(19)
(11)EP 2 820 489 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
09.10.2019 Bulletin 2019/41

(21)Application number: 13706513.2

(22)Date of filing:  26.02.2013
(51)International Patent Classification (IPC): 
G05B 23/02(2006.01)
G07C 5/02(2006.01)
B64D 45/00(2006.01)
G01D 9/00(2006.01)
G07C 5/08(2006.01)
(86)International application number:
PCT/EP2013/053798
(87)International publication number:
WO 2013/127781 (06.09.2013 Gazette  2013/36)

(54)

METHOD OF ANALYSING FLIGHT DATA

VERFAHREN ZUR ANALYSE VON FLUGDATEN

PROCÉDÉ D'ANALYSE DE DONNÉES DE VOL


(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

(30)Priority: 29.02.2012 FR 1251870
29.03.2012 US 201261617601 P

(43)Date of publication of application:
07.01.2015 Bulletin 2015/02

(73)Proprietors:
  • Safran Electronics & Defense
    92100 Boulogne-Billancourt (FR)
  • Université de Technologie de Troyes
    10000 Troyes (FR)

(72)Inventor:
  • CHRYSANTHOS, Nicolas
    F-92100 Boulogne Billancourt (FR)

(74)Representative: Regimbeau 
20, rue de Chazelles
75847 Paris Cedex 17
75847 Paris Cedex 17 (FR)


(56)References cited: : 
US-A1- 2008 091 630
US-B1- 6 937 924
  
      
    Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


    Description

    GENERAL TECHNICAL FIELD



    [0001] The invention relates to a method of analysing a set of flight data recorded during at least one flight of at least one aircraft.

    PRIOR ART



    [0002] The regulations in terms of maintenance and air traffic define standards that airline companies are required to comply with in order to ensure to a user a maximum level of safety.

    [0003] In order to optimise the maintenance phases, airline companies have equipped themselves with flight data analysis systems.

    [0004] Flight data analysis systems known by the name FDM (Flight Data Monitoring) or instead FOQA (Flight Operational Quality Assurance) are known. These systems consist in equipping an aircraft with a flight data recorder. Such a recorder is for example a black box or instead a specific recorder such as an ACMS (Aircraft Condition Monitoring System).

    [0005] These systems enable airline companies to understand in detail the course of a flight from regular recordings of the values of these flight data made during each flight of each of their airplanes.

    [0006] To do this, these systems detect predefined events occurring during the flight and an expert then analyses these events, which indicate that a technical incident has occurred during the flight, that a practice or a condition provided by a flight procedure has not been complied with, thus giving warning at a very advanced stage of any incidents or accidents that could arise.

    [0007] These techniques necessitate predefining rules for detecting events, usually defined as exceeding thresholds of one or more parameters, which may trigger alerts so that the expert analyses more closely the flight.

    [0008] A problem is that these techniques do not make it possible to detect singular events beyond the predefined rules, which may lead to a non detection of an abnormal flight.

    [0009] US2008/091630 A1 discloses a method for defining normal operating regions and identifying anomalous behaviour of units within a fleet, operating in a complex, dynamic environment.

    DESCRIPTION OF THE INVENTION



    [0010] An aim of the invention is to make it possible to detect abnormal flights without having the need to define detection rules.

    [0011] To this end, the invention proposes a method of analysing flight data recorded during N flights of at least one aircraft, the data being grouped together by flight i in a signature vector of the flight Xi of size d, the components of which correspond to data recorded during said flight i of the aircraft, a flight i being thus defined by the signature vector Xi, the method comprising the following steps:
    • Gaussian kernel entropy component analysis of the flight signatures Xi to obtain a zone of normal flights and classifying the flight signatures Xi with respect to their distance to said zone;
    • determining, for each flight i, an abnormality score zi defined by the distance of a flight signature Xi with respect to the zone of normal flights;
    • detecting, as a function of the abnormality score zi, at least one abnormal flight; and
    • determining for each abnormal flight, a phantom flight the nearest to the abnormal flight while being in the zone of normal flights, wherein the parameters of the abnormal flight detected are compared with those of the phantom flight determined in order to detect at least one parameter of the abnormal flight which has rendered abnormal said abnormal flight detected.


