• 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:

    公开号:FR3017705A1
    申请号:FR1451290
    申请日:2014-02-18
    公开日:2015-08-21
    发明作者:Alain Guillet；Jabiri Ahmed Taha Zaari
    申请人:Airbus Operations SAS；
    IPC主号:
    专利说明:

    [0001] METHOD OF MELTING SENSOR DATA. TECHNICAL FIELD The present invention relates to the field of the fusion of sensor measurements, more particularly to estimate 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. Because of the criticality of these systems, the sensors are redundant, that is to say that several sensors provide measurements of the same parameter. These different measurements are then processed by a measurement fusion method in order to provide an estimate of the most accurate parameter possible. The parameter estimate is for example an average value, or a median value of several measurements each provided by a separate sensor. The parameter estimate is also called "consolidated value", or "estimated value". In order to improve the estimation of the parameter, the methods for merging measurements generally comprise a search for a possible anomaly on one of the sensors, in order not to take into account an aberrant measurement provided by a sensor presenting an anomaly.
    [0002] For example, a measurement fusion method is known comprising, at each instant, the following steps: calculating the median value of the measurements provided by the different sensors; positioning the measurements provided by the sensors relative to a tolerance band of predetermined width, centered on said median value; elimination of measurements outside this tolerance band; estimating the value of the parameter from the remaining measurements. The step of positioning the measurements relative to a tolerance band performs a search for a possible anomaly on one of the sensors, this anomaly being revealed by the provision of an aberrant measurement.
    [0003] A disadvantage of such a measurement merging method is that it does not reliably determine a probability of false alarm, that is to say the probability of considering that a sensor has an anomaly, whereas he does not present any. An object of the present invention is to provide a method for merging sensor measurements, more particularly for estimating a flight parameter of an aircraft, which makes it possible to reliably determine a false alarm probability. DISCLOSURE OF THE INVENTION This objective is achieved with a method of merging measurements of a parameter, in particular a flight parameter of an aircraft, from measurements of this parameter provided respectively by a plurality of sensors. The method according to the invention comprises the following steps and sub-steps: 1) search for a possible anomaly on one of at least two sensors called sensors of interest, comprising the following substeps: a) for each sensor of interest, calculation of a so-called detection difference, proportional to the absolute value of a difference between a measurement provided by this sensor of interest, and an estimate of the parameter calculated from the measurements provided by the other sensors of interest; lb) comparing each detection deviation to a corresponding predetermined threshold; 1c) according to the result of the comparisons, determining the presence or the absence of an anomaly on one of the sensors of interest with a total probability of false alarm determined; and 2) fusion of measurements provided by the sensors of interest, to provide an estimate of the parameter, called the final estimate.
    [0004] Each sensor of interest is associated with a normal law characteristic of the total error on the measurement it provides, in the absence of anomaly. This normal distribution is denoted N (μ,; o-,), where a- is a standard deviation, and p, is an average. We denote by y, the measurement provided by the sensor of interest i, and) F1, an estimate of the parameter calculated from the measurements provided by the other sensors of interest, that is to say excluding the sensor of interest i. The detection difference relating to the sensor of interest i is denoted T, and according to the invention we have: (1) Let x be the real value of the parameter, we have, when all the sensors of interest function correctly: V3i = Yi - = (yi - x) + (x - = Vli + V2i (2) = (y, - x) is the difference between the measurement provided by the sensor of interest i and the actual value of the parameter. is the error on the measurement provided by the sensor of interest i, it is thus a random variable defined, in the absence of anomaly on the sensor of interest i, by a normal law of average p. 'of standard deviation o-, and of variance 0-7, denoted N (μ,; o-,) This normal distribution is determined from the characterization of the errors of each sensor i V2i = (x -5C1-3 is the difference between the estimate obtained from the measurements provided by the sensors of interest other than the sensor i and the actual value of the parameter.This is the error on the estimate obtained from measurements s provided by sensors of interest other than the sensor i. It is thus a random variable defined, in the absence of anomaly on these sensors of interest, by a normal law of average of variance a2i, denoted 6_,). This normal law is determined from the characterization of the errors of each sensor i.
    [0005] The variance a_i depends on the values of the different al, j # i, and on the calculation mode of the mean. Kt_i depends on the values of the different pti, j # and the calculation mode of. For each sensor, the normal law characteristic of the total error on the measure that it provides is advantageously brought back to a normal centered law, in order to be free from calculation of the average The variables V1, and V2, are independent random variables, because V1, is calculated only from the measurement provided by the sensor of interest i, and V2, is calculated from the measurements provided by all the other sensors of interest j = i. Their covariance is zero, so that in the absence of anomaly on the sensors of interest, the variable V3, = V1, + V2, is defined by a normal distribution whose variance is equal to 0-7 + 0 -2i. We note this variance aT2i. This is because V1, and V2, are independent random variables that the variance of V3, is equal to the sum of the variances of V1, and V2 ,.
    [0006] The fact that the random variables V1, and V2 are independent also implies that the random variable V3, is defined by a normal law of average pT, equal to the sum of the means p., And the variable V3, can thus be defined, in the absence of anomaly on the sensors of interest, by a normal law whose variance is equal to = 0-7 + 0-2i, and whose average is equal to pT, = μi + noted .N (I aTi). From this normal law, it is possible to deduce the value of a predetermined threshold associated with the detection deviation Ti distinguishing the case where one of the sensors of interest has an anomaly, of the case where none of the sensors of interest does not present an anomaly. This threshold is associated with a probability of false alarm noted PFAI, which corresponds to the probability of considering, from the comparison between the detection deviation Ti and a threshold Td, that a sensor of interest has an anomaly whereas he does not present any. The probability of false alarm PFA1 is related to the detection deviation Ti, but not necessarily to the sensor i itself. It may be another sensor that has the anomaly. In the following, we detail the example in which Ti = lyi = 1V3il. Those skilled in the art will readily adapt this example according to the proportionality factor between 1V3,1 and Ti (see equation (3)).
    [0007] It is assumed that only one of the sensors may have an anomaly at a time t. We note: H0 the hypothesis according to which all the sensors of interest function correctly; H1 the assumption that one of the sensors of interest has an anomaly; and Td, the value of the predetermined threshold (related to the detection deviation Ti), such that Ti> Td, implies that we accept the hypothesis H1 (and we reject the hypothesis H0).
    [0008] It may be noted that it is of little importance to know which deviation Ti has exceeded the detection threshold Tdi, because that will not allow to deduce that it is the sensor of interest i which has an anomaly. We therefore do not distinguish several hypotheses H1, which would each be related to a deviation Ti.
    [0009] The probability of false alarm then corresponds to the probability of rejecting the hypothesis H0 whereas the hypothesis H0 is true. This probability of false alarm can be expressed from the detection deviation Ti: PFAi = P i 7 'of (IiTi; aT i)) + co PFAi = T di 67, i) (t) dt, where fi / - (, Ti, is the probability density corresponding to the law .7 ^ f (piTi; an); PFAi = 1 - where aTi) is the distribution function of the law .N (I an). So that the decision threshold Tdi is defined by: Tdi = p, Ti; 0-Ti) (1-PFAi) (3) In other words, a desired value of the false alarm probability PFAi is set, then the value of a predetermined threshold is deduced for the detection difference Tdi, separating the case where all the sensors of interest function correctly of the case where one of the sensors of interest has an anomaly.
