Title of Invention	METHOD FOR PROCESSING THE SIGNALS OF A DANGER DETECTOR AND DANGER DETECTOR HAVING MEANS FOR PERFORMING THE METHOD
Abstract	The signals from an alarm, comprising at least one sensor (2, 3, 4), for monitoring characteristic hazard values and an analytical electronic unit (1), connected to the at least one sensor (2, 3, 4), are compared with pre-set parameters. Furthermore, the signals are analysed for repeated or regular occurrence and repeated, or regularly occurring alarm signals are classed as error signals. The classification of signals as error signals gives rise to a corresponding adjustment of the parameter. When an error signal arises, before the parameter is adjusted, the validity of the signal analysis for the at least one sensor (2, 3, 4) is checked and the parameter adjustment is carried out, depending upon the result of said validity check. An alarm with the means for carrying out said method comprises at least one sensor (2, 3, 4), for a characteristic hazard value and an electronic analysis unit (1), containing a microprocessor (6), for the evaluation and analysis of the signal from the at least one sensor (2, 3, 4). The microprocessor (6) has a software programme with an adaptive algorithm based on multiple solutions for the analysis of the signals from the at least one sensor (2, 3, 4).

Title of Invention

METHOD FOR PROCESSING THE SIGNALS OF A DANGER DETECTOR AND DANGER DETECTOR HAVING MEANS FOR PERFORMING THE METHOD

Abstract

The signals from an alarm, comprising at least one sensor (2, 3, 4), for monitoring characteristic hazard values and an analytical electronic unit (1), connected to the at least one sensor (2, 3, 4), are compared with pre-set parameters. Furthermore, the signals are analysed for repeated or regular occurrence and repeated, or regularly occurring alarm signals are classed as error signals. The classification of signals as error signals gives rise to a corresponding adjustment of the parameter. When an error signal arises, before the parameter is adjusted, the validity of the signal analysis for the at least one sensor (2, 3, 4) is checked and the parameter adjustment is carried out, depending upon the result of said validity check. An alarm with the means for carrying out said method comprises at least one sensor (2, 3, 4), for a characteristic hazard value and an electronic analysis unit (1), containing a microprocessor (6), for the evaluation and analysis of the signal from the at least one sensor (2, 3, 4). The microprocessor (6) has a software programme with an adaptive algorithm based on multiple solutions for the analysis of the signals from the at least one sensor (2, 3, 4).