    [0012] The invention is advantageously completed by the following characteristics, taken singly or in any technically possible combination thereof:
    • the Gaussian entropy component analysis comprises the following sub-steps:

      ∘ determination of a matrix of similarity K of size N × N, the components of which quantify the proximity between two flight signatures Xi;

      ∘ breakdown into eigen vectors of the matrix of similarity K to obtain N eigen vectors a1,...,aN and N eigen values λ1,...,λN such that ∀i = 1,...,N K · ai = λ · ai;

      ∘ determination, for each eigen vector, of its entropy coefficient γi

      ∘ selection of a sub-assembly of eigen vectors {am}m∈{1,...,N} such that the sum of the entropies γm is greater than the percentage of the sum of the N entropies γi.

    • the entropy coefficient is defined by

    • the abnormality score of a flight i is defined by:

    • the components of the matrix of similarity K are defined by:

      where σ2 is a predetermined bandwidth parameter.
    • the bandwidth parameter σ2 is determined in the following manner:

      ∘ a matrix D of size N × N is determined corresponding to the distance between two flights and is defined in the following manner

      ∘ each component of each column is arranged in ascending order to obtain a matrix D';

      k first lines of the matrix D' thereby obtained are selected and the others are eliminated to obtain a matrix D' of size k × N ;

      ∘ the average of each column of the matrix D' is determined to obtain N values y1,...,yN;

      ∘ the median absolute deviation of the values y1,...,yN defined by mad =mediane{|yi -med|} with med =mediane{y1,..., yN} is determined;

      ∘ the bandwidth parameter σ2 is determined from the median absolute deviation of the values y1,...,yN by the following functional

    • the data of a flight i are grouped together in a matrix Fi of dimension T×P with T the number of data recorded during the flight i and P the number of parameters recorded, the vector Xi having for components the columns of the matrix Fi end to end, the vector Xi being then of dimension d = T × P and is defined by

    • the data of a flight i are grouped together in a matrix Fi of dimension T × P with T the number of data recorded during the flight i and P the number of parameters recorded, the vector Xi having for components the columns of the matrix Fi sampled to select n < T recordings

      of parameters, the vector Xi being then of dimension d = n × P and is defined by

    • the data of a flight i are grouped together in a matrix Fi of dimension T × P with T the number of data recorded during the flight i and P the number of parameters recorded, the vector Xi having for components the average, the variance, the minimal value, the maximal value of a parameter among all the recordings of this parameter, the vector Xi being then defined by:



    [0013] The invention has numerous advantages.

    [0014] With the invention, the detection of abnormal flights is automatic and does not necessitate the intervention of an expert for said detection.

    [0015] With the invention the detection is implemented in a statistical manner while taking into account parameters. In this way, there is an automatic learning that takes place.

    [0016] The invention enables unexpected problems to be highlighted.

    [0017] The invention enables problems stemming from the contribution of several parameters to be detected.

    [0018] The method of the invention may be used by a non expert in statistics or data mining.

    [0019] Moreover, with the invention, the diagnosis of abnormal flights is simple because it makes it possible to create a reference flight, the phantom flight for each abnormal flight detected.

    DESCRIPTION OF FIGURES



    [0020] Other characteristics, aims and advantages of the invention will become clearer from the description that follows, which is purely illustrative and non limiting, and which should be read with reference to the appended drawings, in which:
    • figures 1a and 1b schematically illustrate steps of a method according to an embodiment of the invention;
    • figure 2 illustrates a representation of flight data according to an embodiment of the invention.


    [0021] In all of the figures, similar elements have identical numerical references.

    DETAILED DESCRIPTION OF THE INVENTION



    [0022] In a method of analysing flight data in a preliminary step (not detailed) flight data recorded during several flights made by at least one aircraft are recovered.