    [0010] The other decision thresholds Tdi, each corresponding to a detection deviation Ti relative to the sensor of interest j # i, can be defined in the same way. Each decision threshold Tdi is then defined on the basis of a false alarm probability PFAi which corresponds to the probability of considering, from the comparison of the detection difference Ti with a threshold Tdi, that a sensor of interest has an anomaly when it does not have one. In practice, each detection gap is tested by comparing it with its corresponding decision threshold. If at least one of these tests shows that one of the sensors of interest has an anomaly, it is deduced that the hypothesis H1 is true. The total probability of false alarm, noted PFA is the sum of the probabilities of false alarm associated with each of the detection differences: PFA = PFAi, are M sensors of interest (4) It is thus seen that thanks to the use of a decision criterion defined by independent random variables y, and it is possible to know the total probability of false alarm associated with the diagnosis according to which all the sensors of interest function normally or in which one of the sensors of interest has an anomaly . We propose a method of fusion of measurements offering a rigorous characterization of a final estimate of a parameter. The detailed description of particular embodiments provides additional details relating to the implementation of the invention, in particular the determination of normal laws characterizing the errors of the sensors, the determination of the variables V1, and V2, etc. According to a first embodiment of the invention, the method comprises a search for a possible anomaly on one of two sensors of interest, and a fusion of the measurements provided by these two sensors of interest, to provide an estimate. of the parameter, called final estimate. According to a second embodiment of the invention, the search for a possible anomaly carries out a search for a possible anomaly on one of at least three sensors of interest, and furthermore comprises the following substep, when the presence of an anomaly on one of the sensors of interest is determined: 1d) identification of the sensor of interest having an anomaly, the step of merging measurements merging the measurements provided by: the sensors of interest, if it has been determined that there is no anomaly on one of said sensors of interest; the sensors of interest with the exception of a sensor of interest having an anomaly, if it has been determined that there is an anomaly on this sensor of interest. According to this second embodiment of the invention, the measurement merging step can provide an estimate of the parameter, calculated from the measurements of the parameter provided by: the sensors of interest, except for a sensor of interest having an anomaly , if the presence of an anomaly on this sensor of interest has been determined at each measurement instant for a predetermined duration, and; the sensors of interest, in other cases.
    [0011] Preferably, for each sensor of interest, the detection difference is proportional to said absolute value of a difference, divided by the standard deviation of a normal centered distribution characterizing, in the absence of anomaly on the sensors of interest, said difference between a measurement provided by this sensor of interest, and an estimate of the parameter calculated from the measurements provided by the other sensors of interest. For each detection difference, the predetermined threshold is advantageously determined from a desired probability of false alarm corresponding to the probability of determining, from this detection difference, the presence of an anomaly on one of the sensors. interest, while this anomaly does not exist. The step of identifying the sensor of interest having an anomaly preferably comprises: 1d1) for each sensor of interest, a calculation of an estimation of the parameter, from the measurements of this parameter provided by the other sensors of interest, the sensor of interest considered being excluded; 1d2) for each estimate thus calculated, a calculation of a residual which depends on the distances between said measurements provided by the other sensors of interest and said estimate; and 1d3) a search for a minimum residual, the excluded interest sensor associated with the minimum residue being the sensor of interest having an anomaly. Each of the parameter estimates can be a weighted average of measurements each provided by a sensor of interest. Each estimate of the parameter may be a weighted average of measurements each provided by a sensor of interest, each measurement provided by a sensor of interest being weighted by the inverse of the variance of a normal law characterizing a total error of relative measurement. auditory sensor of interest, in the absence of anomaly on this sensor of interest. The final estimate of the parameter may be a weighted average of measurements each provided by a sensor of interest, each measurement provided by a sensor of interest being weighted by a coefficient minimizing the value of the maximum error on the final estimate, introduced by an anomaly of a sensor of interest used to calculate the final estimate. The method according to the invention advantageously comprises a new step of searching for a possible anomaly, considering as new sensors of interest the sensors of interest taken into account to provide said final estimate.
    [0012] As a variant, the method according to the invention may comprise a new step of searching for a possible anomaly, considering as new sensors of interest the sensors of interest taken into account to provide said final estimate, as well as at least one interest sensor previously identified as having an anomaly.
    [0013] Preferably, the method according to the invention comprises a calculation of a precision error of the final estimate, comprising the following steps: a normal law is determined characterizing a total measurement error associated with the final estimate, in the no anomaly on the sensors of interest taken into account for the final estimate; a value of a probability is determined that the absolute value of the total measurement error associated with the final estimate is greater than a threshold to be determined; the value of said threshold is deduced from it, called precision error of the final estimate. The method according to the invention may comprise the following steps, for each sensor of interest: a probability of bad detection is fixed, common to all the sensors of interest, corresponding to the probability of determining the absence of an anomaly on one of the sensors of interest, while this anomaly exists on the sensor of interest considered; and - the value of the minimum detectable bias corresponding to the smallest bias introduced by an anomaly of the sensor of interest, on the measurement it provides, and making it possible to determine the presence of an anomaly on one of the sensors; of interest, with the probability of poor detection as fixed. A calculation of an error of integrity of the final estimate can then comprise the following steps: for each sensor of interest used to calculate the final estimate, calculation of an indicator proportional to the error introduced on the estimate final, by a failure of said sensor of interest, this failure being characterized by a bias corresponding to the minimum detectable bias; - determination of the largest indicator, named error of integrity of the final estimate. BRIEF DESCRIPTION OF THE DRAWINGS The present invention will be better understood on reading the description of exemplary embodiments given purely by way of indication and in no way limiting, with reference to the appended drawings in which: FIG. 1 schematically illustrates the context of FIG. application of the present invention; FIG. 2 illustrates an example of determining a normal law characteristic of the error on a measurement provided by a sensor; FIG. 3 schematically illustrates a method for merging measurements according to one embodiment of the invention; FIG. 4 schematically illustrates steps of a method according to one embodiment of the invention, making it possible to identify a sensor of interest having an anomaly; FIG. 5 schematically illustrates steps of a method according to one embodiment of the invention, making it possible to determine a precision error of an estimate; FIG. 6 schematically illustrates steps of a method according to an embodiment of the invention, making it possible to determine an integrity error of an estimate; Figure 7 schematically illustrates a total probability of false alarm and a total probability of poor detection; FIG. 8 schematically illustrates steps of a method according to one embodiment of the invention, implementing at least two successive steps of searching for a possible anomaly on a sensor of interest; and FIG. 9 illustrates an exemplary implementation of the method according to the invention. 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, its acceleration, its angular rotation speed, its trajectory angle, its heading angle or still his position. It may also be a temperature or pressure inside the aircraft or in the vicinity thereof. The invention is however 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.
    [0014] 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 to from the 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. Each sensor is for example an IRer (inertial reference system), an inertial sensor AHRS (Attitude & Heading Reference System), a satellite positioning sensor (GNSS, GPS or Galileo for example), an air sensor standard ADR ("Air Data Reference"), an LIDAR ADR laser anemometer, tachometer-type angular sensor, tachometer-based or odometer-based angular sensor, optical sensor (camera), height sensor conventional RA ("Radio Altimeter"), a laser height sensor RA LIDAR, a radio positioning sensor (for example VOR type, DME, ADF, Navy Radio, a radio landing sensor (for example ILS or MLS), a satellite landing sensor (SLS, GLS for example).
    [0015] FIG. 1 shows the context of application of the invention with the notations that will be used later. The parameter to be estimated is x (t), denoted x for the sake of simplification of the notations.
    [0016] At each moment, the sensors C1, C2, C3, C4 respectively acquire the measurements y1 (t), y2 (t), y3 (t), and y4 (t) of the parameter x. They then provide these measurements to a data fusion module 100. In order to lighten the notations, in the following, these measurements will be simply noted yi, y2, y3, and y4.