Full Text	Description The present invention relates to a method for processing the signals of a danger detector that las at least one sensor for monitoring danger parameters and an electronic evaluation system :hat is assigned to the at least one sensor, the danger parameters being monitored by :omparing the signals of the at least one sensor with specified parameters. The danger detector nay be for example a smoke detector, a flame detector, a passive infrared detector, a microwave detector, a dual detector (passive infrared sensor + microwave sensor) or a noise detector. Modern danger detectors have achieved a sensitivity with regard to the detection of danger parameters that is such that the main problem is no longer to detect a danger parameter as early as possible, but to distinguish interference signals from true danger signals reliably and thereby to avoid false alarms. Danger signals and interference signals are distinguished in th connection substantially by using a plurality of different sensors and correlating their signals or by analysing various features of the signals of a single sensor and/or by appropriate signal processing, in which connection a substantial improvement in interference immunity has already been achieved recently by using fuzzy logic. Fuzzy logic is generally known. With regard to the evaluation of the signals of danger detectors, it is to be emphasized that signal values are allocated to fuzzy sets in accordance with a membership function, the value of the membership function, or the degree of membership of a fuzzy set, being between 0 and 1. Important in this connection is the fact that the membership functions can be normalized, i.e. the sum of all the values of the membership function is equal to one, as a result of which the fuzzy logic evaluation permits an unambiguous interpretation of the signals. The object of the present invention is now to specify a method of the type mentioned at the outset for processing the signals of a danger detector that is further improved with regard to insensitivity to interference and interference immunity. The method according to the invention is characterized in that the signals of the at least one sensor are analysed on the basis of whether they occur increasingly frequently or regularly and in that signals occurring increasingly frequently or regularly are classified as interference signals. A first preferred development of the method according to the invention is characterized in that the classification of signals as interference signals triggers an appropriate adjustment of the parameters. The method according to the invention is based on the novel insight that, for example, a fire detector never "sees" more than a few real fires between two inspections or two power failures and that signals occurring increasingly frequently or regularly indicate the presence of sources I of interference. The interference signals due to the interference sources are recognized as such J and the detector parameters are adjusted accordingly. In this way, the detectors operated by the method according to the invention are capable of learning and are better able to distinguish between true danger signals and interference signals. A second preferred development of the method according to the invention is characterized in that if interference signals occur, the validity of the result of the analysis of the signals of the at least one sensor is checked prior to the adjustment of the parameters and in that the parameters are adjusted as a function of the result of this validity test. A third preferred development is characterized in that the validity is tested by methods based on multiple resolution. A fourth preferred development of the method according to the invention is characterized in that wavelets, preferably "biorthogonal" or "second generation" wavelets or "lifting schemes" are used for the validity test. The wavelet transformation is a transformation or imaging of a signal of the time domain into the frequency domain (in this connection, see, for example, "The Fast Wavelet-Transform" by Mac A. Cody in Dr. Dobb's Journal, April 1992); it is therefore basically similar to the Fourier transformation or fast Fourier transformation. However, it differs from the latter in the basic function of the transformation by which the signal is developed. In a Fourier transformation, a sine function and cosine function are used that are sharply localized in the frequency domain and indefinite in the time domain. In a wavelet transformation, a so-called wavelet or wave packet is used. Of the latter, there are various types, such as, for example, a Gauss, spline or hair wavelet that can each be displaced as desired in the time domain and expanded or compressed in the frequency domain by two parameters. Recently, novel wavelet methods have been disclosed that are often described as "second generation". Such wavelets are constructed using the so-called "lifting schemes" (Sweldens). This results in a series of approximations to the original signal, each of which has a coarser resolution than the preceding one. The number of operations necessary for the transformation is always proportional to the length of the original signal, whereas this number is disproportionate with respect to the signal length in the case of the Fourier transformation. The fast wavelet transformation can also be carried out inversely by restoring the original signal from the approximated values and coefficients for the reconstruction. The algorithm for resolving and reconstructing the signal and a table of resolving and reconstruction coefficients are given on the basis of an example for a spline wavelet in "An Introduction to Wavelets" by Charles K. Chui (Academic Press, San Diego, 1992). On this topic see also "A Wavelet Tour of Signal Processing" by S. Mallat (Academic Press, 1998). A further preferred development of the method according to the invention is characterized in that the expected values for the approximation coefficients or the approximation coefficients and detailed coefficients of the wavelets are determined and compared at different resolutions. Preferably, the said coefficients are determined in an estimator or by means of a neuronal network. The invention furthermore relates to a danger detector having means for carrying out the said method, having at least one sensor for a danger parameter and having an electronic evaluation system, comprising a microprocessor, for evaluating and analysing the signals of the at least one sensor. The danger detector according to the invention is characterized in that the microprocessor comprises a software program having