    [0023] These flight data correspond to flight parameters such as for example the speed, the altitude, the temperature, etc.

    [0024] Then, in a first step E1, it is advisable to extract for each flight i, among all the data recorded during the flight i, those that characterise the flight i in other words the signature of each flight i.

    [0025] For N ≥ 1 flights, there are a number P0 of parameters recorded over a flight time T0. By way of example, in certain recorders, it is possible to have up to P0 = 2000 parameters.

    [0026] It is considered that the parameters are recorded/sampled at the same frequency, if necessary, well known techniques of oversampling and linear approximation are implemented on the data.

    [0027] To extract E1 the data characteristic of a flight i, according to one embodiment, a restriction is made to a flight phase (for example, the landing) of time T < T0 and only a restricted number P < P0 of pertinent parameters are retained for the analysis of the data of this flight i. By way of example, it is possible to make a restriction to T = 1000 seconds and P = 30parameters.

    [0028] Hereafter, the value of each data of the flight i is designated in the following manner: Fi(t = 5, p = 1) in other words the value of the first parameter recorded at the 5th second (if the data are sampled in seconds) for the flight i.

    [0029] Then, after the extraction E1, in a signature vector Xi of a flight i of dimension d, the data is grouped together E2.

    [0030] Each signature vector Xi thus contains the set of pertinent information relative to the flight i.

    [0031] The grouping together E2 may be implemented according to three embodiments.

    [0032] According to a first embodiment, the grouping together E2 in the vector consists in keeping all of the data. This is then known as an exhaustive approach.

    [0033] According to this first embodiment, to obtain a vector Xi it is advisable to take the matrix Fi and to stick columns one to one.

    [0034] The vector Xi is then of dimension d = T × P and is defined by:



    [0035] For example if T = 1000 and P = 30 then there is a vector Xi of size 30000 (and not a matrix of size 1000 x 30) .

    [0036] The grouping together according to this first embodiment is useful when one has little extensive job knowledge, and makes it possible to spot quite easily any flight path deviation.

    [0037] According to a second embodiment, the grouping together E2 in the vector Xi consists in focusing on the value of the parameters at precise instants, which are known as snapshots. These precise instants have an operational sense, for example for the landing phase a snapshot may be made every 1000 feet from 10000 feet of altitude, or then snapshots may be made at instants where particular events take place: landing gear down, etc.

    [0038] In assuming that n instants of snapshots have been defined then for a flight i the instants of snapshots

    are noted. It should be noted these instants do not necessarily take place at the same moment for each flight.

    [0039] The vector Xi is then of dimension d = n × P and is defined by



    [0040] This second embodiment makes it possible to obtain better results than with the first embodiment. Indeed, in this second embodiment, since T>>n there is much less data to process than in the first embodiment.

    [0041] According to a third embodiment, the grouping together E2 in the vector Xi consists in "summarising" each "curve" of each parameter in a small set of k≈5 or 10 values: these k values may for example be the average, the variance, the max. value, the min. value of each parameter recorded.

    [0042] These k values are then combined for all of the parameters to obtain the signature vector Xi of dimension d = k × P for k ≈ 4 is defined by



    [0043] If for example there are P=30 parameters, one then has a signature vector Xi of size d = 120.

    [0044] At the end of the grouping together step E2, there are available N ≥ 1 signature vectors Xi of size d, the components of which correspond to pertinent data recorded during said flight i of the aircraft.

    [0045] In a complementary manner, each of the components of each signature vector Xi may be normalised E3.

    [0046] Such a normalisation proves useful when the components of each signature vector Xi have varied orders of magnitude.

    [0047] In a preferred manner, the normalisation E3 leads to having components of zero average and standard deviation 1 on the set of all of the flights.

    [0048] Xi is hereafter employed to designate the signature vector of a flight i, the components of which are normalised or not.

    [0049] The signature vectors Xi are going to undergo several treatments to make it possible to identify if a flight is abnormal.