    [0017] The sensors can be of the same nature or of different natures. At least two sensors are provided. Nevertheless, advantageously, three or more sensors will be preferred. In the following, and in a nonlimiting manner, we will take the example of four sensors: two IRS sensors C1, C2, and two sensors AHRS C3, C4. The parameter x here corresponds to the attitude angles that are the angles of roll and pitch. The data merging module 100 receives as input the measurements yi, y2, y3, and y4. The data fusion module comprises a memory 101 receiving the predetermined thresholds Td1, Td2, Td3, Td4, respectively associated with the sensor C1, C2, C3, C4. From the measurements yi, y2, y3, and y4 and the predetermined thresholds Tm, Td2, Td3, Td4, the data fusion module provides an estimate x (t) of the parameter x, denoted simply 2. We will name in the following "Final estimate" the estimate of the x parameter output from the merge module.
    [0018] The method according to the invention is implemented by a data fusion module 100 comprising electronic means and computer and / or software means. The data fusion module comprises for example a microcontroller.
    [0019] The method according to the invention is carried out more generally within an assembly comprising the sensors C1, C2, C3, C4 and the data fusion module 100 in data communication with the sensors. The data merging module 100 is generally in communication with at least one system, here two avionic systems S1 and S2, to transmit them at each instant the final estimate 2. The avionics systems are for example a navigation system, a system autopilot, a system collecting data to be transmitted to the pilot, an engine control system, a flight control system, etc.
    [0020] Prior to the implementation of the method according to the invention, it will be possible to implement a calibration step aimed at determining, for each sensor, a normal law characteristic of a total error on the measurement provided by this sensor, when works normally. This normal law is determined from the characterization of the errors of the sensor.
    [0021] In order to facilitate the calculations, preference will be given to normal centered laws. It is always possible to characterize a sensor by a normal centered law. For example, if a sensor has a measurement bias, it is enough to subtract this bias to reduce to a centered law. Figure 2 illustrates an example of determining such a normal centered law. In the context of a normal operation of a sensor C ,, we put: Yi = x + ei (5) where y, is the measurement by the sensor C ,, x is the real value of the parameter, and e, is the total measurement error by the sensor C.
    [0022] For example, it is assumed that the error ei comprises a measurement error specific to the sensor erres i, and an error of installation of the sensor relative to the reference frame of the aircraft eins i. Depending on the case, one can consider other sources of error, for example an aerodynamic error in the case of an air sensor. The error erres i is defined by a normal law, also called Gaussian law, of mean iiin 'i and standard deviation o-in' i. This normal law is noted -W H - G-mes i; Gmes i) - The value of the standard deviation times i is known (for example 0.05 ° for an IRS sensor, and 0.5 ° for an AHRS sensor), as well as the interval bounding the values of / Lm , i (for example [-0.05 °, 0.05 °] for an IRS sensor, and [-0.5, 0.5 °] for an AHRS sensor). The error eins i is defined by a simple bias noted pins i. This error is due to imperfections in the installation of the sensor and other phenomena such as the deformation of the fuselage according to the flight phases. We have, for example, Pins i E [The total error ei is thus defined by a normal law of standard deviation times i and of mean (μ mes T = limes i Pins i- The mean of the total error ei is therefore bounded by the interval [-0.1 °, 0.1 °] for an IRS sensor, and [-0.55, 0.55 °] for an AHRS sensor, so that the algorithm corresponding to the process according to The invention remains valid whatever the actual distribution of the total error of measurement, we take a conservative hypothesis, in other words, we determine a normal centered law which makes a compromise between the different normal centered laws corresponding to the different possible values. The average of the total error of measurement is shown in Figure 2. The curve 21 represents the normal law of standard deviation times i and of average min Ointes T) , minimum value of the average [torus T - 0,05 '; 0.05 °] for an IRS sensor and for an AHRS sensor.
    [0023] Curve 22 represents the normal standard deviation law times, and average mean (μm 'T), mean value of the mean [tin' T. Curve 23 represents the normal law of standard deviation times, and of average max Ointes T), maximum value of the average r-mes T - The curve 24 represents the normal law of standard deviation a, and of average p., = 0, including the curves 21, 22 and 23. It s is therefore a normal centered law, denoted by "7"; o-,), whose standard deviation is defined so as to obtain a normal centered law which makes a compromise between the different normal centered laws corresponding to the different possible values of the average limes i ktins i- We have for example a, = 0.1 ° for an IRS sensor and a = 0.8 ° for an AHRS sensor. The values of a, (0.1 ° and 0.8 °) are determined theoretically and more particularly graphically, starting from FIG. 2. For this, a Gaussian curve (referenced 24) is plotted in such a way that it is always above the two extreme Gaussian curves (referenced 21 and 23) from a certain error value (about 2 ° for AHRS and about 0.2 ° for IRS), so from a certain probability value. Other theoretical methods could be used, for example a real-time error estimation method based for example on a Kalman filter. The values of a can also be determined conveniently, for example by flight tests. The values 0.1 ° and 0.8 ° for a, are given simply by way of example. These values are however realistic, with regard to the type of sensor considered (IRS or AHRS) and installations in an aircraft. Figure 3 schematically illustrates a method of merging measurements according to one embodiment of the invention.
    [0024] A preliminary step of acquiring measurements of the parameter by the sensors is assumed. Among the sensors C1, C2, C3, C4 sensors of interest are selected, among which we will look for a possible sensor faulty (that is to say having an anomaly). At least three sensors of interest are selected. It is assumed for example initially that all the sensors C1, C2, C3, C4 function normally although one of them could be brought to present an anomaly. The sensors of interest are then the four sensors C1, lo C2, C3, C4. For each sensor C1, C2, C3, C4, the value of a detection difference T1, T2, T3, respectively T4 (step 301, 302, 303, respectively 304) is calculated. The detection difference T is proportional to the absolute value of the difference between the measurement supplied by the sensor i and an estimation of the parameter, calculated from the measurements provided by the other sensors of interest (see formula (1) ). This difference is named useful difference. In the example detailed here, the detection difference T, is equal to the useful difference divided by the standard deviation of a normal law characterizing this useful difference in the absence of anomaly on the sensors of interest: T. = 137i.-x, 1 (6) aTi As detailed previously, the variable V3i = Yi - is defined, in the absence of anomaly on the sensors of interest, by a normal law whose variance is equal at 6Ti = 6i2 + 62i. The variable V3i corresponds to a normal centered law, which can be determined from the variances associated with the total measurement errors associated with each of the sensors of interest.
    [0025] It will be considered that the total measurement errors on the sensors are each characterized by a normal centered law, so that V3i is defined, in the absence of anomaly on the sensors of interest, by a normal centered law. The denominator o-T, allows the detection deviation Ti to be finally defined, for each sensor of interest i, by the same normal centered reduced law, of variance equal to unity, denoted .7 "0; 1). Thus, the detection deviation as defined in equation (6) is particularly advantageous because it allows all detection deviations Ti to be defined by the same normal centered law, namely a normal centered reduced law.
    [0026] Then, each detection difference T1, T2, T3, respectively T4 is compared with a corresponding predetermined threshold Tm, Td2, Td3, respectively Td4 (step 311, 312, 313, respectively 314). During a step 32, the presence or absence of an anomaly on one of the sensors of interest is determined according to the results of the comparisons. In particular, if there is at least one detection difference T1, T2, T3, respectively T4 which is greater than or equal to Tm, Td2, Td3 or Td4, it can be deduced that one of the sensors of interest has a anomaly. The invention thus realizes at this stage a failure or disturbance detection affecting one of the sensors of interest. The steps 30 ,, 31, and 32 together form a step of searching for a possible anomaly on one of the sensors of interest. We can note that if we have Ti> Tdi, this does not necessarily mean that it is the sensor of interest i that has broken down, because the value of Ti depends on yi, but also on 5c-j- ; which is determined by the measurements provided by the sensors of interest other than the sensor of interest i. In the following steps are presented to identify the sensor of interest having an anomaly. As previously presented, when the detection difference is defined by a normal law .N (I aTi), the corresponding predetermined threshold is Tou = Fjv - 1 (itTi; 6Ti) (1 - PFAi) (see equation (3) ).