a learning algorithm, based on multiple resolution, for analysing the signals of the at least one sensor. A first preferred embodiment of the danger detector according to the invention is characterized in that, on the one hand, the said sensor signals are analysed by the learning algorithm for their repeated or regular occurrence and, on the other hand, a validity test is carried out on the result, and in that the learning algorithm for the validity test uses wavelets, preferably "biorthogonal" or "second generation" wavelets. A second preferred embodiment of the danger detector according to the invention is characterized in that the learning algorithm uses neuro-fuzzy methods. A third preferred embodiment of the danger detector according to the invention is characterized in that the learning algorithm comprises the two equations in which cpmn denotes wavelet scaling functions, cmn denotes approximation coefficients and yk denotes the kth input point of the neuronal network and n is the dual function of cpmin (for definition of dual function see S. Mallat). The invention is explained in greater detail below with reference to exemplary embodiments and the drawings; in the drawings: Figure 1 shows a function explanation diagram, Figure 2 shows a block diagram of a danger detector equipped with means for carrying out the method according to the invention, Figures 3a, 3b show two variants of a detail of the danger detector of Figure 2; and Figure 4 shows a further variant of a detail of the danger detector of Figure 3. The method according to the invention processes the signals of a danger detector in such a way that typical interference signals are detected and characterized. If fire detectors are predominantly mentioned in a present description, this does not mean that the method according to the invention is restricted to fire detectors. On the contrary, the method is suitable for danger detectors of all kinds, in particular also for intruder detectors and movement detectors. The interference signals mentioned are analysed by a simple and reliable method. An important feature of this method is that the interference signals are not only detected and characterized, but that the result of the analysis is checked. Wavelet theory and multiple resolution analysis (multiresolution analysis) are used. Depending on the result of the check, the detector parameters or the algorithms are adjusted. That means that, for example, the sensitivity is reduced or that certain automatic switchings between different sets of parameters are interlocked. The latter may be explained by means of an example: EuropejanJ^fe^ 975.8 describes a fire detector that has an optical sensor for scattered light, a temperature sensor and a fire gas sensor. The electronic evaluation system of the detector comprises a fuzzy controller in which the signals of the individual sensors are combined and the particular type of fire is diagnosed. A special application-specific algorithm is provided for each type of fire and can be selected on the basis of the diagnosis. In addition, the detector comprises various sets of parameters for personnel protection and property protection between which on-line switching takes place under normal circumstances. If interference signals are now diagnosed in the case of the temperature sensor and/or in the case of the fire gas sensor, the switching between these sets of parameters is interlocked. If fuzzy logic is used, one of the problems to be solved is to translate the knowledge stored in a database into linguistically interpretable fuzzy rules. Neuro-fuzzy methods developed for this purpose have not been convincing because they partly yield only fuzzy rules that are very difficult to interpret. On the other hand, so-called multiple resolution procedures offer a possibility of obtaining interpretable fuzzy rules. Their idea is to use a dictionary of membership functions that form a multiple resolution and to determine which are suitable membership functions for describing a control surface. Figure 1 shows a diagram of such a multiple resolution. Row a shows the characteristic of a signal whose amplitude varies in the ranges small, medium and large. Correspondingly, row b shows the membership functions d "fairly small", c2 "medium" and c3 "rather large". These membership functions form a multiple resolution, which means that each membership function can be resolved into a sum of membership functions of a higher resolution level. This results in the membership functions c5 "very small", c6 "small to very small", c7 "very medium", c8 "large to very large" and c9 "very large" entered in row c. In accordance with row d, the triangular spline function c2 can therefore be converted, for example into the sum of the translated triangle functions of the higher level of row c. In the Tagaki-Sugeno model, the fuzzy rules are expressed by the equation (1) Here Aj's are linguistic expressions, x is the linguistic input variable and y is the output variable. The value of the linguistic input variables can be sharp or fuzzy. If, for example, xs is a linguistic variable for temperature, the value x may be a sharp number such as "30(°C)" or a fuzzy quantity such as "approximately 25(°C)", "approximately 25" being itself a fuzzy set. For a sharp input value, the output value of the fuzzy system is given by: (2) where the degree of fulfilment fy is given by the expression fy = \|iAi (x) in which \|aAi(x) denotes the membership function of the linguistic term Aj. In many applications, a linear function is taken: f (x) = aTi ■ x + bj. If a constant b; is taken to describe the sharp output value y, the system becomes: (3) If spline functions Nkare taken, for example as membership function jjAi(x) = Nk[2m(x-n)], then the system of equation (3) is equivalent to (4) In this special case, the output y is a linear sum of translated and expanded spline functions. And that means that, given equation (4), the Tagaki-Sugeno model is equivalent to a multiple resolution spline model. And it follows from this that wavelet procedures can be applied. Figure 2 shows a block diagram of a danger detector equipped with a neuro-fuzzy learning algorithm. The detector denoted by the reference symbol M is, for example, a fire detector and has three sensors 2 to 4 for fire parameters. For example, an optical sensor 2 is provided for scattered light measurement or transmitted light measurement, a temperature sensor 3 and a fire gas sensor, for example a CO sensor, 4, are provided. The output signals of the sensors 2 to 4 are fed to a processing stage 1 that has suitable means for processing the signals, such as, for example, amplifiers, and