• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    A data fusion method capable of merging the measurements of a parameter, for example a flight parameter of an aircraft, taken by a plurality of sensors comprising a set of main sensors and sets of secondary sensors. The difference between each principal measurement and the secondary measurements (410) is determined and a consistency score is deduced therefrom (420). For each fault configuration of the main sensors, a first, conditional estimate of said parameter (430) is made and a weighting coefficient relative to this failure is calculated, taking into account the coherence scores of the main measurements (440). The parameter is then estimated by performing a combination of the conditional measures with the associated weighting coefficients (450).
    公开号:FR3013834A1
    申请号:FR1361754
    申请日:2013-11-28
    公开日:2015-05-29
    发明作者:Patrice Brot；Emmanuel Fall；Corinne Mailhes；Jean-Yves Tourneret
    申请人:Airbus Operations SAS；
    IPC主号:
    专利说明:

    [0001] TECHNICAL FIELD The present invention relates to the field of sensor data fusion, more particularly to estimating a flight parameter of an aircraft. STATE OF THE PRIOR ART An aircraft is equipped with a large number of sensors making it possible to measure its flight parameters (speed, attitude, position, altitude, etc.) and more generally its state at each instant. These flight parameters are then used by avionics systems, including the autopilot system, the Flight Control Computer Systems, the Flight Guidance System, systems among the more critical of the aircraft. Due to the criticality of these systems, the measured parameters must have high integrity and availability. Integrity means that the values of the parameters used by the avionics systems are not erroneous because of any failure. By availability, it is meant that the sensors providing these parameters must be sufficiently redundant to be able to permanently have a measurement of each parameter. If one sensor fails and can not provide measurements, another sensor takes over. In general, an avionics system receives measurements of the same parameter from several redundant sensors. When these measures differ, the system carries out a processing (data fusion) in order to estimate, by consolidation of said measurements, the parameter with the lowest risk of error. This treatment is usually an average or median of the measures in question.
    [0002] Fig. 1 illustrates a first example of processing measurements provided by redundant sensors. This figure shows the measurements of the same flight parameter, made respectively by three sensors, Ap A2, A3, as a function of time.
    [0003] The treatment here consists in calculating at each instant the median value of the measurements. Thus, in the illustrated example, the measurement a2 (t) of A2 is selected up to the time t1 and, beyond that, the measurement ai (t) of A. The diagram also shows a tolerance band of width 2A around the median. If one of the measurements comes out of the tolerance band (for example the measurement a2 (t) from time t2), it is considered that this measurement is erroneous and it is no longer taken into account in the estimation of the parameter in question. The corresponding sensor (here A2) is disabled for further processing. In general, this treatment can be applied when the number of redundant sensors is odd.
    [0004] When the number of sensors is even, or if the number of sensors is odd but a sensor has already been invalidated, the values of the measurements are simply averaged at each instant to obtain an estimate of the parameter in question. All sensors belonging to the same technology may be affected by a common fault (for example, presence of ice in the Pitot tubes, closed static pressure taps, frozen incidence probes, failure on the same electronic component, etc.). In this case, the aforementioned methods of treatment are not able to identify the erroneous sources. It is then advantageous to use additional sensors using one or more different technology (s). Thus, there are generally several sets of sensors for measuring the same parameter, the technologies of the different sets of sensors being chosen dissimilar. By dissimilar technologies, we mean that these technologies use different physical principles or different implementations.
    [0005] For example, it is possible to use a first set of sensors capable of measuring the speed of the aircraft from pressure probes (total pressure and static pressure), also called ADRs (Air Data Reference units), to a first estimator capable of to estimate the velocity as a function of the angle of incidence and the lift (using the lift equation), and a second estimator capable of estimating this velocity from the engine data. A first approach to estimate the parameter is to merge all the measurements taken by the sensors, dissimilar or not, according to the same principle as before. For example at a given moment, we will take the median or the average of the values measured by the various sensors. This first approach improves the robustness of the estimation of the parameter by freeing it from failures that may affect a particular technology. On the other hand, it can lead to a significant reduction in the precision of the estimate as illustrated by the example below.
    [0006] Fig. 2A represents the values of a flight parameter (here the speed of the aircraft) measured by a first set of three sensors (denoted A1, A2, k) using a first technology and a second set of two sensors (denoted B1 , B2) using a second technology dissimilar to the first. The measurements are respectively noted ai (t), a2 (t), a3 (t) for the first set of sensors and bi (t), b2 (t) for the second set of sensors. It is assumed that the measures ai (t), a2 (t), a3 (t) are significantly more accurate than the measures b1 (t), b2 (t). V (t) represents the actual speed of the aircraft. It is assumed that at time tf the sensors A1 and k of the first set are affected by the same failure. As can be seen in the figure, from the time tf, the measurements a1 (t), a2 (t) derive and deviate substantially from the real value of the parameter, V (t).
    [0007] Fig. 2B represents the estimate i (t) of the velocity obtained as the median of the measures ai (t), a2 (t), a3 (t), bi (t), b2 (t). We see that from the time tc, the calculation of the median is to select the measurement b2 (t) of the B2 sensor. However, this measurement is much less precise than the measurement a3 (t) of the A3 sensor, which is available and valid.