    [0027] In the example presented here, each detection difference is defined by the reduced normal centered law N (0; 1). Thus, for each detection difference, the predetermined threshold is Td, = F17.7 (0; 1) (1 - PFA,). Preferably, a desired probability of identical false PFAI alarm is set for all the detection differences each associated with a sensor i. Thus, the predetermined threshold Td is the same for each detection deviation associated with a sensor i. The definition of the detection difference as given in equation (6) is therefore particularly advantageous since it makes it possible to consider a single predetermined threshold Td for all the detection deviations Ti according to the invention. The process is greatly simplified. If Ti> Td then it means that we accept the hypothesis H1 as presented in the introduction, that is to say that one of the sensors of interest has an anomaly (not necessarily the sensor j.). It is assumed that the determination of the predetermined thresholds Td, was carried out prior to the implementation of the method according to the invention. However, it can be provided that the method according to the invention includes this step of determining the predetermined threshold, for each detection difference. In this case, the fusion module 100 can be supplied with: the desired false alarm probability, associated with each detection difference; and the normal centered laws characterizing each detection difference, when none of the sensors of interest have an anomaly. All the estimates used according to the invention are advantageously weighted averages. Each sensor is associated with a specific weighting coefficient. The so-called intermediate estimates 5c-j-; are preferably weighted averages such that the weighting coefficient associated with a measurement by a sensor of interest is equal to the inverse of the variance of the normal law characterizing a total error of measurement relative to this sensor of interest in l absence of anomaly on this sensor of interest. This estimate is called the least squares method. The estimate of x from the measurements provided by the sensors of interest, except the sensor of interest i, is then defined as follows: =, n4 (7) M is the number of sensors of interest, and wi = Z (inverse of the variance). In other words, we try to minimize the following criterion: 2_, = argmin (X, 2 (x)) 2 where X (x) = Em 1 x (Yi-), which amounts to solving gradX, 2 (x) = Ô. 2 j # i The value of o-_, can be determined if necessary. For example, in the case 1 vm 1 of a least squares estimation, we have: 2 = Lj = 1 2 "j; i The least squares method, that is, a weighted average of which the coefficients are equal to the inverses of the variances, is a method known to be optimal for minimizing the difference between a measurement and the estimate of the measurement, since it minimizes the mean squared error, so the least squares method provides an estimate The method according to the invention then comprises a step 33 of merging measurements Step 33 corresponds to a calculation of the estimate x of the parameter x, called "final estimate" In the example illustrated in FIG. reference to the figures, there are four sensors of interest.
    [0028] If, after step 32, it is concluded that none of the sensors of interest has an anomaly, the final estimate 2 takes into account the measurements provided by all the sensors of interest. If, at the end of step 32, it is concluded that one of the sensors of interest has an anomaly, then identified this sensor as being the sensor j., The final estimate 2 takes into account the measurements provided. by all the sensors of interest except the sensor i. It can be considered that an exclusion of the sensor of interest with an anomaly is carried out, so that the measurements it provides are no longer taken into account for calculating the final estimate. This improves the tolerance of avionic systems especially to failures and disturbances of the sensors, since the systems receive a final estimate freed from the effect of these failures and disturbances. In the case (not shown) where there are only two sensors of interest, the final estimate 2 takes into account the measurements provided by all the sensors, even if it has been determined that one of the sensors exhibits anomaly. The final estimate is therefore at each moment "as accurate as possible", thanks to the real-time exclusion of a sensor identified as having an anomaly. Advantageously, at a given instant t, and after excluding a faulty sensor, one does not seek to exclude a second one. The exclusion of a possible second faulty sensor will then be performed at time t + 1. In practice, the probability that two sensors fail at exactly the same instant t is almost zero. The hypothesis according to which, at each instant t, there is at most one defective sensor, is therefore credible. This assumption includes the case where the failure of the two sensors is very slightly offset in time, for example one sensor fails at time t and the other sensor fails at time t + 1.
    [0029] The final estimate is preferably an estimate by the method of least squares vm: 2 = ~ j = 1Wj As stated above, the least squares method provides the optimum degree of precision. Where appropriate, other types of weighted averages using other coefficients, noted to replace wi = 2, may be used. For example, it may be possible to improve the integrity of the final estimate ( at the expense of its accuracy). The values of the Wi coefficients must be determined by justifying their validity. An example of such a determination is detailed in the following (new coefficients w'1, W'A).
    [0030] Thus, the estimates other than the final estimate will preferably be least squares estimates, while the final estimate may be a weighted average whose weighting coefficients are adjusted to favor a criterion other than the maximum accuracy. The measurement fusion method according to the invention is a centralized fusion, in which the different measurements are processed centrally (unlike a cascade merge where some data are pre-processed). The method for merging measurements according to the invention is also an instantaneous merger, in which each calculation takes into account only measurements at the same time t (as opposed to a recursive merge). Calculations implemented are simple, so fast execution and little consumption of computing resources. FIG. 4 schematically illustrates steps of a method according to an embodiment of the invention, making it possible to identify the sensor of interest having an anomaly, when the presence of an anomaly on the one of the sensors of interest.
    [0031] For each sensor of interest, an estimate .f1-1, respectively .f1-4, is calculated from the measurements of the parameter provided by the other sensors of interest {Y2, Y3Y4} '13 / 1Y3Y4MY1Y2, Y4} respectively {Y1, Y2, Y3}. These are the steps referenced 401, 402, 403, respectively 404.
    [0032] For example, each estimate is computed by the method of the least EITLi wiYi ji squares has the advantage of providing the optimal degree of precision. For each estimate thus calculated, a residual r1, r2, r3 or r4 is calculated, which depends on the distances between the estimate and each measurement provided by the said other sensors of interest. A distance here denotes a positive difference between two values. The residue depends in particular on: the absolute value of the difference between the estimate and each measurement provided by the said other sensors of interest; or - the square of the difference between the estimate and each measurement provided by said other sensors of interest. In Figure 4, the residues ri are denoted ri, and the measures y, are noted yi. The residue is for example a mean square deviation between said estimate and the measurements provided by said other sensors of interest. We have, for example: rl = 41 fr2 j = 1 Wi (Yi (8) j # i J j # i These are the steps referenced 411, 412, 413, respectively 414.
    [0033] This residual, based on least-squares estimates and on a mean squared difference calculation divided by the 012 variances, is a 5c-1-i squares. As mentioned above, the method of the least optimal method to minimize the probability of being mistaken in the exclusion of the sensor affected by an anomaly. If necessary, other residue calculations can be implemented, using other weighting coefficients.