are passed from the latter to a microprocessor or microcontroller denoted below as \|^P 6. In the \|aP 6, the sensor signals are compared both with one another and also individually with certain sets of parameters for the individual fire parameters. Of course, the number of sensors is not limited to three. Thus, only a single sensor may also be provided, and in this case, various characteristics, for example the signal gradient or the signal fluctuation, are extracted from the signal of the one sensor and investigated. Incorporated in the ^P 6 are a neuro-fuzzy network 7 software and a validity test (validation) 8. If the signal resulting from the neuro-fuzzy network 7 is regarded as an alarm signal, an appropriate alarm signal is fed to an alarm-emitting device 9 or to an alarm centre. If the validation 8 reveals that interference signals occur repeatedly or regularly, the sets of parameters stored in the jaP 6 are correspondingly corrected. The neuro-fuzzy network 7 is a series of neuronal networks which use the symmetrical scaling functions (pm,n(x) = (pm,n,(x) = where yk(x) is the kth input point and cpmn (x) is the dual function of q>min(x). The two equations (5) and (6) form the main algorithm of the neuro-fuzzy network. In each iteration step, the values of the various neuronal networks are checked crosswise (validated), for which purpose a characteristic of the wavelet resolution is used, namely the one that the approximation coefficient c mn of a level m can be obtained from the approximation coefficients and wavelet coefficients of the level m-1 using the reconstruction algorithm or resolving algorithm. In a preferred version, function. In a second version, (pm,n(x) is a spline function and #>mi„0) is the dual function of cp mn(x). In a third version, (f>m^{x) = cpm,n(x), where cpmin(x) is the hair function. In these cases, it is possible to implement the learning algorithm in a simple microprocessor. Figures 3a and 3b show two variants of a neuro-fuzzy network 7 and the associated validation stage 8. In the example of Figure 3a, the input signal is approximated in various resolution stages as the weighted sum of wavelets y¥m%n and scaling functions 9mn having a given resolution. The validation stage 8 compares the approximation coefficients cmn with the approximation coefficients and detailed coefficients of the wavelets at the level of the next lower resolution stage. Wavelet reconstruction filter coefficients are denoted by p and q. In the example of Figure 3b, the input signal is approximated in various resolution stages as a weighted sum of scaling functions cpmn having a given resolution. The validation stage 8 compares the approximation coefficients cmn with the approximation coefficients at the next-deeper resolution stage. Wavelet low-pass resolving coefficients are denoted by g. The said coefficients can be determined in an estimator of the type shown in Figure 4 instead of in a neuro-fuzzy network 7. Said estimator is a so-called multiple resolution spline estimator that uses dual spline estimators based on the functions (pmJXx)to estimate the coefficients cmn in the equation fm(x) = cmin- (pm,n(x). Wavelet spline estimators are used for adaptively determining the appropriate resolution for locally describing a basic hypersurface in an on-line learning process. A known estimator is the Nadaraya-Watson estimator with which the equation of the hypersurface f(x) is estimated using the following expression: (6) Nadaraya-Watson estimators have two interesting characteristics - they are estimators of the local mean quadratic deviation and it can be shown that they are so-called Bayes estimators of (Xk.Yk) in the case of a random design, where (xk,yk) are iid copies of a continuous random variable (X, Y). The spline functions Using the symmetry of ip(x), equation (6) for the dual spline function is equivalent to the use of an estimator centred at xn: (7) The expected value of the numerator in equation (7) is proportional to the approximation coefficients cmn. Equation (6) yields an estimate of cmn in fm(x) = E cmn ■ (8) In Figure 4, the available data (values) are denoted by a small square, their projection on dual spline functions by a small circle and the estimate on a regular grid by a small cross. To validate the coefficient cn, two conditions are necessary: (9) where the filter coefficients g correspond to the low-pass resolving coefficients for spline functions. In addition it is required that (10) so that divisions by very small values are prevented. The strength of this method is that the calculation of a coefficient cmn requires the storage of only two values, the numerator and the denominator in equation (7). The method is therefore well suited for on-line learning using a simple microprocessor having low storage capacity. The method can easily be adapted to density estimation by replacing equations (7) and (8) by the following equation: WE CLAIM : 1. Method for processing the signals of a detector unit that has at least one sensor (2, 3,4) for monitoring danger parameters and an electronic evaluation system (1) that is assigned to the at least one sensor (2, 3, 4) and in which the signals of the at least one sensor (2, 3, 4) are compared with specified detector parameters, characterized in that the signals of the at least one sensor (2, 3,4) are analysed on the basis of whether they occur increasingly frequently or regularly, in that signals occurring increasingly frequently or regularly are classified as interference signals, and in that the classification of signals as interference signals triggers an appropriate adjustment of the detector parameters. 2. Method according to claim 1, wherein if interference signals occur, the validity of the result of the analysis of the signals of the at least one sensor (2,3,4) is checked prior to the adjustment of the detector parameters and in that the parameters are adjusted as a function of the result of this validity test. 3. Method according to claim 2, wherein the validity is tested by methods based on multiple resolution. 4. Method according to claim 3, wherein wavelets, preferably "biorthogonal" or "second generation" wavelets or "lifting schemes" are used for the validity test. 5. Method according to claim 4, wherein the expected values for the approximation coefficients or the approximation coefficients and detailed coefficients of the wavelets are determined and compared at different resolutions. 6. Method according to claim 5, wherein the said coefficients are determined i an estimator or by means of a neuronal network, 7. A danger detector having means for carrying out the method according to claims 1 to 6.