    [0008] It can be seen that the data fusion applied to all the measurements leads here to a suboptimal estimation precision. The purpose of the present invention is therefore to propose a data fusion method capable of merging the measurements of a parameter, for example a flight parameter of an aircraft, taken by a plurality of sensors, technologies and accuracies. different, to obtain an estimate of this parameter which is not only available and robust but which also has a better accuracy than in the prior art. DISCLOSURE OF THE INVENTION The present invention is defined by a method of merging measurements of a parameter, in particular of a flight parameter of an aircraft, from a first set of measurements, called principal, taken by a first set of sensors, called main and a plurality of second sets of measurements, called secondary, taken by second sets of sensors, called secondary, the main measurements having a degree of precision greater than the secondary measures, in which: a difference is calculated between each principal measure and said secondary measures; a consistency score of each principal measurement is determined with the secondary measurements by means of the difference thus obtained; for each possible failure configuration of the main sensors, a first, conditional estimate of said parameter is made from the main measurements taken by the sensors that are not in fault in said configuration; for each fault configuration, a weighting coefficient is determined from the coherence scores of the main measurements previously obtained; an estimate of said parameter is made by weighting the conditional estimates relating to the various failure configurations by the weighting coefficients corresponding to these configurations. The main measures that have the highest probability of being valid because of their consistency with the secondary measures are thus preferred in estimating the parameter. Preferably, for each set of secondary measurements, calculating the difference between a main measurement and the secondary measurements comprises calculating the difference between said main measurement and each secondary measurement of said set. The coherence score of a main measurement with the secondary measurements is advantageously obtained: by calculating, from the difference between said main measurement and each secondary measurement, masses respectively allocated to a first focal set corresponding to a first hypothesis. of coherence of said principal measurement with said secondary measurement, with a second focal set corresponding to a second hypothesis of lack of coherence of said principal measurement with said secondary measurement and with a third hypothesis corresponding to an uncertainty of the coherence of the measurement. principal with said secondary measure; estimating a belief that the principal measure is consistent with at least one of the secondary measures from the masses thus obtained; 25 estimating a plausibility that the main measure is consistent with at least one of the secondary measures from the masses thus obtained; calculating said coherence score by combining said belief and said plausibility by means of a combination function. According to one variant, for each set of secondary measurements, the calculation of the difference between a main measurement and the secondary measurements comprises calculating the difference between said main measurement and a secondary merged measurement obtained by merging the secondary measurements of said set. . The coherence score of a main measurement with the secondary measurements is then advantageously obtained: by calculating, from the difference between said main measurement and each merged secondary measurement, masses respectively allocated to a first focal set corresponding to a first coherence assumption of said principal measure with said secondary merged measure, a second focal set corresponding to a second hypothesis of lack of coherence of said main measure with said merged secondary measure and a third hypothesis corresponding to an uncertainty of coherence of the main measurement with said secondary merged measure; estimating a belief that the main measure is consistent with at least one of the merged secondary measures from the masses thus obtained; estimating a plausibility that the main measure is consistent with at least one of the merged secondary measures from the masses thus obtained; calculating said coherence score by combining said belief and said plausibility by means of a combination function. Said combination function may in particular be an average. the coherence score of a principal measurement with the secondary measurements is obtained: by calculating fuzzy values of strong coherence, average coherence, weak coherence between said principal measurement and each secondary measurement starting from the difference between said principal measurement and this secondary measure; calculating fuzzy rules of coherence operating on said fuzzy values; calculating said coherence score by combining said fuzzy rules thus calculated. Alternatively, the coherence score of a principal measurement with the secondary measurements is obtained: by calculating fuzzy values of high coherence, average coherence, low coherence between said principal measurement and each secondary measure merged from the difference between said measurement principal and this merged secondary measure; calculating fuzzy rules of coherence operating on said fuzzy values; calculating said coherence score by combining said fuzzy rules thus calculated using a combination operator. The fuzzy rules of coherence advantageously use blurred operators OR, AND and NOT of Lukasiewicz. The combination operator is for example an OR function.
    [0009] The fuzzy rules can be weighted with weighting factors before said combination, a weighting factor of a rule being all the higher as it involves a fuzzy value of high coherence with a greater number of elementary measures. .
    [0010] Alternatively, the consistency score of a primary measure with the secondary measures can be obtained by means of a supervised learning method of the "one class SVM" type. Similarly, the consistency score of a main measurement with the merged secondary measurements can be obtained by means of a "one class SVM" supervised learning method. According to a first advantageous example of embodiment, the said conditional estimate, relating to a fault configuration of the main sensors, is obtained as the median of the main measurements of the sensors in working order of this configuration, when the number of said sensors in working order is odd. According to a second advantageous example of embodiment, said conditional estimate, relating to a fault configuration of the main sensors, is obtained as the average of the main measurements of the sensors in working order of this configuration.
    [0011] BRIEF DESCRIPTION OF THE DRAWINGS Other features and advantages of the invention will appear on reading preferred embodiments of the invention, with reference to the appended figures in which: FIG. 1 represents an estimate of a flight parameter of an aircraft from measurements taken by sensors belonging to the same technology, according to a data fusion method known from the state of the art; Fig. 2A represents measurements of a flight parameter of an aircraft by two sets of sensors belonging to two dissimilar technologies; Fig. 2B represents an estimate of this flight parameter from the measurements in FIG. 2A, according to a data fusion method known from the state of the art; Fig. Figure 3 schematically illustrates the application context of the present invention; Fig. 4 schematically shows a flow chart of a data fusion method according to one embodiment of the invention; Fig. 5 represents masses of Dempster-Shafer focal sets as a function of the difference between a main measurement and a secondary measurement; Fig. 6 represents a first variant of calculation of a coherence score of a main measurement with the set of secondary measures; Fig. 7 represents a membership function of a linguistic term relating to a strong coherence between a main measurement and a secondary measurement; Fig. 8 represents a second variant of calculation of a coherence score of a main measurement with the set of secondary measures; Fig. 9 represents an estimate of the flight parameter of an aircraft from the measurements of FIG. 2A, by means of a data fusion method according to the present invention; Fig. 10 represents an example of a coherence classification of a principal measure by supervised learning.
    [0012] DETAILED DESCRIPTION OF PARTICULAR EMBODIMENTS We will consider in the following the estimation of a parameter, by means of a plurality of measurements of this parameter, obtained by different sensors. The present invention applies more particularly to the estimation of a flight parameter of an aircraft, for example its speed, its attitude or its position. However, it is not limited to such an application but can instead be applied to many technical fields in which it is necessary to merge measurements of a plurality of sensors. By sensor is meant here a physical sensor capable of directly measuring the parameter in question but also a system that can include one or more physical sensor (s) and signal processing means to provide an estimate of the parameter from measurements provided by these physical sensors. Similarly, the measurement of this parameter will be used to denote both a gross measurement of a physical sensor and a measurement obtained by a more or less complex signal processing from raw measurements. It is assumed that there is a first set of so-called main sensors capable of each providing a measurement of the parameter with a first degree of precision. The main sensors use a first technology in the sense defined above, that is to say use a first physical principle or a first specific implementation. It is also assumed that there is a plurality of so-called secondary sensor assemblies. These secondary sensors are capable of each providing a measurement, called secondary measurement, of the parameter in question but with a second degree of precision less than the first degree of precision. The degree of precision of the secondary measurements may vary between a second set to another, but it is in any case less than the degree of precision of the main measurements. Each set of secondary sensors is based on a second technology dissimilar to the first technology. This second technology therefore uses a different physical principle or a different implementation of the one used by the first technology, so that the probabilities of failure of a main sensor and a secondary sensor are independent. The technologies used by the different sets of secondary sensors are also dissimilar to each other.