    [0034] The smallest residue is then sought in absolute value (step 42). Let ri be the residue associated with the sensor of interest i, the sensor of interest having an anomaly is the sensor of interest associated with this smaller residue. In the example represented in FIG. 4, r2 is the smallest residue, so C2 is the sensor of interest that has an anomaly (sensor associated with the estimate x` = "2, itself associated with the sensor of interest The invention thus provides means for automatically detecting and identifying a fault affecting a sensor, and FIG. 5 schematically illustrates steps of a method according to the invention for determining a precision error. of the final estimate, called RPo.The precision error of the final estimate 2, also called "failure-free protection radius", refers to the error in calculating the final parameter estimate, assuming that the Measurements taken only from non-anomalous sensors are taken into account, followed by the example of a final least squares estimate, ie a weighted average the weights are equal to the inve A normal law characterizing a total measurement error associated with the estimation of the parameter (in the absence of anomaly on the sensors of interest taken into account for the final estimate) is determined. Given a least-squares estimation, the total error of measurement eT linked to this estimate is defined by: rD zq, c = iWkek eT = ^ Lk = iWk are D sensors of interest taken into account to calculate .2 , ek the total error of (where measurement related to the sensor k, and wk = (or ok is the standard deviation of the normal distribution crk centered characterizing the total measurement error of the sensor k in the absence of anomaly on If necessary, the coefficient wk is replaced by the weighting coefficient associated with the sensor k and used for the final estimation carried out in the method, In the following, for example, the determination of new values has been detailed. weighting coefficients, in an example in which the three sensors of interest are an IRS and two AHRSs, it follows that eT follows a normal centered distribution of standard deviation oT, denoted vp .7 "0; GT), with - a2 = z-ik = 1 2 "ak In order to determine this normal distribution, the memo is advantageously Figure 101 of the data merging module 100 includes the different values of a for each of the available sensors i. The value of the desired probability y is then set so that the absolute value of the total error of measurement eT is greater than a threshold £ to be determined. In other words, 1 - y is the probability that the error eT is included in the precision range [-E; We choose for example y = 5.10-2 (error at 2o), or y = 10-3 (error at 3o), or y = 10-5 (error at 4o). This condition is formalized as follows: 13 (- £ <eT <+ e) = 1 - y (10) We deduce (step 50) the value of the threshold E, which is the precision error of the final estimate. , also noted RP0 faultless protection radius. Indeed, we have: ## EQU1 ## with fu (0; T) the probability density of the law .7 "0; WG). hence: £ = F 0-, T) 1 (-1 2Y + F r (3; 6, T) (0)), (12) with F, ,, r (0;, T) the function of distribution associated with the probability density f ,, ,, r (0;, T) such that F r (3; 0-7,) (X) = fxco fi, r (o;, T) (t) dt. FIG. 6 schematically illustrates steps of a method according to the invention, making it possible to determine the error of integrity of the final estimate.
    [0035] The integrity error of the final estimate, also referred to as the "fail protection radius," refers to the error in calculating the parameter estimate, assuming that one of the sensors whose measurement was taken account has an anomaly. This is particularly the error on the final estimate before a faulty sensor has been identified and excluded.
    [0036] For each detection gap, a single probability of PMI detection is fixed). A bad detection corresponds to the probability of determining, from the comparisons between the detection deviations T, and the thresholds Tdi, the absence of an anomaly whereas this anomaly exists on the sensor i = j.
    [0037] For each sensor of interest j, we deduce the value of the smallest borin bias introduced by an anomaly of the sensor of interest j on the measurement it provides, making it possible to detect the presence of an anomaly with the probability bad detection as fixed. These steps are referenced 601, 602, 603, respectively 604, and make it possible to determine the minimum detectable biases bmin1, bmin2, bmin3, respectively bmin4. The determination of borin j implements a determination step, for each detection deviation Ti, of a normal law followed by it when the sensor of interest j has an anomaly.
    [0038] At a time t, an anomaly on the sensor is manifested by an offset of the measurement with respect to an expected measurement. Thus, in the case where the sensor of interest] breaks down, the fault introduces the bias b on the measure y: yi = x + ei + b (13) In order to simplify the calculations, the following example is continued: , in the absence of anomaly, each sensor has an error characterized by a normal centered law. The variable V3, as defined above (equation (2)) is the sum of the variable V1, which then follows a normal law of mean b noted .W (bi, garlic and variable V2, which always follows the normal law centered .W (0, Let T- = 3 / i- (Ti as defined in equation (6)), we deduce that Ti then follows a non-centered reduced normal law of mean -bi, denoted 0 The skilled person will readily be able to determine the normal law followed by Ti for other definitions of Ti such as T] oc. note: H0 the assumption that the set of interest sensors is working correctly, H1 the hypothesis that one of the sensors of interest has an anomaly, and Td, the value of the predetermined threshold (linked to the Detection deviation Ti), such that T,> Td, implies that we accept the hypothesis H1 (and we reject the hypothesis H0) .The probability of mis-detection then corresponds to the probability of rejecting hypothesis H1, while hypothesis H1 is true. To calculate borin j, we consider in particular that the probability of bad detection is related to T- and the sensor], and corresponds to the probability of CrTi reject hypothesis H1, while hypothesis H1 is true and concerns the sensor j. In other words, the probability of verifying 7 ', <7', /, for any i, whereas a sensor i = j has an anomaly, that is to say that the detection difference (corresponding is characterized by a normal law .7V 'b -, 1. In order to simplify 0-T the calculations, we suppose that 7', /, = 1PMD = P [Ti Tc117) -J "bmini)] Td PMD = f fbmin (chisel 1) (t) Read T PMD = F.7 r bmin (Ta) (1) crT. Thus by fixing the value of Pym and knowing that of Td, we can calculate bm, 'j. We can in the same way calculate all the different bmin In order to determine the normal law .7 "bmin; , 1), 0-T is predicted. advantageously, the memory 101 of the data fusion module 100 comprises the different values of o 'for each of the available sensors i. Each step 601, 602, 603, respectively 604 is followed by a step 611, 612, 613, respectively 614 of calculation of an indicator Ica, 412, 413, respectively Id4. The indicator In, Ide, Id3, respectively Id4 corresponds to the error introduced on the final estimate 2, by a failure (not detected) of the sensor of interest C1, C2, C3, respectively C4, this failure being characterized by a bias corresponding to the minimum detectable bias. In a step 62, the largest indicator is sought, which corresponds to the integrity error of the final estimate 2, denoted RP1. The error in the integrity of the final estimate 2, also called "protection radius with failure", refers to the error in the calculation of the final parameter estimate, assuming that a sensor used is affected by a failure not detected by the method described above.
    [0039] Thus, for each sensor of interest j used to calculate the final estimate 2, corresponds: the detection deviation Ti, the minimum detectable bias bmin j, and the indicator / di which corresponds to the error introduced on the final estimate 2, by a failure of the sensor of interest j, this failure being characterized by said minimum detectable bias. The final estimate is made here by the least squares method.
    [0040] The total error of measurement eT linked to such an estimate is defined by vp eT = `k = D1 '(see equation (9)) The indicator Id, then, is worth: Idi = wibmin i (14) In others In other words, the indicator / d, is equal to the minimum detectable chisel bias, weighted by a weighting coefficient equal to the inverse of the variance of the normal centered law corresponding to the sensor of interest i, divided by the sum of the inverses of the variances of the normal centered laws associated with each sensor of interest used to calculate the final parameter estimate.
    [0041] If necessary, the coefficients wk, w are replaced by the weighting coefficients associated with the sensor k, respectively i, used for the final estimation implemented in the method. For example, four sensors of interest formed by two IRS sensors and two AHRS sensors, (index I corresponds to an IRS sensor and index A corresponds to an AHRS sensor): = W (11) min / wAbmin AI d1, and -, 2 ', w / -I-wA) 2 (, wi- + wA) "Then: RP1 = max (1.11, 1, m).
    [0042] We see that if several sensors of interest are characterized by the same normal centered law, we reduce the number of calculations necessary to determine all Idi indicators. This remark concerns all the treatments described throughout the text. Advantageously, the memory 101 of the data fusion module 100 comprises the different values of o- 'for each of the available sensors. In some cases, it may be advantageous to adjust the weighting coefficient associated with each measurement in the estimates implemented according to the invention, in particular in the final estimate of the parameter x. Indeed, if one carries out a estimation by the least squares method, the weight corresponding to each sensor of interest is inversely proportional to its variance.