Full Text

Description
The present invention relates to a method for processing the signals of a danger detector that las at least one sensor for monitoring danger parameters and an electronic evaluation system :hat is assigned to the at least one sensor, the danger parameters being monitored by :omparing the signals of the at least one sensor with specified parameters. The danger detector nay be for example a smoke detector, a flame detector, a passive infrared detector, a microwave detector, a dual detector (passive infrared sensor + microwave sensor) or a noise detector.
Modern danger detectors have achieved a sensitivity with regard to the detection of danger parameters that is such that the main problem is no longer to detect a danger parameter as early as possible, but to distinguish interference signals from true danger signals reliably and thereby to avoid false alarms. Danger signals and interference signals are distinguished in th connection substantially by using a plurality of different sensors and correlating their signals or by analysing various features of the signals of a single sensor and/or by appropriate signal processing, in which connection a substantial improvement in interference immunity has already been achieved recently by using fuzzy logic.
Fuzzy logic is generally known. With regard to the evaluation of the signals of danger detectors, it is to be emphasized that signal values are allocated to fuzzy sets in accordance with a membership function, the value of the membership function, or the degree of membership of a fuzzy set, being between 0 and 1. Important in this connection is the fact that the membership functions can be normalized, i.e. the sum of all the values of the membership function is equal to one, as a result of which the fuzzy logic evaluation permits an unambiguous interpretation of the signals.
The object of the present invention is now to specify a method of the type mentioned at the outset for processing the signals of a danger detector that is further improved with regard to insensitivity to interference and interference immunity.
The method according to the invention is characterized in that the signals of the at least one sensor are analysed on the basis of whether they occur increasingly frequently or regularly and in that signals occurring increasingly frequently or regularly are classified as interference signals.

A first preferred development of the method according to the invention is characterized in that the classification of signals as interference signals triggers an appropriate adjustment of the parameters.
The method according to the invention is based on the novel insight that, for example, a fire detector never "sees" more than a few real fires between two inspections or two power failures and that signals occurring increasingly frequently or regularly indicate the presence of sources I of interference. The interference signals due to the interference sources are recognized as such J and the detector parameters are adjusted accordingly. In this way, the detectors operated by the method according to the invention are capable of learning and are better able to distinguish between true danger signals and interference signals.
A second preferred development of the method according to the invention is characterized in that if interference signals occur, the validity of the result of the analysis of the signals of the at least one sensor is checked prior to the adjustment of the parameters and in that the parameters are adjusted as a function of the result of this validity test.
A third preferred development is characterized in that the validity is tested by methods based on multiple resolution.
A fourth preferred development of the method according to the invention is characterized in that wavelets, preferably "biorthogonal" or "second generation" wavelets or "lifting schemes" are used for the validity test.
The wavelet transformation is a transformation or imaging of a signal of the time domain into the frequency domain (in this connection, see, for example, "The Fast Wavelet-Transform" by Mac A. Cody in Dr. Dobb's Journal, April 1992); it is therefore basically similar to the Fourier transformation or fast Fourier transformation. However, it differs from the latter in the basic function of the transformation by which the signal is developed. In a Fourier transformation, a sine function and cosine function are used that are sharply localized in the frequency domain and indefinite in the time domain. In a wavelet transformation, a so-called wavelet or wave packet is used. Of the latter, there are various types, such as, for example, a Gauss, spline or hair wavelet that can each be displaced as desired in the time domain and expanded or compressed in the frequency domain by two parameters. Recently, novel wavelet methods have been disclosed that are often described as "second generation". Such wavelets are constructed using the so-called "lifting schemes" (Sweldens).
This results in a series of approximations to the original signal, each of which has a coarser resolution than the preceding one. The number of operations necessary for the transformation is