    [0013] It is shown in FIG. 3 the context of application of the invention with the notations that will be used later. The parameter to be estimated is noted V (for example the speed of the aircraft). The measurements provided by the main sensors, Ai, ..., AN, are denoted by ai (t), ..., a N (t). It is assumed that P sets of secondary sensors (P 2) are available. For each set p = 1, ..., P, we denote Bil ', ... BmP the sensors and biP (t), ..., bmP (t) the measurements provided by these sensors (we assumed without loss generality that each set of secondary sensors had an identical number of sensors). The data fusion module, 300, receives, on the one hand, all the main measurements al (t), ..., a N (t) and, on the other hand, the sets of secondary measurements, biP (t), ..., b! '', (t), p = 1, ..., P. From the primary and secondary measurements, the data fusion module provides an estimate of the parameter, denoted V (t). It is shown in FIG. 4 an embodiment of the data fusion method implemented in the module 300. In a first step, 410, a difference dnmp is calculated between each principal measurement an (t), n = 1, ..., N and each secondary measure binP (t), m = 1, ..., M, p = 1, ..., P. This difference can be expressed as a modulus an (t) -bmP (t) 1, a quadratic difference (an (t) -bmP (t)) 2, a logarithmic log an (t), and so on. b fn (t) In a second step, 420, for each principal measure, an (a), from the previously obtained dnmp deviations, is calculated a consistency score, yr, of this main measurement with the set of measurements. secondary, binP (t). This consistency score expresses the extent to which the primary measure is consistent with the set of secondary measures. It will be assumed in the following, without loss of generality, that the coherence scores are between 0 and 1.
    [0014] Alternatively, steps 410 and 420 can be simplified by pre-fusing the measurements of each set of secondary sensors. In other words, the secondary measures binP (t), m = 1, .., M, are merged, for example by calculating their mean or their median. The measure thus merged is denoted bp (t). In this case, it will be understood that step 410 consists in calculating the differences between each principal measurement an (t), n = 1, ..., N and each merged secondary measure bp (t), p = 1,. , P and that step 420 consists of calculating, for each principal measurement, an (t), the coherence score, an, of this principal measurement with the set of merged secondary measurements, bp (t), p = 1 , .., P.
    [0015] In all cases, at step 430, for each configuration k possible failure main sensors, An, n = 1, ..., N, a first estimate of the parameter, V, said conditional estimate, noted Vk . A failure configuration is called an N-uplet of binary values vik, ..., vNk, each binary value vnk indicating whether the sensor An has failed or not. Without loss of generality, we will assume that the value 0 means a failure of the sensor and the value 1, an absence of failure. The failure configuration index k is the binary word vik, ..., vNk. It is therefore understood that 0 k 2N -1 and that the number of possible fault configurations is 2N. For a failure configuration k = vik, ..., vNk given, the conditional estimate Ésk only involves the main measurements of the sensors that are not out of order in this configuration, ie the main measurements an (t), such as vnk = 1, excluding other main measures.
    [0016] Advantageously, when the number of sensors that have not failed N in the configuration k, that is to say nckoh = Ivnk (where the sum is here calculated in N) n = 1 is odd, the conditional estimate is obtained as the median of the main measurements of the sensors in working order, ie: V k = mediantan (t) vnk = 1, n = 1, ..., N1 (1) On the other hand, when the number of main sensors in state of the configuration k, nckoh, is even, the conditional estimate is obtained as the average of the principal measurements of these sensors, ie: Vk = meansan (t) vnk = 1, n = 1, ..., N1 ( 2) Alternatively to the calculation of the average, we can obtain a first median value on the Ilk coh -1 (odd number) the weakest main measurements of the sensors in working order and a second median value on the nckoh -1 main measurements. higher, the estimate Vk then being calculated as the average between the first and second median values. It will be noted further that the conditional estimate may also be obtained as the average of the main measurements in the case where the number nckoh of the main sensors in operating state of the configuration k is odd. Whatever the parity of k, one can, if necessary, take into account in conditional estimation 12k, one or more secondary measures, merged or not. In this case, the conditional estimate 12k can be obtained by a combination of the main measurements of the sensors in working order of the configuration k and the secondary measurements, weighted by their respective degrees of precision. Thus, if one notes 12kh the conditional estimate of the parameter V based on the only principal measures of the configuration k and 1 '7' p the estimation of this same parameter by means of the secondary measures binP (t), m = 1, .., M, the conditional estimate 1-7 can have the following form: 1177 77 (2 ') 1 771 V1 ± IPP p = 1 12 k - P h X-, 1 1 ± L11 P p = 1 where i 'is the degree of precision of the principal measures an (t), n = 1, ..., N (assumed to be identical regardless of the principal measure) and 77 / P is the degree of precision of the measurements secondary binP (t), m = 1, .., M, the degrees of accuracy being even higher than the 10 measurements are more accurate (qh> 74 /, p = 1, ..., P). Finally, in the particular case where k = 0, in other words when all the main sensors are considered to be out of order, the conditional estimate 12k can be obtained from the secondary measurements, bin (0, for example as the median value or the average value of these measures.