    [0043] The measurement provided by a low variance interest sensor therefore strongly influences the value of the final estimate. In other words, the degree of confidence given to a low variance interest sensor is high. The disadvantage is that if this sensor of interest breaks down, the error on the final estimate is high, as long as we have not excluded this sensor of interest. In other words, the integrity error of the final estimate is high (protection radius with failure). One can consider adjusting the weight given to a sensor of interest to reduce the integrity error of the final estimate. This results in an increase in the precision error of the final estimate, which is compensated for by the reduction of the integrity error. In particular, it will be possible to define the weighting coefficients so as to obtain the minimum integrity error. In other words, we are looking for the weighting coefficients minimizing the value of the maximum error on the final estimate, introduced by a failure of a sensor of interest used to calculate the final estimate. For example, in the case where the sensors of interest consist of two AHRS sensors and one IRS sensor, the IRS sensor is the only sensor whose error follows a normal centered distribution of low standard deviation. If a least squares estimation is used, the weight associated with the IRS sensor for the estimate is high. Consequently, in case of failure of this IRS sensor, the error on the estimate x is very high as long as the sensor is not excluded. Thus, the indicator Idi as defined above, associated with the IRS sensor, takes a very high value, and defines the integrity error of the estimate. This integrity error is even higher, as the minimum detectable bias associated with the IRS sensor is higher in the presence of a single IRS sensor and two AHRS sensors, than in the presence of two IRS sensors and two AHRS sensors. We therefore seek two new weighting coefficients W'A for the IRS sensor and for the two AHRS sensors, which minimize the maximum error introduced on the final estimate 2, by a failure of a sensor of interest. We have: dl WIAtiminA W111) min / .., and I 2 (., 2 14 // i- -EW / A) 14 /// -EWA) "We search for the pair () which minimizes the maximum among / di and IcIA- These weighting coefficients then make it possible to define an estimate making it possible to obtain the minimum integrity error (maximum integrity): w, cr + w, AYAi + w, AYA2 x- w'rEw'A + w A with y, the measurement by the IRS sensor, ym the measurement by the first AHRS sensor and yA2 the measurement by the second AHRS sensor.
    [0044] Note that such an estimate has a higher accuracy error than the Least Squares estimate. In other words, the final estimate is a weighted average in which the weights are not variances. We assign a lower weight to the IRS.
    [0045] Thus, the ability to detect a future anomaly on this IRS is favored, with however a slight deterioration in the accuracy of the final estimate of the parameter. As an illustration, there is shown in Figure 7 a total probability of false alarm and a total probability of poor detection. The curve 71 represents the normal centered law corresponding to the measurement error on the overall estimate, when all the sensors of interest taken into account for the estimation function normally. The curve 73 represents the normal law corresponding to the measurement error on the overall estimate, when one of the sensors of interest taken into account for the estimation has an anomaly. The horizontal straight line 72 defines under the curve 73, to the left of the line 72, an area representative of the probability of poor detection. Line 72 also defines, under curve 71, to the right of line 72, an area representative of the total probability of false alarm. FIG. 8 schematically illustrates steps of a method for merging measurements according to the invention, implementing at least two successive steps of searching for a possible anomaly on a sensor of interest.
    [0046] The sensors of interest are denoted C. The fusion module performs a monitoring function referenced 814, consisting of searching, at each instant t, for a possible anomaly on one of the sensors of interest C. When one of the sensors of interest is identified as having an anomaly, the monitoring function 814 also makes it possible to determine the sensor of interest Cp which has an anomaly. An exclusion function 804 makes it possible to exclude the sensor Cp from the group of the sensors of interest.
    [0047] The fusion module (see FIG. 1) takes into account the sensors of interest C ,, except for a possible excluded sensor Cp, to calculate a final estimate x of the parameter x at the instant t (function 834 called the combination or consolidation or merger of measures). The fusion module also calculates the precision error RP0 at time t and the integrity error RP1 at time t, associated with the final estimate (function referenced 824). These different values are supplied to a system, including an avionics system. It is advantageously provided that the fusion module calculates the precision error RP0 and the integrity error RP1 only when at least three sensors of interest are available.
    [0048] For the identification and exclusion functions to be of real interest, it is advantageous to always have at least three sensors of interest. It is preferably provided not to implement the identification and exclusion functions when only two sensors of interest are available. There may be a step of transmitting to a system, a message signaling the fact that there are only two sensors of interest taken into account for the final estimate, to warn the pilot of the aircraft or a maintenance center. The exclusion can be immediate, as soon as it has been determined that the sensor of interest Cp has an anomaly. Alternatively, the exclusion may depend on an exclusion confirmation duration. In this case, the sensor Cp is excluded if it is identified as having a fault, at any time during the exclusion confirmation period. It is possible to provide a step of transmitting to a system, an identifier of the sensor Cp, in order to warn the pilot of the aircraft or a maintenance center that the sensor Cp has failed.
    [0049] As the sensor Cp is excluded, it is no longer taken into account to calculate a final estimate 2 of the parameter x, as well as the precision error and the integrity error of this final estimate. In addition, following this exclusion, the monitoring function will only monitor the remaining sensors of interest. Several successive exclusions can be provided in the event of successive failures of the sensors. It is possible to reinject into the group of interest sensors C 1, the sensor or sensors previously identified as having an anomaly. If the exclusion function does not perform a new sensor exclusion, it means that all the sensors are functioning normally again. We can thus ensure that we take into account at each moment a maximum number of sensors operating normally. This characteristic is particularly advantageous in combination with a maintenance step on the sensor identified as having an anomaly. Preferably, all the sensors previously identified as having an anomaly are reinjected, because it is possible that several sensors return at the same time to the normal state. In the same way as for the exclusion of a sensor, a reinjection confirmation duration can be provided. A sensor is again taken into account for the calculation of the final estimate of the parameter x if it is not considered to be abnormal, at any time during the reinjection confirmation period. The reinjection of sensors previously identified as having an anomaly is of increased interest when only two sensors of interest are available, and the presence of an anomaly on one of these two sensors has been detected. interest. In this case, it is impossible to identify and then exclude the sensor of interest having an anomaly. However, it is possible to reinject the sensors of interest so that at least three sensors of interest are available, thus making it possible to identify and exclude the one that has an anomaly.
    [0050] It can be provided that the exclusion is immediate, and that the reinjection is related to a reinjection confirmation period, in order to avoid instability at the limit of the anomaly situation. As already stated above, it is assumed that at each moment t, there is at most one faulty sensor. At a given instant t, and after excluding a faulty sensor, we do not try to exclude a second one before calculating the final estimate. The exclusion of a possible second faulty sensor will then be performed at time t + 1. The method of fusion of measurements according to the invention has the advantage of allowing a rigorous characterization of the final estimate made. It is indeed possible to easily determine a precision error and / or an integrity error of the final estimate. In addition, the probability of false alarm related to the final estimate is known, as well as a probability of false detection related to the calculated integrity error. Depending on its needs, for example precision and / or integrity, each system may or may not use the final estimate provided at a time t. Thanks to the exclusion of faulty sensors and to a rigorous characterization of the final estimate, estimates are easily available with a certain accuracy error and / or a low integrity error. Systems requiring a certain level of precision and / or integrity may use these final estimates. Thus, the invention increases the availability of the final estimates that can be provided to the systems. We can then consider new functions through the provision of these final estimates certainly presenting a high accuracy (low accuracy error) and / or high integrity (low integrity error). Since the precision and / or the integrity of the final estimates is controlled, it is no longer necessary to oversize the number and / or the technical complexity of the sensors used.
    [0051] FIG. 9 illustrates an exemplary implementation of the method according to the invention, made from actual flight data. The abscissa axis is time. The y-axis corresponds to the pitch angle value in degrees.