    [0050] To do this, a Gaussian kernel entropy component analysis E4 of the signature vectors Xi is going to be implemented to obtain a zone of normal flights E and classifying the signature vectors Xi with respect to their distance to said zone.

    [0051] In other words, it involves, from all of the flight signatures Xi grouping them together in order to demarcate a zone E of normal flights.

    [0052] Such an analysis E4 is for example described in the document R. Jenssen: "Kernel entropy component analysis", IEEE transactions on pattern analysis and machine intelligence, vol. 32, n°5, pages 847-860, May 2010.

    [0053] To carry out the analysis, a matrix of similarity K is determined E41 of size N × N, the components of which quantify the proximity between two signature vectors Xi (in other words between two flights).

    [0054] This matrix K is known by the name Gram matrix.

    [0055] The matrix K is obtained from a matrix of the distances D to which a Gaussian function with a predetermined bandwidth parameter σ2 is applied (the determination of this parameter will be described later).

    [0056] The matrix K has for expression ∀i = 1,...,N



    [0057] The bandwidth parameter is obtained by calculation of the distances of each signature vector Xi to its k nearest neighbours with k a whole number between 4 and 10. The value of k depends on the number of flights N and the dimension d.

    [0058] To determine E42 the bandwidth parameter σ2, a matrix D is determined of size N × N corresponding to the distance between two flights and which is defined in the following manner ∀i = 1,...,N Di,j = ∥Xi - Xj2.

    [0059] Next, for each of the columns of the matrix D, its elements are arranged in ascending order to obtain a matrix D' such that ∀j = 1,...,N



    [0060] Next, the k first lines of the matrix D' thereby obtained are selected and the others are eliminated to obtain a matrix D' of size k × N.

    [0061] Then, the average of each column of the matrix D' is determined to obtain N values y1,...,yN which represent in fact the average distance of each vector to its k nearest neighbours.

    [0062] Next, the values y1,...,yN that are aberrant are eliminated. To do this, one begins by determining med the median value of the set y1,...,yN, which is written med = median{yl, ..., yN}. Next the absolute median deviation of the values y1,...,yN defined as being the median value of the set {|y1 - med|, ... |yN - med|} which is written mad = mediane{|yi - med|} is determined.

    [0063] Then, the bandwidth parameter σ2 is determined from the median absolute deviation of the values y1,...,yN by the following functional

    In other words, all of the values yi that are greater than med+20.mad have been eliminated and the largest of the remaining values is chosen as bandwidth parameter.

    [0064] Once the matrix K is obtained, the matrix of similarity K is broken down E43 into eigen vectors to obtain N eigen vectors a1,...,aN and N eigen values λ1,...,λN such that ∀i = 1,...,N K·ai = λi·ai.

    [0065] It should be noted that each of the eigen vectors ai is of size N :

    and ai,j designates the jth component of the vector ai. The λi are also called the energy coefficients.

    [0066] From the vectors and eigen values obtained one determines E44 for each eigen vector its entropy coefficient γi defined by



    [0067] The entropy coefficients are a pertinent criterion to select only the most pertinent data in the set of signature vectors of all the flights.

    [0068] In particular, a sub-assembly of eigen vectors {am}m∈{1,...,N} is selected E45 such that the sum of the corresponding entropies γm is greater than the percentage of the sum of the N entropies γi.

    [0069] In a preferred manner, the percentage is comprised between 75 and 95%, preferably 90%.

    [0070] The selection of this sub-assembly makes it possible to define a zone of normal flights E going back to the parameters of each vector Xi associated with the selected eigen vectors.

    [0071] In relation with figure 2, if a simple example is considered with two parameters, it is then possible to represent all the flights in a two dimensional graph and it is then possible to determine the zone of normal flights E as being the circle surrounding the cluster of data.

    [0072] Then, for each flight i, an abnormality score zi is determined E5, defined by the distance of the flight Xi with respect to the zone of normal flights E.

    [0073] The abnormality score is defined by

    with ak the eigen vectors selected and λk their associated eigen values.