    [0017] In step 440, we calculate weighting factors for the different possible failure configurations of the main sensors from the consistency scores of the main measurements with the secondary measurements, obtained in step 420. The weighting factor relative to one failure configuration of the main sensors reflects the probability of this configuration, given the consistency observed between each of the main measurements and the set of secondary measures. For a given configuration k = vil ', ..., vNk, the weighting factor fik of this configuration is calculated using: N / ,,' flk -n (an) vk '(1-601 1 n = 1 (3) From Expression (3) it is understood that the weighting factor of the failure pattern k is the product of the coherence scores for the working sensors in this configuration and the non-coherence scores for the sensors In other words, the coherence scores of each principal measurement are deduced from the set of secondary measurements, the probabilities of the different failure configurations Finally, in step 450, the estimation of the parameter is calculated. by weighting the conditional estimates 1 -, /,, relative to the different failure configurations, k = 0, ..., 2N -1 by their respective weighting coefficients, ie: 2N-1 12 = 1 igif7k (4) kO calculation of coherence scores between each principal measure an (t) and the set of secondary measures binP ( t), p = 1, ..., P, m = 1, ..., M (or in the case of prior fusion between each main measurement an (t) and the set of secondary measures bp (t), p = 1, ..., P) can be made according to several variants. In order to simplify the presentation, we will assume that P = 2 sets of secondary sensors are available and that the measurements for each of these two sets have been merged. We will not mention in the ratings of the main / secondary measures, the temporal variable t, which is henceforth considered to be implied. Thus, the measurements of the main sensors, b1 the measurement (merged) of the first set of secondary sensors, and b2 the measurement (merged) of the second set of secondary sensors.
    [0018] According to a first variant, the coherence scores are obtained by the Dempster-Shafer belief method. An account of this method can be found in the article by S. Le Hégarat et al. "Application of Dempster-Shafer evidence theory to unsupervised classification in multisource remote sensing", published in IEEE Trans. On Geoscience and Remote Sensing, Vol. 35, No. 4, July 1997, pp. 1018-1031. The Dempster-Shafer method supposes that we define on the one hand a set 0 of hypotheses, called framework of discernment and that one also has a plurality of sources of information providing a credit to this or that hypothesis. For each principal measure an, we can consider the following hypotheses, represented by subsets of 0: anbi: the measure an is consistent with the secondary measure b1; anbi: the measure an is not consistent with the secondary measure b1; X ,, bi: the consistency between the measures an and b1 is uncertain; anb2: the measure an is consistent with the secondary measure b2; anb2: the measure an is not consistent with the secondary measure b2; X ab2: the consistency between the measures an and b2 is uncertain.
    [0019] The descriptors capable of providing information on these hypotheses are the differences between the main measurements and the secondary measurements as calculated in step 410 of FIG. 4, and more precisely: - the difference c / n1 between the measures an and b1 for the assumptions anbi, anbi, X a ,, b; - the difference c / n2 between the measures an and b2 for the assumptions anb2, anb2, X a ,, b2.
    [0020] In the terminology used in the Dempster-Shafer theory, the subsets of 0 for which the descriptors can bring credit (or belief) are referred to as focal sets.
    [0021] Thus, the subsets anbi, anbi, X abi are the focal sets associated with the descriptor c / n1 and the subsets anb2, anb2, X ab2 are the focal sets associated with the descriptor c / n2. All possible intersections / meetings of focal groups and their intersections / meetings form the framework of discernment 0. In other words, 0 contains the focal sets and is stable by intersection and meeting operations. The amount of credit a descriptor allocates to an associated focal set is referred to as mass.
    [0022] Fig. 5 represents an example of allocation of masses to the focal sets anbi, anbi, X abi by the descriptor c / n1. The mass values are between 0 and 1. Note that the mass m (anbi) allocated to the focal set anbi is a decreasing function of the difference c / n1 and reaches the value zero for a threshold value g. The mass allocated to the focal set anbi, on the other hand, is an increasing function of the difference c / n1, starting from the value zero when the difference is equal to the threshold g. Finally, the mass allocated to the uncertainty X abi is maximal (M (X ab) -1) when the masses allocated to the focal sets a nbi and anbi are minimal (in other words for c / n1 = 8) and this is minimal. (m (Xa) = 1-, u) when one of the masses allocated to the focal sets anbi and a nbi is maximal (m (anbi) = of or m (anbi) =, u). In all cases, the sum of the masses allocated to the focal sets by the descriptor is equal to 1: m (anbi) + m (anbi) + m (X a'b) = 1 (5-1). a: m (anb2) + m (anb2) + m (X anbz = 1 (5-2) The intersections between the focal sets are subsets of 0 (and thus elements of the set 0 (0) of parts of 0) For a given principal measure year, the intersections can be represented by the following table: anbi anbi X a, bi anb2 a nbi n anb2 Xabi n anb2 a nbi n anb2 anb2 anbi n anb2 a nbi n anb2 Xabi n anb2 X anb2 ank n X anb2 anbi n xab, Xci ,, bi n Xa, b2 ,, Table I We now consider the following hypothesis H n, element of 0: H n: the principal measure an is consistent with at least l one of the secondary measures b1 and b2 This hypothesis can be represented by the union of the subsets appearing in the first column and the first row of Table I, ie those which appear in the following table: Anbi anbi X a, bi anb2 anbi n anb2 Xabi n anb2 a nbi n anb2 anb2 a nbi n anb2 X anb2 ank n X anb2 Table II Indeed, the first column of the table corresponds to the consistency of the main measure an with the measure secondary b1 and the first row of the table corresponds to the consistency of the principal measure an with the secondary measure b2.