always proportional to the length of the original signal, whereas this number is disproportionate with respect to the signal length in the case of the Fourier transformation. The fast wavelet transformation can also be carried out inversely by restoring the original signal from the approximated values and coefficients for the reconstruction. The algorithm for resolving and reconstructing the signal and a table of resolving and reconstruction coefficients are given on the basis of an example for a spline wavelet in "An Introduction to Wavelets" by Charles K. Chui (Academic Press, San Diego, 1992). On this topic see also "A Wavelet Tour of Signal Processing" by S. Mallat (Academic Press, 1998).
A further preferred development of the method according to the invention is characterized in that the expected values for the approximation coefficients or the approximation coefficients and detailed coefficients of the wavelets are determined and compared at different resolutions. Preferably, the said coefficients are determined in an estimator or by means of a neuronal network.
The invention furthermore relates to a danger detector having means for carrying out the said method, having at least one sensor for a danger parameter and having an electronic evaluation system, comprising a microprocessor, for evaluating and analysing the signals of the at least one sensor.
The danger detector according to the invention is characterized in that the microprocessor comprises a software program having a learning algorithm, based on multiple resolution, for analysing the signals of the at least one sensor.
A first preferred embodiment of the danger detector according to the invention is characterized in that, on the one hand, the said sensor signals are analysed by the learning algorithm for their repeated or regular occurrence and, on the other hand, a validity test is carried out on the result, and in that the learning algorithm for the validity test uses wavelets, preferably "biorthogonal" or "second generation" wavelets.
A second preferred embodiment of the danger detector according to the invention is characterized in that the learning algorithm uses neuro-fuzzy methods.
A third preferred embodiment of the danger detector according to the invention is characterized in that the learning algorithm comprises the two equations

in which cpmn denotes wavelet scaling functions, cmn denotes approximation coefficients and yk denotes the kth input point of the neuronal network and

n is the dual function of cpmin (for definition of dual function see S. Mallat).
The invention is explained in greater detail below with reference to exemplary embodiments and
the drawings; in the drawings:
Figure 1 shows a function explanation diagram,
Figure 2 shows a block diagram of a danger detector equipped with means for
carrying out the method according to the invention,
Figures 3a, 3b show two variants of a detail of the danger detector of Figure 2; and
Figure 4 shows a further variant of a detail of the danger detector of Figure 3.
The method according to the invention processes the signals of a danger detector in such a way that typical interference signals are detected and characterized. If fire detectors are predominantly mentioned in a present description, this does not mean that the method according to the invention is restricted to fire detectors. On the contrary, the method is suitable for danger detectors of all kinds, in particular also for intruder detectors and movement detectors.
The interference signals mentioned are analysed by a simple and reliable method. An important feature of this method is that the interference signals are not only detected and characterized, but that the result of the analysis is checked. Wavelet theory and multiple resolution analysis (multiresolution analysis) are used. Depending on the result of the check, the detector parameters or the algorithms are adjusted. That means that, for example, the sensitivity is reduced or that certain automatic switchings between different sets of parameters are interlocked.
The latter may be explained by means of an example: EuropejanJ^fe^ 975.8 describes a fire detector that has an optical sensor for scattered light, a temperature sensor and a fire gas sensor. The electronic evaluation system of the detector comprises a fuzzy controller in which the signals of the individual sensors are combined and the particular type of fire is diagnosed. A special application-specific algorithm is provided for each type of fire and can be selected on the basis of the diagnosis. In addition, the detector comprises various sets of parameters for personnel protection and property protection between which on-line switching takes place under normal circumstances. If interference signals are now diagnosed in the case of the temperature sensor and/or in the case of the fire gas sensor, the switching between these sets of parameters is interlocked.