    [0074] The abnormality score is comprised between 0 and 1. The closer the score is to 1, the more the flight is considered abnormal.

    [0075] With reference to figure 2, the flights far from the zone of normal flights E have "abnormal" scores.

    [0076] In a preferred manner, it is considered that the flights i for which the abnormality score is greater than 0.99 are very probably abnormal and the flights i for which the score is greater than 0.999 are very decidedly abnormal.

    [0077] Consequently, one detects E6, as a function of the abnormality score zi, if at least one flight is abnormal.

    [0078] The abnormality score may be displayed to be viewed by an analyst.

    [0079] As will have been understood, to determine whether a flight is abnormal, the method uses all the data at its disposal to, itself, determine in what manner a flight may be considered normal and thereby determine which are not. Thus, it is not necessary to put in place detection rules.

    [0080] If an abnormal flight is detected, for the latter a nominal reference flight known as "phantom flight" the nearest to this abnormal flight detected is determined E7 while being in the zone of normal flights, the phantom flight being able to be different to a signature flight Xi.

    [0081] In other words, the phantom flight is defined as the theoretical flight the nearest to the abnormal flight detected while being in the zone of normal flights.

    [0082] It is in particular the signature Z0 of the phantom flight that is going to be determined.

    [0083] The phantom flight has a score z less than 0.99 (or 0.95 if it is really wished to ensure the normality of this flight).

    [0084] The phantom flight is thus the projection of the abnormal flight detected in the zone of normal flights. In particular, it is the nearest projection.

    [0085] The phantom flight is generated during the method, it is potentially independent of the flights analysed.

    [0086] The phantom flight of signature Z0 associated with an abnormal flight detected of signature Xi is generated via a constrained non linear optimisation procedure. This optimisation procedure, for a fixed threshold value seuil, is expressed in the following manner:



    [0087] It should be noted that the norm 1 above makes it possible to ensure that the minimum of components are changed between the abnormal flight and its phantom flight.

    [0088] This optimisation procedure necessitates the calculation of the score z(Z) to obtain the signature Z0 of the phantom flight.

    [0089] To do this, the following vector is defined as kZ, valid for all





    [0090] Thus, the function z giving the score of any vector of

    is defined by:

    with a1,...,am the m eigen vectors retained previously (covering 90% of the total entropy) and λ1,...,λm their associated eigen values.

    [0091] In relation with figure 2, if the flight corresponding to the parameters referenced 20 is abnormal, the corresponding phantom flight is the flight for which the parameters are referenced 20'.

    [0092] This makes it possible to compare E8 the parameters of the abnormal flight detected with those of the phantom flight determined in order to detect at least one parameter of the abnormal flight that has rendered abnormal said abnormal flight detected.


    Claims

    1. Method of analysing flight data recorded during N flights of at least one aircraft, by means of a flight data recorder of an aircraft, the data being grouped together (E2) by flight i in a signature vector of the flight Xi of size d, the components of which correspond to data recorded during said flight i of the aircraft, a flight i being thus defined by the signature vector Xi, the method comprising the following steps:

    - Gaussian kernel entropy component analysis (E4) of the flight signatures Xi to obtain a zone of normal flights (E) and classifying the flight signatures Xi with respect to their distance to said zone;

    - determining (E5), for each flight i, an abnormality score zi defined by the distance of a flight signature Xi with respect to the zone of normal flights (E);

    - detecting (E6), as a function of the abnormality score zi, at least one abnormal flight;

    characterised in that the method further comprises the step of

    - determining for each abnormal flight, a phantom flight the nearest to the abnormal flight while being in the zone of normal flights, wherein the parameters of the abnormal flight detected are compared with those of the phantom flight determined in order to detect at least one parameter of the abnormal flight which has rendered abnormal said abnormal flight detected.