    [0023] The belief of the hypothesis II n, is defined as the sum of the beliefs of the sets included in II n, in other words: Bel (Hn) = 1 M (n) (6) E0 (0) 11 cii 'In this case , the belief of hypothesis II can be expressed in the following way: Bel (Hn) = m (anbinanb2) + m (anbinanb2) + ~ n (Xabna'b2) + m (anbina'b2) + m ( anbinx,) and, assuming no conflict between the secondary measures: Bel (H n) = m (anbi) .m (anb2) + m (anbi) .m (anb2) + m (X, b) .m (anb2) + m (anbi) .m (anb2) + m (anbi) m (X ,,, b2) The Dempster-Shafer method also makes it possible to define the plausibility of the hypothesis II n as the sum of the beliefs of the sets with non-empty intersection with I In, ie: P / S (H) = / in (n) (9) E0 (0) 11n1 / '# 0 (7) (8) Considering again in Table I, we understand that the sum of the expression (9) involves all the elements of the table except the central element, which can be represented by: anbi anbi X a, bi anb2 anbi n anb2 Xabi n anb2 anbi na nb2 anb2 anbi n anb2 Xabi n anb2 X anb2 anbi n X a ,, b2 anbi n xa ,, Xci ,, bi n Xa, b2 Table III The plausibility of Hn can be expressed simply from the belief of Hn, in taking into account the additional elements appearing in Table III: Pis (H n) = BI (H n) + m (a nbi) .m (X an2) + m (X a) .m (anb2) + m (X an) .m (X an2) (10) Belief and plausibility frame the probability that the hypothesis Ha is well verified. The consistency score of the main measure an is then defined by combining the belief and the plausibility of Ha using a combination function. One can for example define the coherence score of the principal measurement an by the arithmetic mean: 1 year = - 2 (Bel (H n) + Pls (Ha) 20 Other combination functions (for example geometric mean) of the Belief and plausibility may be contemplated by those skilled in the art without departing from the scope of the present invention. Returning to the general case of a number / 32, any set of secondary sensors, FIG. schematic calculation of the coherence score of a principal measurement an with the set of secondary measures bp, p = 1, ..., P. At step 610, the masses allocated to the focal sets anbp, anbp, are calculated X ab p, p = 1, ..., P, by the clip descriptor, expressing the difference between the measurements an and bp, for example the mass functions are heuristically determined in advance. calculates the belief Bel (11p) of the hypothesis H p of coherence of the measure an with at least one of the secondary measurements bp, p = 1, ..., P, from the masses calculated in the previous step. In step 630, the plausibility P / s (1-1) of the hypothesis H p of consistency of the measurement an with at least one of the secondary measures bp, p = 1, ..., is computed. P, from the Bel belief (11p) computed in the previous step and masses calculated in step 620. In step 640, the coherence score an is computed from the main measurement an with the set secondary measures by combining Bel belief (11p) and Pls pliability (11p), oep = F (Bel (1-1p), P1s (Hp)) where F is a combination function, for example an average. In the case where the steps 410, 420 are not merged beforehand, there are MP secondary measurements bli, p = 1, ..., P, m = 1, ... , M. The masses allocated to the three sets of focal lengths can be calculated for each of these MP measurements. The calculation of the coherence score then continues in a similar way, the consistency assumption Ilp being replaced by a looser, finer assumption, that the measure an is consistent with at least one of the secondary measures b'n, p = 1, ..., P, m = 1, ..., M.
    [0024] According to a second variant, the coherence scores are obtained by a fuzzy logic method. To do this, we define a plurality of linguistic variables Lm (in the sense of fuzzy logic) as the consistency of the measurement an with the secondary measure bp (merger of the secondary measures binP, m -1, ..., M). The linguistic variable Lm, can be expressed in the form of three linguistic terms {strong coherence, average consistency; low coherence}, each of these terms being semantically defined as a fuzzy set on the clip gap values, between the measures an and bp.
    [0025] Fig. 7 represents an example of membership function defining the linguistic term "strong coherence" for the linguistic variable Lm ,. The abscissa has been indicated as the clip difference, between the measures an and bp. This membership function can be parameterized by a transition threshold value 6 and a slope y of the line segment connecting the values 0 and 1. The membership function can follow a more complex law, for example a nonlinear law ( for example an arctan law) or else a law with linear parts connected by polynomial functions (for example splines). The parameters of these membership functions can be obtained heuristically or by learning. In the same way, membership functions give the definition of the terms "average consistency" and "low coherence".
    [0026] If we take the previous case of two secondary (merged) measures b1, b2, the computation of the coherence score can make use of the following fuzzy rules: Rl: if strong coherence with b1 and strong coherence with b2, the measure an is coherent with all the secondary measures; R2: if strong coherence with bt, the measure an is consistent with the set of secondary measures; R3: if strong consistency with b2, the measure an is consistent with the set of secondary measures. The coherence score of the measure an with the set of secondary measures is then determined by: an = (anbi AND an192) OR (anbi) OR (a'b2) (12) where AND and OR are respectively AND operators ( intersection) and fuzzy OR (meeting) and where anbi, anb2 are respectively the fuzzy values relative to the linguistic term "strong coherence" for linguistic variables Ln1 and 42. Other sets of fuzzy rules can be used alternately or cumulatively. For a number P of secondary measurements, it will be possible for example to consider that if the measurement an is coherent with at least a predetermined number P <P of these measurements, then it is coherent with the set of secondary measures. In a general way, if we denote Rkn, k = 1, ..., K, the fuzzy coherence rules for the measure an, the coherence score will be determined by: a = OR (K) k = 1 K ( 13) where the rules R kn involve the membership functions of the linguistic terms "strong coherence", "average consistency" and "low coherence" of the linguistic variables Ln ,, and the fuzzy operators AND, OR and NOT.
    [0027] The fuzzy operators AND, OR and NOT will preferably be the fuzzy operators of Lukasiewicz defined by: a AND b = max (0, a + b -1) (14-1) a OR b = min (1, a + b) ( 14-2) NOT (a) = 1-a (14-3) Alternatively, we can use probabilistic operators or Zadeh operators. The fuzzy rules intervening in expressions (12) and (13) can be advantageously weighted. For example, in the case of expression (12), it is conceivable that a larger weight is attributed to the conjunctive term anbi AND anb2 than to the terms anbi, a'b2. The consistency score obtained by weighting the fuzzy rules can then be expressed as follows: an - ((1-22) [anbi AND anb2]) OR (2 [anbi]) OR (2 [anb2]) (15) where 0 <2 <1/2. Similarly, in the expression (13) a higher weight can be given to the R kn rules, which translate the coherence with a large number of secondary measures and a lower weight to those resulting in coherence with a smaller number of such measures. . The weighting factor 2 may be adaptive. For example, if the parameter to be estimated is the speed of the aircraft, the weighting factor 2 may be a function of the Mach number. For example, for high Mach numbers one will be able to tolerate more widely a coherence with one of the secondary measures only and thus to envisage a weighting factor 2 greater than for lower Mach numbers. Finally, other weighting rules than (15) can be envisaged without departing from the scope of the present invention.