If fuzzy logic is used, one of the problems to be solved is to translate the knowledge stored in a database into linguistically interpretable fuzzy rules. Neuro-fuzzy methods developed for this purpose have not been convincing because they partly yield only fuzzy rules that are very difficult to interpret. On the other hand, so-called multiple resolution procedures offer a possibility of obtaining interpretable fuzzy rules. Their idea is to use a dictionary of membership functions that form a multiple resolution and to determine which are suitable membership functions for describing a control surface.
Figure 1 shows a diagram of such a multiple resolution. Row a shows the characteristic of a signal whose amplitude varies in the ranges small, medium and large. Correspondingly, row b shows the membership functions d "fairly small", c2 "medium" and c3 "rather large". These membership functions form a multiple resolution, which means that each membership function can be resolved into a sum of membership functions of a higher resolution level. This results in the membership functions c5 "very small", c6 "small to very small", c7 "very medium", c8 "large to very large" and c9 "very large" entered in row c. In accordance with row d, the triangular spline function c2 can therefore be converted, for example into the sum of the translated triangle functions of the higher level of row c.
In the Tagaki-Sugeno model, the fuzzy rules are expressed by the equation
(1)
Here Aj's are linguistic expressions, x is the linguistic input variable and y is the output variable. The value of the linguistic input variables can be sharp or fuzzy. If, for example, xs is a linguistic
variable for temperature, the value x may be a sharp number such as "30(°C)" or a fuzzy
quantity such as "approximately 25(°C)", "approximately 25" being itself a fuzzy set.
For a sharp input value, the output value of the fuzzy system is given by:
(2)
where the degree of fulfilment fy is given by the expression fy = |iAi (x) in which |aAi(x) denotes the membership function of the linguistic term Aj. In many applications, a linear function is taken:
f (x) = aTi ■ x + bj. If a constant b; is taken to describe the sharp output value y, the system becomes:
(3)
If spline functions Nkare taken, for example as membership function jjAi(x) = Nk[2m(x-n)], then the system of equation (3) is equivalent to

(4)
In this special case, the output y is a linear sum of translated and expanded spline functions. And that means that, given equation (4), the Tagaki-Sugeno model is equivalent to a multiple resolution spline model. And it follows from this that wavelet procedures can be applied.
Figure 2 shows a block diagram of a danger detector equipped with a neuro-fuzzy learning algorithm. The detector denoted by the reference symbol M is, for example, a fire detector and has three sensors 2 to 4 for fire parameters. For example, an optical sensor 2 is provided for scattered light measurement or transmitted light measurement, a temperature sensor 3 and a fire gas sensor, for example a CO sensor, 4, are provided. The output signals of the sensors 2 to 4 are fed to a processing stage 1 that has suitable means for processing the signals, such as, for example, amplifiers, and are passed from the latter to a microprocessor or microcontroller
denoted below as |^P 6.
In the |aP 6, the sensor signals are compared both with one another and also individually with certain sets of parameters for the individual fire parameters. Of course, the number of sensors is not limited to three. Thus, only a single sensor may also be provided, and in this case, various characteristics, for example the signal gradient or the signal fluctuation, are extracted from the
signal of the one sensor and investigated. Incorporated in the ^P 6 are a neuro-fuzzy network 7 software and a validity test (validation) 8. If the signal resulting from the neuro-fuzzy network 7 is regarded as an alarm signal, an appropriate alarm signal is fed to an alarm-emitting device 9 or to an alarm centre. If the validation 8 reveals that interference signals occur repeatedly or regularly, the sets of parameters stored in the jaP 6 are correspondingly corrected.
The neuro-fuzzy network 7 is a series of neuronal networks which use the symmetrical scaling functions (pm,n(x) = (pm,n,(x) =

where yk(x) is the kth input point and cpmn (x) is the dual function of q>min(x). The two equations (5) and (6) form the main algorithm of the neuro-fuzzy network.
In each iteration step, the values of the various neuronal networks are checked crosswise (validated), for which purpose a characteristic of the wavelet resolution is used, namely the one that the approximation coefficient c mn of a level m can be obtained from the approximation coefficients and wavelet coefficients of the level m-1 using the reconstruction algorithm or resolving algorithm.
In a preferred version, function. In a second version, (pm,n(x) is a spline function and #>mi„0) is the dual function of cp
mn(x). In a third version, (f>m^{x) = cpm,n(x), where cpmin(x) is the hair function. In these cases, it is possible to implement the learning algorithm in a simple microprocessor.
Figures 3a and 3b show two variants of a neuro-fuzzy network 7 and the associated validation stage 8. In the example of Figure 3a, the input signal is approximated in various resolution
stages as the weighted sum of wavelets y¥m%n and scaling functions 9mn having a given
resolution. The validation stage 8 compares the approximation coefficients cmn with the approximation coefficients and detailed coefficients of the wavelets at the level of the next lower resolution stage. Wavelet reconstruction filter coefficients are denoted by p and q.
In the example of Figure 3b, the input signal is approximated in various resolution stages as a
weighted sum of scaling functions cpmn having a given resolution. The validation stage 8
compares the approximation coefficients cmn with the approximation coefficients at the next-deeper resolution stage. Wavelet low-pass resolving coefficients are denoted by g.
The said coefficients can be determined in an estimator of the type shown in Figure 4 instead of in a neuro-fuzzy network 7. Said estimator is a so-called multiple resolution spline estimator that
uses dual spline estimators based on the functions (pmJXx)to estimate the coefficients cmn in
the equation fm(x) = cmin- (pm,n(x). Wavelet spline estimators are used for adaptively determining the appropriate resolution for locally describing a basic hypersurface in an on-line learning process. A known estimator is the Nadaraya-Watson estimator with which the equation of the hypersurface f(x) is estimated using the following expression:
(6)