     
    2. Analysis method according to claim 1, in which the Gaussian entropy component analysis comprises the following sub-steps:

    - determination of a matrix of similarity K of size N × N, the components of which quantify the proximity between two flight signatures Xi;

    - breakdown into eigen vectors of the matrix of similarity K to obtain N eigen vectors a1,...,aN and N eigen values λ1,...,λN such that ∀i = 1,...,N K·ai = λi·ai;

    - determination for each eigen vector of its entropy coefficient γi;

    - selection of a sub-assembly of eigen vectors {am}m∈{1,...,N} such that the sum of the entropies γm is greater than the percentage of the sum of the N entropies γi.


     
    3. Analysis method according to claim 2, in which the entropy coefficient is defined by ∀i = 1,...,N


     
    4. Analysis method according to one of claims 2 to 3 in which the abnormality score of a flight i is defined by:


     
    5. Analysis method according to one of claims 2 to 4, in which the components of the matrix of similarity K are defined by:

    where σ2 is a predetermined bandwidth parameter.
     
    6. Analysis method according to the preceding claim, in which the bandwidth parameter σ2 is determined in the following manner:

    - a matrix D of size N × N corresponding to the distance between two flights is determined and is defined in the following manner ∀i = 1,...,N Di,j = ∥Xi - Xj2;

    - each component of each column is arranged in ascending order to obtain a matrix D';

    - k first lines of the matrix D' thereby obtained are selected and the others are eliminated to obtain a matrix D' of size k × N;

    - the average of each column of the matrix D' to obtain N values y1,...,yN is determined;

    - the median absolute deviation of the values y1,...,yN defined by mad = mediane{|yi - med|} with med = mediane{y1,...,yN} is determined;

    - the bandwidth parameter σ2 is determined from the median absolute deviation of the values y1,...,yN by the following functional


     
    7. Analysis method according to one of the preceding claims, in which the data of a flight i are grouped together in a matrix Fi of dimension T × P with T the number of data recorded during the flight i and P the number of parameters recorded, the vector Xi having for components the columns of the matrix Fi end to end, the vector Xi being then of dimension d = T × P and is defined by


     
    8. Analysis method according to one of claims 1 to 5, in which the data of a flight i are grouped together in a matrix Fi of dimension T × P with T the number of data recorded during the flight i and P the number of parameters recorded, the vector Xi having for components the columns of the matrix Fi sampled to select n < T recordings

    of parameters, the vector Xi being then of dimension d = n × P and is defined by


     
    9. Analysis method according to one of claims 1 to 6, in which the data of a flight i are grouped together in a matrix Fi of dimension T × P with T the number of data recorded during the flight i and P the number of parameters recorded, the vector Xi having for components the average, the variance, the minimal value, the maximal value of a parameter among all the recordings of this parameter, the vector Xi being then defined by:


     


    Ansprüche

    1. Verfahren zum Analysieren von Flugdaten, die während N Flügen mindestens eines Flugzeugs aufgezeichnet werden, mittels eines Flugdatenschreibers eines Flugzeugs, wobei die Daten nach Flug i in einem Signaturvektor der Flüge Xi der Größe d zusammengefasst (E2) werden, dessen Komponenten Daten entsprechen, die während des Fluges i des Flugzeugs aufgezeichnet werden, wobei ein Flug i somit durch den Signaturvektor Xi definiert ist, wobei das Verfahren die folgenden Schritte umfasst:

    - Gauß-Kernel-Entropiekomponentenanalyse (E4) der Flugsignaturen Xi, um einen Bereich normaler Flüge (E) zu erhalten, und Klassifizieren der Flugsignaturen Xi in Bezug auf ihren Abstand zu diesem Bereich;

    - Bestimmen (E5), für jeden Flug i, eines Anomaliewerts zi, der durch den Abstand einer Flugsignatur Xi in Bezug auf den Bereich normaler Flüge (E) definiert ist;

    - Erkennen (E6), in Abhängigkeit vom Anomaliewert zi, mindestens eines anormalen Fluges;

    dadurch gekennzeichnet, dass das Verfahren weiter den Schritt umfasst des

    - Bestimmens, für jeden anormalen Flug, eines Phantomfluges, der dem anormalen Flug an nächsten liegt, während er sich gleichzeitig im Bereich normaler Flüge befindet, wobei die Parameter des erkannten anormalen Fluges mit denjenigen des bestimmten Phantomfluges verglichen werden, um mindestens einen Parameter des anormalen Fluges zu erkennen, der den erkannten anormalen Flug anormal werden ließ.