    [0028] Fig. 8 represents the second variant of calculation of a coherence score of a principal measurement an with the set of secondary measures bp, p = 1, ..., P according to said second variant. In step 810, the fuzzy values relating to the "strong coherence" linguistic term, apbp, are calculated as a function of the differences cIpp between the main measurement an and the secondary measures bp, p = 1, ..., P. This calculation of fuzzy values (or fuzzification) is performed from the membership functions such as that shown in FIG. 7 (with a membership function by secondary measure). These can be obtained heuristically (setting parameters ci and y) or result from a learning phase. As indicated above, other more complex membership functions, including in particular non-linear laws may be used without departing from the scope of the present invention, as indicated above. In step 820, fuzzy predefined coherency rules are applied, Rkn, k = 1, ..., K, operating on the fuzzy values obtained in the previous step.
    [0029] In step 830, the consistency score oen = OR (R1) where the operator k = 1 K OR is a fuzzy operator OR, preferably an OR operator of Lukasiewicz, is calculated. The different rules can also be weighted as explained above. If we do not proceed to the preliminary fusion of the secondary measures in steps 410, 420, we have MP secondary measures bli, ',, p = 1, ..., P, m = 1, ..., M . We then define linguistic variables Lnmp as the consistency of the measure an with the secondary measure binP, m = 1, ..., M. The linguistic variable Lnmp can be expressed in the form of three linguistic terms {strong coherence; average consistency; low coherence}, each of these terms being semantically defined as a fuzzy set on the dnmp difference values between the an and bin measurements. The computation of the coherence score continues in a similar way, starting from predefined fuzzy rules Rkn, k = 1, ..., K, each rule involving the membership functions of the linguistic terms "strong coherence", "average coherence" And "weak consistency" of Lnmp linguistic variables, and fuzzy operators AND, OR, and NOT. The coherence score can be obtained by means of the relation a = OR (R1) or by means of a weighted relation n k = 1 K as explained above.
    [0030] Fig. 9 represents an estimate of the flight parameter of an aircraft (here the speed) from the measurements of FIG.
    [0031] 2A by means of a data fusion method according to the present invention. In the present case, N = 3 main sensors and two secondary sensors (P = 2, M = 1) are available. The estimation of the flight parameter using a calculation of coherence scores according to the first variant (Dempster-Shafer method) and VFL (t) has been designated by 12 DS (t), the estimation of this parameter using a calculation of the scores of consistency according to the second variant (fuzzy logic). Note that the two estimates 121) s (t) and VFL (t) of the flight parameter V (t) are valid and accurate. Only a transient deflection of small amplitude appears. This amplitude can be controlled in particular by appropriately choosing the parameters of the mass functions in the first variant and the parameters of the membership functions in the second variant. The data fusion method described in connection with FIG. 4 uses (steps 410,420) to calculate the differences between each main measurement and the set of secondary measures and then a calculation of the coherence score according to these differences. Alternatively, a supervised learning method can be used to obtain a coherence score directly. In a learning phase, the algebraic deviations between a main measurement and the secondary measurements are recorded for a plurality of typical cases and the main measurement is classified as coherent or incoherent. This classification can be performed according to a parameter. These typical cases make it possible to determine a coherence zone, an inconsistency zone and an undecidable zone in a given space. It will be possible to use, for supervised learning, a support vector machine or a class, called a single-class support vector machine or "one ciass SVM", as described, for example, in the article by B. Scheilkopf and al. entitled "Estimating the support of a high dimensional distribution" published in the journal Neural Computation, Vol. 13, pp. 1443-1471, 2001. After this learning phase, an automatic classification of a new main measurement can be carried out according to the part of the space in which it is located. The "one class SVM" method provides a positive or negative score depending on whether the main measure is coherent or not. This score can then be transformed into a consistency score of between 0 and 1. FIG. 10 represents an example of coherence classification of a principal measure by supervised learning. The abscissa is the Mach number and the ordinate is the algebraic difference between the main measure and a secondary measure. The crosses correspond to the learning cases of the SVM method. These learning cases make it possible to distinguish three distinct zones in the classification diagram: a zone 1010 corresponding to the coherence zone, a zone 1030 corresponding to the zone of incoherence and a transition zone 1020. To determine the coherence or the inconsistency of a main measurement, it is then enough to see in which zone of the diagram is the measurement. The coherence scores are then given on the one hand by the Mach number and on the other hand by the algebraic difference between the main measurement and the secondary measurement. According to one variant, the coherence scores are not given directly by the Mach number and the difference between the main measurement and a secondary measurement but the parameters of the membership functions (for example, the transition threshold a and the slope y in the case of the law illustrated in Fig. 7) are adaptive. Parameters can depend on the number of Mach in an analytical way (heuristic dependence law) or the values of these parameters can be stored in a correspondence table (also known as "look-up" table), addressed by the number from Mach. More generally, the supervised coherence classification or the adaptive parameterization of the membership functions can be performed, for different phases or different flight conditions, in relation or not with the Mach number. For example, there may be more critical consistency conditions for some key measurements during the landing and / or takeoff phases. 1 0
    权利要求:
    Claims (15)
    [0001]
    REVENDICATIONS1. A method for merging the measurements of a parameter, in particular a flight parameter of an aircraft, from a first set of so-called main measurements taken by a first set of sensors, called the main ones, and a plurality of second sets of measurements, called secondary, taken by second sets of sensors, called secondary, the main measures having a degree of precision greater than the secondary measures, characterized in that: - a difference between each measurement is calculated (410) principal and said secondary measurements; determining (420) a coherence score of each principal measurement with the secondary measurements by means of the difference thus obtained; for each possible failure configuration of the main sensors, (430) a first, so-called conditional, estimate of said parameter is made from the main measurements taken by the sensors that have not failed in said configuration; for each fault configuration, a weighting coefficient is determined (440) from the coherence scores of the main measurements previously obtained; an estimate of said parameter (450) is made by weighting the conditional estimates relating to the different failure configurations by the weighting coefficients corresponding to these configurations.
    [0002]
    A method of merging measurements according to claim 1, characterized in that, for each set of secondary measurements, calculating the difference between a main measurement and the secondary measurements comprises calculating the difference between said main measurement and each secondary measure of said set.