Nadaraya-Watson estimators have two interesting characteristics - they are estimators of the local mean quadratic deviation and it can be shown that they are so-called Bayes estimators of (Xk.Yk) in the case of a random design, where (xk,yk) are iid copies of a continuous random variable (X, Y).
The spline functions

Using the symmetry of ip(x), equation (6) for the dual spline function is equivalent to the use of an estimator centred at xn:
(7)
The expected value of the numerator in equation (7) is proportional to the approximation coefficients cmn. Equation (6) yields an estimate of cmn in fm(x) = E cmn ■ (8)
In Figure 4, the available data (values) are denoted by a small square, their projection on dual spline functions by a small circle and the estimate on a regular grid by a small cross.
To validate the coefficient cn, two conditions are necessary:
(9)
where the filter coefficients g correspond to the low-pass resolving coefficients for spline functions. In addition it is required that
(10)
so that divisions by very small values are prevented.
The strength of this method is that the calculation of a coefficient cmn requires the storage of only two values, the numerator and the denominator in equation (7). The method is therefore well suited for on-line learning using a simple microprocessor having low storage capacity.
The method can easily be adapted to density estimation by replacing equations (7) and (8) by the following equation:

WE CLAIM :
1. Method for processing the signals of a detector unit that has at least one sensor (2, 3,4) for monitoring danger parameters and an electronic evaluation system (1) that is assigned to the at least one sensor (2, 3, 4) and in which the signals of the at least one sensor (2, 3, 4) are compared with specified detector parameters, characterized in that the signals of the at least one sensor (2, 3,4) are analysed on the basis of whether they occur increasingly frequently or regularly, in that signals occurring increasingly frequently or regularly are classified as interference signals, and in that the classification of signals as interference signals triggers an appropriate adjustment of the detector parameters.
2. Method according to claim 1, wherein if interference signals occur, the validity of the result of the analysis of the signals of the at least one sensor (2,3,4) is checked prior to the adjustment of the detector parameters and in that the parameters are adjusted as a function of the result of this validity test.
3. Method according to claim 2, wherein the validity is tested by methods based on multiple resolution.
4. Method according to claim 3, wherein wavelets, preferably "biorthogonal" or "second generation" wavelets or "lifting schemes" are used for the validity test.
5. Method according to claim 4, wherein the expected values for the approximation coefficients or the approximation coefficients and detailed coefficients of the wavelets are determined and compared at different resolutions.

6. Method according to claim 5, wherein the said coefficients are determined i an estimator or by means of a neuronal network,
7. A danger detector having means for carrying out the method according to claims 1 to 6.

Documents:

in-pct-2001-1413-che-abstract.pdf

in-pct-2001-1413-che-claims .pdf

in-pct-2001-1413-che-correspondance others.pdf

in-pct-2001-1413-che-correspondance po.pdf

in-pct-2001-1413-che-description complete.pdf

in-pct-2001-1413-che-drawings.pdf

in-pct-2001-1413-che-form 1.pdf

in-pct-2001-1413-che-form 18.pdf

in-pct-2001-1413-che-form 26.pdf

in-pct-2001-1413-che-form 3.pdf

in-pct-2001-1413-che-form 5.pdf

in-pct-2001-1413-che-pct.pdf

« Previous Patent

Next Patent »

Patent Number

218946

Indian Patent Application Number

IN/PCT/2001/1413/CHE

PG Journal Number

23/2008

Publication Date

06-Jun-2008

Grant Date

16-Apr-2008

Date of Filing

12-Oct-2001

Name of Patentee

SIEMENS BUILDING TECHNOLOGIES AG

Applicant Address

Inventors:

#	Inventor's Name	Inventor's Address
1	MARE PIERRE TIIUILLARD

PCT International Classification Number

G08B 29/26

PCT International Application Number

PCT/CH01/00136

PCT International Filing date

2001-03-06

PCT Conventions:

#	PCT Application Number	Date of Convention	Priority Country
1	00105438.6	2000-03-15	EUROPEAN UNION