     
    2. Analyseverfahren nach Anspruch 1, wobei die Gauß-Entropiekomponentenanalyse die folgenden Teilschritte umfasst:

    - Bestimmen einer Ähnlichkeitsmatrix K der Größe N x N, deren Komponenten die Nähe zwischen zwei Flugsignaturen Xi quantifizieren;

    - Zerlegen der Ähnlichkeitsmatrix K in Eigenvektoren, um N Eigenvektoren a1,...,aN und N Eigenwerte λ1,...,λN zu erhalten, derart, dass ∀i = 1,...,N K·ai = λi·ai;

    - Bestimmen, für jeden Eigenvektor, dessen Entropiekoeffizienten γi;

    - Auswählen eines Teilsatzes von Eigenvektoren {am}m∈{1,...,N} derart, dass die Summe der Entropien γm größer ist als der Prozentsatz der Summe der N Entropien γi.


     
    3. Analyseverfahren nach Anspruch 2, wobei der Entropiekoeffizient definiert ist durch ∀i = 1,...,N


     
    4. Analyseverfahren nach einem der Ansprüche 2 bis 3, wobei der Anomaliewert eines Fluges i definiert ist durch:


     
    5. Analyseverfahren nach einem der Ansprüche 2 bis 4, wobei die Komponenten der Ähnlichkeitsmatrix K definiert sind durch:

    wobei σ2 ein vorbestimmter Bandbreitenparameter ist.
     
    6. Analyseverfahren nach dem vorstehenden Anspruch, wobei der Bandbreitenparameter σ2 in der folgenden Weise bestimmt wird:

    - eine Matrix D der Größe N x N, die dem Abstand zwischen zwei Flügen entspricht, wird bestimmt und ist in der folgenden Weise definiert ∀i = 1,...,N Di,j = ∥Xi - Xj2;

    - jede Komponente jeder Spalte wird in aufsteigender Reihenfolge angeordnet, um eine Matrix D' zu erhalten;

    - k erste Zeilen der dadurch erhaltenen Matrix D' werden ausgewählt und die anderen werden verworfen, um eine Matrix D' der Größe k x N zu erhalten;

    - der Durchschnitt jeder Spalte der Matrix D' wird bestimmt, um N Werte y1,...,yN zu erhalten;

    - die mittlere absolute Abweichung der Werte y1,...,yN, definiert durch mad = mediane{|yl - med|} mit med = mediane {y1,...,yN), wird bestimmt;

    - der Bandbreitenparameter σ2 wird aus der mittleren absoluten Abweichung der Werte y1,...,yN durch die folgende Funktion

    bestimmt.


     
    7. Analyseverfahren nach einem der vorstehenden Ansprüche, wobei die Daten eines Fluges i in einer Matrix Fi der Dimension T x P zusammengefasst werden, wobei T die Anzahl während des Fluges i aufgezeichneter Daten und P die Anzahl aufgezeichneter Parameter ist, wobei der Vektor Xi als Komponenten die Spalten der Matrix Fi von Ende zu Ende aufweist, wobei der Vektor Xi dann der Dimension d = T x P ist und definiert ist durch


     
    8. Analyseverfahren nach einem der Ansprüche 1 bis 5, wobei die Daten eines Fluges i in einer Matrix Fi der Dimension T x P zusammengefasst werden, wobei T die Anzahl während des Fluges i aufgezeichneter Daten und P die Anzahl aufgezeichneter Parameter ist, wobei der Vektor Xi als Komponenten die Spalten der Matrix Fi aufweist, die abgetastet werden, um n < T Aufzeichnungen

    von Parametern auszuwählen, wobei der Vektor Xi dann der Dimension d = n x P ist und definiert ist durch