    [0003]
    3. A method for merging measurements according to one of the preceding claims, characterized in that the coherence score of a main measurement with the secondary measurements is obtained: by calculating (620), from the difference between said main measurement and each secondary measurement, masses respectively allocated to a first focal set corresponding to a first coherence assumption of said main measurement with said secondary measurement, to a second focal set corresponding to a second hypothesis of non-coherence of said measurement principal with said secondary measure and with a third hypothesis corresponding to an uncertainty of the consistency of the main measurement with said secondary measure; estimating (630) a belief that the main measure is consistent with at least one of the secondary measures from the masses thus obtained; estimating (640) a plausibility that the principal measure is consistent with at least one of the secondary measures from the masses thus obtained; calculating (650) said coherence score by combining said belief and said plausibility by means of a combination function.
    [0004]
    A method of merging measurements according to claim 1, characterized in that, for each set of secondary measurements, the calculation of the difference between a main measurement and the secondary measurements comprises calculating the difference between said main measurement and a secondary merged measurement obtained by merging the secondary measurements of said set.
    [0005]
    A method of merging measurements according to claim 4, characterized in that the coherence score of a main measurement with the secondary measurements is obtained: by calculating (610), from the difference between said main measurement and each merged secondary measurement, masses respectively allocated to a first focal set corresponding to a first coherence assumption of said main measurement with said secondary merged measure, to a second focal set corresponding to a second hypothesis of lack of coherence of said main measurement with said secondary merged measure and a third hypothesis corresponding to an uncertainty of the consistency of the main measurement with said merged secondary measure; estimating (620) a belief that the principal measure is consistent with at least one of the merged secondary measures from the masses thus obtained; estimating (630) a plausibility that the principal measure is consistent with at least one of the merged secondary measures from the masses thus obtained; calculating (640) said coherence score by combining said belief and said plausibility by means of a combination function.
    [0006]
    The method of merging measurements according to claim 3 or 5, characterized in that said combining function is an average.
    [0007]
    7. A method for merging measurements according to claim 2, characterized in that the coherence score of a main measurement with the secondary measurements is obtained: by calculating (810) fuzzy values of high coherence, average consistency, low consistency between said primary measure and each secondary measure from the difference between the primary measure and that secondary measure; calculating (820) fuzzy coherence rules operating on said fuzzy values; calculating (830) said coherence score by combining said fuzzy rules thus calculated.
    [0008]
    8. A method for merging measurements according to claim 4, characterized in that the coherence score of a main measurement with the secondary measurements is obtained: by calculating (810) fuzzy values of high coherence, average consistency, low consistency between said main measure and each secondary measure merged from the difference between said main measure and this merged secondary measure by calculating (820) fuzzy coherence rules operating on said fuzzy values; calculating (830) said coherence score by combining said fl uid rules thus calculated using a combination operator.
    [0009]
    9. A method of merging measurements according to claim 7 or 8, characterized in that the fuzzy coherence rules use fuzzy operators OR, AND and NOTde Lukasiewicz.
    [0010]
    10. A method of merging measurements according to claim 9, characterized in that the combination operator is a function OR.
    [0011]
    11. A method of merging measurements according to claim 10, characterized in that the fuzzy rules are weighted with weighting factors before said combination, a weighting factor of a rule being all the higher as it puts into effect. game a fuzzy value of strong consistency with a larger number of elementary mnesurey.
    [0012]
    A method of merging measurements according to claim 2, characterized in that the coherence score of a main measurement with the secondary measurements is obtained by means of a supervised learning method using a monO-medium support vector machine. classroom.
    [0013]
    A method of merging measurements according to claim 4, characterized in that the consistency score of a main measurement with the merged secondary measurements is obtained by means of a supervised learning method using a mono carrier vector machine. -classroom.
    [0014]
    14. The method for merging measurements according to one of the preceding claims, characterized in that said conditional estimate, relating to a configuration of the main sensors, is obtained as the median of the main measurements of the sensors in working order of this configuration. when the number of said operating sensors is odd.
    [0015]
    15. A method for merging measurements according to one of claims 1 to 13, characterized in that said conditional estimate, relating to a fault configuration of the main sensors, is obtained as the average of the main measurements of the sensors in working condition of this configuration. I. 0
    类似技术:
    公开号 | 公开日 | 专利标题
    FR3013834A1|2015-05-29|METHOD FOR MERGING SENSOR DATA USING A COHERENCE CRITERION
    FR3017705A1|2015-08-21|METHOD OF MELTING SENSOR DATA.
    EP2473867B1|2013-10-09|Multi-target data processing for multi-receiver passive radars in an sfn or mfn mode
    EP1914561B1|2017-09-06|Method of satellite location of a vehicle and guarantee of integrity with selection of a subgroup of satellites
    EP0675374A1|1995-10-04|Method for tracking moving bodies
    EP2105760A1|2009-09-30|Method and system for tracking emitters
    EP0490745A1|1992-06-17|Method and apparatus to improve the reliability of aircraft flight data
    EP2541200B1|2014-03-19|Navigation device and method including a plurality of hybrid inertial navigation systems
    Schwall et al.2002|A probabilistic approach to residual processing for vehicle fault detection
    WO2015150129A1|2015-10-08|Method for geolocating the environment of a carrier
    EP3385677B1|2020-01-22|A system and a method of analyzing and monitoring interfering movements of an inertial unit during a stage of static alignment
    FR3067326A1|2018-12-14|METHOD AND SYSTEM FOR PERMITTING COMPONENT CONTROL REDUNDANCY IN A DIGITAL NETWORK OF INTELLIGENT DETECTION DEVICES
    FR2875626A1|2006-03-24|METHOD FOR DETECTING AND PITCHING TARGET TARGETS IN AN OPTRONIC MONITORING SYSTEM
    EP3547159A1|2019-10-02|Method for assessing the conformity of a tracking system with a set of requirements and associated devices
    FR3000196A1|2014-06-27|Device for providing values of e.g. speed of aircraft to pilot, has computers computing value for navigation parameter and value of error estimation using measurement information and unique fusion algorithm for use by user systems
    EP2105380A1|2009-09-30|Attitude estimation method of a star tracker
    EP2321728A2|2011-05-18|Method and system for generating a supervision device from specified feared behaviours
    EP2491424A1|2012-08-29|Method for simultaneously locating and mapping via resilient non-linear filtering
    Mosin et al.2021|Applicability of Machine Learning Architectural Patterns in Vehicle Architecture: A Case Study
    Damm et al.2018|Perspectives on the Validation and Verification of Machine Learning Systems in the Context of Highly Automated Vehicles.
