig. 8 on the two axes.

Fig. 10 The limit cycles (period four) obtained at 156 ppm hydrogen concentration.

Fig. 11 The limit cycles (period two) obtained at 312 ppm hydrogen concentration.

Fig. 12 Current-voltage characteristics of the MOSFET gas sensor used in the experiments.

The curves shown are for 0, 156, and 312 ppm hydrogen, respectively.

BEST MODE FOR CARRYING OUT THE INVENTION

Example of chemical sensing

Measurements on a non-linear dynamic circuit with a diode-coupled MOSFET gas sensor have been made using the configuration shown in Fig. 8 (implementation of the scheme in Fig. 1). The circuit used is a standard circuit for obtaining a Rossler-attractor under certain parameter choices, but with the modification of including a sensor instead of an ordinary diode at one position. The voltage V in the figure can be varied and thus obtaining different dynamical behaviour. The behaviour also changes when the characteristics of the other components change (in this case the variations in the MOSFET sensor are the only ones of interest). The voltage at the different points X,Y, and Z can be monitored in order to study the behaviour of the circuit. In Fig. 9-11, the attractors in the XY plane for different hydrogen concentrations are shown. The voltage V was tuned in order to find the edge of a chaotic attractor at zero hydrogen concentration (see Fig. 9). The hydrogen concentration was then changed, and the characteristics of the MOSFET sensor subsequently changed, giving rise to a change in the attractor, which became non-chaotic for the concentrations under study (156 and 312 ppm of hydrogen). The pattern changed into limit cycles of period 4 and 2, respectively. This pattern was obtained and became stable in the order of 5-10 seconds (attractor mapping time), compared to several minutes for these sensors to obtain a stable value (sensor response time) when measured using ordinary methods. These patterns were also analysed in an FFT, where the limit cycles were seen as single peaks, while the chaotic state was seen as a signal with continuos power spectrum over a frequency range.

In order to understand the size of the change of the electrical characteristics of the MOSFET sensor when used in these measurements, a measurement of the current- voltage characteristics was also made (see Fig. 12). Under operation, the voltage over the diode varies between -0.4 and +2.7 V. At these voltages, there is a very small shift in the characteristics, but as seen in Figs. 9-11 this small difference is very easy to detect thanks to the invention. Other changes, even dramatic, occur in the circuit if the operating voltage over the MOSFET sensor is increased.

Theoretical framework of the read-out technique and the design approach

The preceding example of embodiment is a straightforward implementation of the scheme in Fig. l. This arrangement requires a design of both the NLDC and the involved sensors in order to have the measurand as a parameter that provokes bifurcation in the system dynamics. An additional constrain is that the regimes after bifurcation should be well distinct. A straightforward design strategy is that of using the well known property of the edge-of-me- chaos regimes, that is they consist of a collection of unstable periodic regimes. First, let us choose a basic circuit in which the scheme in Fig. 1 is on the edge-of-the-chaos with the measurand under the intended minimum detection level (mdl). Then thanks to the prolongation theory the parameters of both NLDC and sensors can be optimised in order to have a bifurcation cascade as the measurand increases over the mdl. Fig. 7 shows the result of such a design with a circuit very similar to the one in Fig. 8. To be remarked that also bifurcation of non-regular regimes in other non-regular regimes can be designed and used according to this invention.

Fig. 2 introduces two sensors whose states affect the dynamics. Here the design should consider exactly the same procedure if the intended measurand is only one. In this case the system is redundant for reasons of precision and noise rejection. In case the measurands are two, the designer will design a double parameter bifurcation diagram with bifurcations placed in convenient locations of the measurand plane.

Information extraction in schemes shown in Fig. 1 and 2 may be discrete or continuous. • Discrete means to recognise the order of the bifurcation we are near to and then to estimate a range of the measurand. Taking as an example the case of the bifurcation diagram shown in Fig. 7, this recognition means to know which interval among P16, P8, P4, P2, PI, P0 the measurand belongs to. These intervals are referred to attractor limit cycles of period 16 (or more), 8, 4, 2, 1, 0 (fixed point), respectively. The period of the actual cycle can be easily recognised using a FFT analysis (so we need the A/D converter shown in Fig. l, 2), or a series of narrow-band filters (in this case the analyser is analogue circuitry and does not require the A/D converter), or many others straightforward techniques. Discrete is superior to any other existing readout strategy, being very simple and constrained only by the resolution of the measure (i.e. the length of the intervals). • Continuous means to estimate the parameter value, that is our measurand, from computation instead of a simple recognition. Continuous requires first the discrete estimation to be done, that is to identify the interval; after a refinement takes place measuring the distance between the actual attractor and the one at the next bifurcation (upper bound of the interval). This distance is properly defined according to the bifurcation type and the dynamics type. In case of limit cycles the simplest distance is the ratio of power of the forthcoming sub-harmonic (e.g. period 4) and of the fundamental harmonic (period 1). There is a defined upper bound of the ratio, which refers to the presence of the bifurcation point (see Fig. 7). A calibration curve can relate the actual ratio to the measurand value in any of the intervals. This approach is much more effective than trying to estimate the standard dynamics parameters, such as generalised Lyapunov exponents and generalised attractor dimensions, as proposed by other authors. The reason is that the double mapping from attractor reconstruction to dynamics parameters then to measurands, is imprecise because difficult to be calibrated and prone to noise and ageing. On the contrary, the design and use of bifurcations is an engineering approach, controllable, precise and repeatable.