    [0052] The sensors of interest are two IRS sensors and two AHRS sensors. The curve 91 represents the value measured by a first IRS sensor, which exhibits a failure characterized by a ramping bias between the times t = 4000 and t = 4200. Curve 92 represents the value measured by the second IRS sensor, which exhibits a breakdown characterized by a constant bias from time t = 4150. The other curves are very close and represent the values measured by the two AHRS sensors, the real value, and the value estimated using the method according to the invention, implementing an estimation by the least squares method and the search. , the identification and exclusion of a sensor with a failure. It can be seen that the estimated value remains very close to the actual value, even when one of the IRS sensors fails, and even when the two IRS sensors fail successively.20
    权利要求:
    Claims (15)
    [0001]
    REVENDICATIONS1. A method of merging measurements of a parameter (x), in particular a flight parameter of an aircraft, from measurements (yi, y2, y3, y4) of this parameter respectively provided by a plurality of sensors (Ci, C2, C3, C4), characterized in that it comprises the following steps and sub-steps: 1) search for a possible anomaly on one of at least two sensors called sensors of interest, including the substeps following: la) for each sensor of interest (Ci; C2; C3; C4), calculation (301; 302; 303; 304) of a so-called detection difference (T1; T2; T3; T4), proportional to the an absolute value of a difference between a measurement (y1; V2; V3; y4) provided by this sensor of interest and an estimate of the parameter (2-1-.1;
    [0002]
    2-1-2; 2-1-.3; 2-1-4. ) calculated from the measurements provided by the other sensors of interest; lb) comparing (311; 312; 313; 314) each detection deviation to a corresponding predetermined threshold (Tm; Td2; Td3; Td4); 1c) according to the result of the comparisons, determining (32) the presence or absence of an anomaly on one of the sensors of interest with a total probability of false alarm determined (PFA); and 2) merging (33) measurements provided by the sensors of interest, to provide an estimate of the parameter (2), called the final estimate. 2. Method according to claim 1, characterized by a search for a possible anomaly on one of two sensors of interest (Ci; C2; C3; C4), and a fusion of the measurements provided by these two sensors of interest , to provide an estimate of the parameter (2), called the final estimate.
    [0003]
    3. Method according to claim 1, characterized in that the search for a possible anomaly carries out a search for a possible anomaly on the unparmi at least three sensors of interest (C1; C2; C3; C4), and comprises in addition, the following sub-step, when determining the presence of an anomaly on one of the sensors of interest: 1d) identifying the sensor of interest having an anomaly, the step of merging measurements performing a merging the measurements provided by: the sensors of interest (C1, C2, C3, C4), if it has been determined that there is no anomaly on one of said sensors of interest; the sensors of interest with the exception of a sensor of interest having an anomaly, if it has been determined that there is an anomaly on this sensor of interest.
    [0004]
    4. Method according to claim 3, characterized in that the step of merging measurements provides an estimate of the parameter (2), calculated from the measurements of the parameter provided by: the sensors of interest (C1, C3, C4) , except for a sensor of interest having an anomaly, if the presence of an anomaly on this sensor of interest has been determined, at each instant of measurement for a predetermined duration, and; the sensors of interest, in other cases.
    [0005]
    5. Method according to any one of claims 1 to 4, characterized in that for each sensor of interest (C1; C2; C3; C4), the detection difference (T1; T2; T3; T4) is proportional to said absolute value of a difference, divided by the standard deviation (o-Ti) of a normal centered law characterizing, in the absence of anomaly on the sensors of interest, said difference between a measurement (y1 yz; y3; y4) provided by this sensor of interest, and an estimate of the parameter (2-1-1; 2-1-2; 2-1-3;) calculated from the measurements provided by the other sensors interest.
    [0006]
    6. Method according to any one of claims 1 to 5, characterized in that for each detection difference, the predetermined threshold (Tm, Td2, Td3, Td4) is determined from a desired probability of false alarm corresponding to the probability of determining, from this detection difference, the presence of an anomaly on one of the sensors of interest, whereas this anomaly does not exist.
    [0007]
    7. Method according to any one of claims 3 to 6, characterized in that the step of identifying the sensor of interest having an anomaly comprises: 1c11) for each sensor of interest (C1; C2; C3; C4 ), a calculation (401; 402; 403; 404) of an estimate of the parameter (.f1-1; .f1-2; x` = "3; .C.4), from the measurements of this parameter provided by the other sensors of interest, the sensor of interest considered being excluded; 1d2) for each estimate thus calculated, a calculation (411; 412; 413; 414) of a residue (r1; r2; r3; r4) which depends on the distances between said measurements (y2, y3, y4, y1, y3, y4, y1, y2, y4, y1, y2, y3) provided by the other sensors of interest and said estimate (x` = "1; f1-2; .t "=" 3; .f1-4.); and 1d3) a search (42) of a minimum residue, the excluded interest sensor associated with the minimum residue being the sensor of interest having an anomaly.
    [0008]
    8. Method according to any one of claims 1 to 7, characterized in that each of the estimates of the parameters (-. ') C; .f1-1; .f1-2; x` = "3; .f1-4) is a weighted average of measurements each provided by a sensor of interest.
    [0009]
    9. The method according to claim 8, characterized in that each estimate (2; 2-1-1; fl- -2; fl- -3; fl- -4) of the parameter is a weighted average of measurements provided. each by a sensor of interest, each measurement (y1; y2; y3; y4) provided by a sensor of interest being weighted by the inverse of the variance of a normal law characterizing a total measurement error relating to said sensor of interest. interest, in the absence of anomaly on this sensor of interest.
    [0010]
    10. Method according to claim 8, characterized in that the final estimate of the parameter (2) is a weighted average of measurements each provided by a sensor of interest, each measurement (y1, y2, y3, y4) provided by a a sensor of interest being weighted by a coefficient minimizing the value of the maximum error on the final estimate, introduced by an anomaly of a sensor of interest used to calculate the final estimate.
    [0011]
    11. Method according to any one of claims 1 to 10, characterized in that it implements a new step of searching for a possible anomaly, considering as new sensors of interest the sensors of interest taken into account. account for providing said final estimate (2).
    [0012]
    12. Method according to any one of claims 1 to 10, characterized in that it implements a new step of searching for a possible anomaly, considering as new sensors of interest the sensors of interest taken into account. to provide said final estimate (2), as well as at least one sensor of interest previously identified as having an abnormality.
    [0013]
    13. Method according to any one of claims 8 to 12, characterized by a calculation of a precision error of the final estimate (RP0), comprising the following steps: - a normal law is determined characterizing a total error of measurement associated with the final estimate, in the absence of anomaly on the sensors of interest taken into account for the final estimate; a value of a probability (y) is fixed that the absolute value of the total measurement error associated with the final estimate (2) is greater than a threshold to be determined; - We deduce the value of said threshold, called accuracy error of the final estimate (RPo).
    [0014]
    14. Method according to any one of claims 8 to 13, characterized in that, for each sensor of interest: - one fixes a probability of poor detection (PMD), common to all the sensors of interest, corresponding to the probability of determining the absence of an anomaly on one of the sensors of interest, whereas this anomaly exists on the sensor of interest considered; and - the value of the detectable minimum bias is deduced therefrom; bmin2; bmin3; bmin4) corresponding to the smallest bias introduced by an anomaly of the sensor of interest, on the measurement it provides, and making it possible to determine the presence of an anomaly on one of the sensors of interest, with the probability of bad detection (PMD) as fixed.
    [0015]
    15. The method of claim 14, characterized by a calculation of an integrity error of the final estimate (RP1), comprising the following steps: for each sensor of interest used to calculate the final estimate (2) , Calculating an indicator proportional to the error (611; 612; 613; 614) introduced on the final estimate, by a failure of said sensor of interest, this failure being characterized by a bias corresponding to the minimum detectable bias - determination (62) of the largest indicator, called error of integrity of the final estimate (RP1).