    EP3623758B1|2021-04-21|Positioning system, and associated method for positioning
    WO2012076661A1|2012-06-14|Method and device making it possible to generate a control system on the basis of specified feared behaviours
    EP2765514B1|2015-08-19|Apparatus for a secured client application of a symbolic display system of the client-server type
    Lukoyanov et al.2020|Functional and test diagnosis of onboard systems using dynamic models
    Sun et al.2021|Hierarchical Variational Auto-Encoding for Unsupervised Domain Generalization
    同族专利:
    公开号 | 公开日
    FR3013834B1|2015-12-25|
    US20150148997A1|2015-05-28|
    CN104677379B|2019-11-29|
    US9268330B2|2016-02-23|
    CN104677379A|2015-06-03|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US5963662A|1996-08-07|1999-10-05|Georgia Tech Research Corporation|Inspection system and method for bond detection and validation of surface mount devices|
    EP1496414A1|2003-07-07|2005-01-12|Airbus France|Device and method for controlling the validity of at least one parameter calculated by an aircraft anemometer station|
    DE102011056644A1|2010-12-22|2012-06-28|General Electric Company|Method and system for compressor condition monitoring|EP3553617A1|2018-04-13|2019-10-16|Airbus Operations S.A.S.|Method and system for merging measurements of flight parameters of an aircraft|
    FR3113518A1|2020-08-24|2022-02-25|Arianegroup Sas|Failure determination method for a set of sensors|US7089099B2|2004-07-30|2006-08-08|Automotive Technologies International, Inc.|Sensor assemblies|
    JP4229358B2|2001-01-22|2009-02-25|株式会社小松製作所|Driving control device for unmanned vehicles|
    GB2392250B|2001-06-25|2004-11-10|Invensys Sys Inc|Sensor fusion using self evaluating process sensors|
    US7099796B2|2001-10-22|2006-08-29|Honeywell International Inc.|Multi-sensor information fusion technique|
    WO2006104552A1|2005-03-29|2006-10-05|Honeywell International Inc.|Method and apparatus for high accuracy relative motion determinatation using inertial sensors|
    FR2905763B1|2006-09-11|2008-12-12|Eurocopter France|METHOD AND SYSTEM FOR DIAGNOSING AN AIRCRAFT FROM MEASUREMENTS CARRIED OUT ON THE AIRCRAFT.|
    ES2400708T3|2008-08-27|2013-04-11|Saab Ab|Use of an image sensor and a time tracking filter to avoid collisions in flight|
    CN102192736B|2010-03-03|2016-02-10|中国船舶重工集团公司第七0七研究所|The optimized treatment method of sensor output data of ship comprehensive control system|
    US8935071B2|2011-05-05|2015-01-13|GM Global Technology Operations LLC|Optimal fusion of electric park brake and hydraulic brake sub-system functions to control vehicle direction|
    CN102306206B|2011-07-18|2013-10-30|福州大学|Self-adapting consistent data fusion method|
    FR2978858B1|2011-08-01|2013-08-30|Airbus Operations Sas|METHOD AND SYSTEM FOR DETERMINING FLIGHT PARAMETERS OF AN AIRCRAFT|
    WO2013037855A1|2011-09-12|2013-03-21|Continental Teves Ag & Co. Ohg|Sensor system comprising a vehicle model unit|
    US9562764B2|2012-07-23|2017-02-07|Trimble Inc.|Use of a sky polarization sensor for absolute orientation determination in position determining systems|
    CN103065293A|2012-12-31|2013-04-24|中国科学院东北地理与农业生态研究所|Correlation weighted remote-sensing image fusion method and fusion effect evaluation method thereof|
    CN103313386B|2013-06-18|2016-04-27|浙江大学|Based on the radio sensing network method for tracking target of consistency on messaging right-value optimization|GB2541939B|2015-09-04|2018-04-18|British Telecomm|Evaluation of network condition|
    CN107543540B|2016-06-27|2020-05-15|杭州海康机器人技术有限公司|Data fusion and flight mode switching method and device for flight equipment|
    CN106778883A|2016-12-23|2017-05-31|贵州电网有限责任公司电力科学研究院|A kind of evidence theory intelligent patrol detection information fusion method based on fuzzy set|
    FR3065543B1|2017-04-19|2019-05-03|Airbus Operations |FLIGHT CONTROL CALCULATOR OF AN AIRCRAFT|
    US10948511B2|2018-03-05|2021-03-16|Honeywell International Inc.|Apparatus and method for verifying operation of air data probes|
    CN109982287B|2019-03-13|2021-10-15|北京工业大学|Forestry multi-sensor fire alarm system based on zigbee wireless sensor network|
    CN111058958B|2019-12-11|2022-01-07|厦门林巴贺航空发动机股份有限公司|Control method of piston type aircraft engine|
    CN111257592A|2020-03-05|2020-06-09|广东零偏科技有限公司|Static discrimination method for detection device|
    CN112003891A|2020-07-16|2020-11-27|山东省网联智能车辆产业技术研究院有限公司|Multi-sensing data fusion method for intelligent networked vehicle controller|
    法律状态:
    2015-11-19| PLFP| Fee payment|Year of fee payment: 3 |
    2016-11-18| PLFP| Fee payment|Year of fee payment: 4 |
    2017-11-21| PLFP| Fee payment|Year of fee payment: 5 |
    2019-11-20| PLFP| Fee payment|Year of fee payment: 7 |
    2020-11-20| PLFP| Fee payment|Year of fee payment: 8 |
    2021-11-19| PLFP| Fee payment|Year of fee payment: 9 |
    优先权:
    申请号 | 申请日 | 专利标题
    FR1361754A|FR3013834B1|2013-11-28|2013-11-28|METHOD FOR MERGING SENSOR DATA USING A COHERENCE CRITERION|FR1361754A| FR3013834B1|2013-11-28|2013-11-28|METHOD FOR MERGING SENSOR DATA USING A COHERENCE CRITERION|
    US14/553,125| US9268330B2|2013-11-28|2014-11-25|Method for fusing data from sensors using a consistency criterion|
    CN201410709508.9A| CN104677379B|2013-11-28|2014-11-28|Method for using data of the conformance criteria fusion from sensor|
    [返回顶部]