Schemes in Fig. 3 and Fig. 4 may be treated in exactly the same way as schemes in Fig. 1 and 2. Nonetheless, there is another theoretical approach that is part of this invention, the synchronisation of NLDCs' dynamics. Here the information extraction is based on the analysis of corresponding state variables in the various coupled NLDCs (say X with X, Y with Y, etc.). These variables may synchronise, i.e. their difference is very small in comparison with the signals absolute value. If so it is well known from the synchronisation theory by TL Carrol and LM Pecora that the connection weights have specific properties. This theory considers the eigenvalues of the matrix of connection weights and proves that if they all are inside a certain region of the complex plane, the synchronisation occurs in all the coupled NLDCs. This region, called stability region, is also proven to be dependent only on the NLDC design and not on the pattern and weights of the comiections. The diagram consisting of the complex plane, the stability region and the eigenvalues of the of the connection matrix is called synchronization diagram. Design of a read-out scheme according to the present invention means to design a proper synchronization diagram together with its changes as measurands change.

Considering two NLDCs, we can design the synchronisation diagram such that if there is synchronisation we know that the measurands are in a certain range (e.g under the intended mdl), otherwise they are out of the range. This is fairly similar to the usage of bifurcation in the example embodiment above discussed. A simplification in the design occurs when the scheme in Fig. 4 is employed. Here the measurand changes can only move the eigenvalues on the plane but the stability region remains the same. The simplest approach is first to choose the NLDC scheme in order to have a comfortable stability region, that is closed, convex, and sharp boundaries. Then to optimise the sensors and the connections in order to design the locus of the eigenvalues as the measurands change. Fig. 6 shows the case of two Rossler-like circuits, with two MOS chemical sensors affecting the connections as in scheme 4. The measurand, that is a gas concentration, moves the eigenvalue along the arrow representing the locus. In this arrangement the occurrence of synchronization means that the concentration is below a value xo.

Further, if there are more than two NLDCs, we may have all of them synchronised, or only sub-clusters synchronised (this called & pattern of synchronisation). A simulation-based design can be done in order to define many suitable ranges of. measurands, each one occurring in coincidence with a certain synchronisation pattern.

As long as the synchronization technique is preferred, the read-out technique can be discrete or continuous. Discrete means simply to recognise the synchronisation pattern and to associate the ranges of the measurands. Continuous means first discrete estimation, and after a refinement, based on measuring the distance between the actual attractor of the whole network and the one at the neighbouring (according to a suitable distance and induced topology) synchronisation pattern.

As a final remark, the detection of synchronisation (when no finite delay is involved) is surely the simplest electronic operation to do: it consists only of a comparator that turns on output high when the inputs' difference is less than a fixed threshold.

INDUTRIAL APPLICABILITY

Measurement of physical and chemical variables is something that necessarily occurs in any industrial activities of every field, which incur in technical processes, especially in fabrications ones. Those processes are quite always based on detecting environment changes.

This invention regards that measurement mentioned above, through an array of suitable sensors. In fact, the invented method allows the measurands to be measured in real-time and both analog and discrete mode. It provides better measurement results in comparison with die same sensors used in standard read-out configuration.

Claims

CLAIMSAlthough the invention has been described with reference to a preferred embodiment, other embodiments can achieve the same results. Variations and modifications of the present invention will be obvious to those skilled in the art, and the claims below are intended to cover all such modifications and equivalents. What is claimed is:

1. Sensor device characterised by including at least one sensor and an electronic non-linear dynamic circuit and an evaluation device coupled to the electronic non-linear dynamic circuit.

2. Sensor device according to claim 1, characterised by the circuit being a self-oscillating circuit.

3. Sensor device according to claim 1 or 2, characterised by that at least one component of the circuit is changeable to adjust a switching point for the response from the circuit.

4. Sensor device according to any of the preceding claims, characterised by that more than one component in the non-linear dynamic circuit is constituted by a sensor

5. Sensor device according to any of the preceding claims, characterised by the connection of two or more circuits each with one ore several components constituted by sensors.

6. Sensor device according to any of the preceding claims, characterised by that the nonlinear circuit contain changeable components that can be used to trim the switching points of the device.

7. Sensor device according to any of the preceding claims characterised by that one or several circuits being chaotic.

8. Sensor device according to any of the preceding claims characterised by the sensor being a chemical sensor.