    类似技术:
    公开号 | 公开日 | 专利标题
    FR3017705A1|2015-08-21|METHOD OF MELTING SENSOR DATA.
    EP2069818B1|2009-12-09|Method and device for mojnitoringthe integrity of information provided by a hybrid ins/gnss system
    EP1914561B1|2017-09-06|Method of satellite location of a vehicle and guarantee of integrity with selection of a subgroup of satellites
    EP2648017B1|2014-09-24|On-board system for piloting an aircraft, based on a GNSS system, with redundant, dissimilar architecture for high level of integrity
    FR2963436A1|2012-02-03|METHOD FOR DETERMINING A VOLUME OF PROTECTION IN THE CASE OF TWO SIMULTANEOUS SATELLITE FAILURES
    EP2598911B1|2014-06-18|Method for detecting and excluding multiple satellite failures
    FR2866423A1|2005-08-19|Position and speed information integrity monitoring device for hybrid system, has test circuits for comparing states of filters and for detecting satellite fault when distance between receiver position points is greater than threshold
    EP2318854B1|2012-10-03|Method for protecting a radio navigation receiver user against aberrant pseudo-range measurements
    FR3013834A1|2015-05-29|METHOD FOR MERGING SENSOR DATA USING A COHERENCE CRITERION
    CN103675854A|2014-03-26|Multichannel time service fitting method based on BeiDou satellite
    EP2411832B1|2013-06-05|Method and device for detecting and excluding satellite malfunctions in a hybrid ins/gnss system
    FR3007841A1|2015-01-02|METHOD FOR DETECTING A FAILURE OF AT LEAST ONE SENSOR PRESENTED ON AN AIRCRAFT, USING A BARO-INERTIAL LOOP AND ASSOCIATED SYSTEM
    CN111060133A|2020-04-24|Integrated navigation integrity monitoring method for urban complex environment
    EP2804016A1|2014-11-19|Improved method for determining the position and/or the speed of a guided vehicle; related system
    FR2916290A1|2008-11-21|SYSTEM FOR SELECTING DATA REPRESENTATIVE OF AN AIR PARAMETER, METHOD AND COMPUTER PROGRAM THEREOF
    FR2816703A1|2002-05-17|INERTIAL NAVIGATION CENTER COMPRISING AN INTEGRATED GPS RECEIVER
    FR2951535A1|2011-04-22|METHOD OF DETECTING PARASITE MOVEMENTS DURING THE ALIGNMENT OF AN INERTIAL POWER PLANT
    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
    EP2799890B1|2016-01-20|Method and system for determining the speed of an aircraft relative to the air
    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
    FR2977314A1|2013-01-04|NAVIGATION DEVICE AND METHOD INTEGRATING MULTIPLE INERTIAL HYBRID NAVIGATION SYSTEMS
    EP3553617B1|2022-02-16|System and method of fusing measurements of aircraft flight parameters
    FR2905006A1|2008-02-22|METHOD OF MONITORING THE INTEGRITY OF A PLANE POSITION CALCULATED ON BOARD
    EP1752786B1|2013-11-20|Hybrid inertial/satellite navigation system and method for the operation of such a system
    EP1956386A1|2008-08-13|Method for determining the position of a moving body and a protection limit around this position
    同族专利:
    公开号 | 公开日
    CN104848873B|2019-08-16|
    CN104848873A|2015-08-19|
    US10041808B2|2018-08-07|
    FR3017705B1|2017-07-07|
    US20150233730A1|2015-08-20|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US5661735A|1994-12-27|1997-08-26|Litef Gmbh|FDIC method for minimizing measuring failures in a measuring system comprising redundant sensors|
    US20080276155A1|2007-05-03|2008-11-06|Chung-An University Industry Academic Cooperation Foundation|Method of detecting and isolating fault in redundant sensors, and method of accommodating fault in redundant sensors using the same|
    US7130772B2|2004-04-07|2006-10-31|United Technologies Corporation|Method and apparatus for estimating a parameter based on a plurality of redundant signals|
    CN101807047B|2010-03-19|2011-11-09|北京航空航天大学|Method for predicting fault of nonlinear system based on fuzzy parity equation and AR model|
    CN102176159B|2011-02-28|2013-05-01|哈尔滨工业大学|Satellite attitude control system failure diagnosis device and method based on state observer and equivalent space|FR3000196B1|2012-12-21|2015-02-06|Airbus Operations Sas|DEVICE FOR PROVIDING NAVIGATION PARAMETER VALUES OF A VEHICLE|
    DE102014211168A1|2014-06-11|2015-12-17|Continental Teves Ag & Co. Ohg|Method and system for verification of measurement data|
    CN105825720B|2016-04-11|2018-03-30|成都民航空管科技发展有限公司|A kind of optimization method for effectively reducing short-term flight collision alarm false-alarm|
    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|
    US11062533B2|2016-08-05|2021-07-13|Siemens Energy, Inc.|Universal high redundancy sensor interface for low cost sensing|
    DE102017201548A1|2017-01-31|2018-08-02|Siemens Aktiengesellschaft|Method for providing measured values of a technical installation, technical system and method for operating the technical system|
    US10373474B2|2017-03-09|2019-08-06|Keithley Instruments, Llc|Auto setting of alarm limits|
    FR3066755B1|2017-05-23|2019-06-07|Airbus Operations|METHOD AND DEVICE FOR MONITORING AND ESTIMATING PARAMETERS RELATING TO THE FLIGHT OF AN AIRCRAFT.|
    CN107749072B|2017-09-21|2021-01-19|黄仁杰|Unmanned aerial vehicle aerial image calibration method suitable for slope surface measurement|
    US20190108760A1|2017-10-10|2019-04-11|Honeywell International Inc.|System and method for developing and maintaining temperature-compensated altitude information|
    CN108762306A|2018-06-22|2018-11-06|深圳市科比特航空科技有限公司|A kind of sensor-based system of unmanned plane and unmanned plane|
    CN109978010B|2019-04-10|2021-03-05|广东省气象探测数据中心|Fusion method of multiple rainfall sensors|
    CN111225440A|2019-11-22|2020-06-02|三一重工股份有限公司|Cooperative positioning method and device and electronic equipment|
    CN111323069B|2020-03-23|2021-08-24|清华大学|Multi-sensor online calibration method and system based on deep reinforcement learning|
    CN111736626A|2020-06-22|2020-10-02|中国人民解放军国防科技大学|Stable missile path data processing method|
    FR3113518A1|2020-08-24|2022-02-25|Arianegroup Sas|Failure determination method for a set of sensors|
    法律状态:
    2016-02-18| PLFP| Fee payment|Year of fee payment: 3 |
    2017-02-17| PLFP| Fee payment|Year of fee payment: 4 |
    2018-02-23| PLFP| Fee payment|Year of fee payment: 5 |
    2020-02-19| PLFP| Fee payment|Year of fee payment: 7 |
    2021-02-24| PLFP| Fee payment|Year of fee payment: 8 |
    2022-02-16| PLFP| Fee payment|Year of fee payment: 9 |
    优先权:
    申请号 | 申请日 | 专利标题
    FR1451290A|FR3017705B1|2014-02-18|2014-02-18|METHOD OF MELTING SENSOR DATA.|FR1451290A| FR3017705B1|2014-02-18|2014-02-18|METHOD OF MELTING SENSOR DATA.|
    US14/623,085| US10041808B2|2014-02-18|2015-02-16|Method of sensor data fusion|
    CN201510086861.0A| CN104848873B|2014-02-18|2015-02-17|The method of Data Fusion of Sensor|
    [返回顶部]