     
    9. Analyseverfahren nach einem der Ansprüche 1 bis 6, wobei die Daten eines Fluges i in einer Matrix Fi der Dimension TxP zusammengefasst werden, wobei T die Anzahl während des Fluges i aufgezeichneter Daten und P die Anzahl aufgezeichneter Parameter ist, wobei der Vektor Xi als Komponenten den Durchschnitt, die Varianz, den Minimalwert, den Maximalwert eines Parameters aus allen den Aufzeichnungen dieses Parameters aufweist, wobei der Vektor Xi dann definiert ist durch:


     


    Revendications

    1. Méthode d'analyse de données de vol enregistrées pendant N vols d'au moins un aéronef, au moyen d'un enregistreur de données de vol d'un aéronef, les données étant regroupées (E2) par vol i dans un vecteur de signature du vol Xi de taille d, dont les composantes correspondent à des données enregistrées pendant ledit vol i de l'aéronef, un vol i étant ainsi défini par le vecteur de signature Xi, la méthode comprenant les étapes suivantes :

    - une analyse en composante d'entropie de noyau gaussien (E4) des signatures de vol Xi pour obtenir une zone de vols normaux (E) et la classification des signatures de vol Xi en relation avec leur distance par rapport à ladite zone ;

    - la détermination (E5), pour chaque vol i, d'un score d'anomalie zi défini par la distance d'une signature de vol Xi en relation avec la zone de vols normaux (E) ;

    - la détection (E6), en fonction du score d'anomalie zi, d'au moins un vol anormal ;

    caractérisée en ce que la méthode comprend en outre l'étape de :

    - détermination pour chaque vol anormal, d'un vol fantôme le plus proche du vol anormal tout en étant dans la zone de vols normaux, dans laquelle les paramètres du vol anormal détecté sont comparés à ceux du vol fantôme déterminé afin de détecter au moins un paramètre du vol anormal qui a rendu anormal ledit vol anormal détecté.


     
    2. Méthode d'analyse selon la revendication 1, dans laquelle l'analyse de composante d'entropie gaussienne comprend les sous-étapes suivantes :

    - la détermination d'une matrice de similarité K de taille N x N, dont les composantes quantifient la proximité entre deux signatures de vol Xi ;

    - la décomposition en vecteurs propres de la matrice de similarité K pour obtenir N vecteurs propres a1,...,aN et N valeurs propres λ1,...,λN de sorte que ∀i = 1,...,N K·ai = λi·ai;

    - la détermination pour chaque vecteur propre de son coefficient d'entropie γi ;

    - la sélection d'un sous-ensemble de vecteurs propres {am}m∈{1,....,N} de sorte que la somme des entropies γm soit supérieure au pourcentage de la somme des N entropies γi.


     
    3. Méthode d'analyse selon la revendication 2, dans laquelle le coefficient d'entropie est défini par ∀i = 1,...,N


     
    4. Méthode d'analyse selon l'une des revendications 2 et 3, dans laquelle le score d'anomalie d'un vol i est défini par :


     
    5. Méthode d'analyse selon l'une des revendications 2 à 4, dans laquelle les composantes de la matrice de similarité K sont définies par :

    où σ2 est un paramètre de largeur de bande prédéterminé.
     
    6. Méthode d'analyse selon la revendication précédente, dans laquelle le paramètre de largeur de bande σ2 est déterminé de la manière suivante :

    - une matrice D de taille N x N correspondant à la distance entre deux vols est déterminée et est définie de la manière suivante ∀i = 1,...,N Di,j = ∥Xi - Xj2;

    - chaque composante de chaque colonne est agencée dans un ordre croissant pour obtenir une matrice D' ;

    - k premières lignes de la matrice D' ainsi obtenue sont sélectionnées et les autres sont éliminées pour obtenir une matrice D' de taille k x N ;

    - la moyenne de chaque colonne de la matrice D' pour obtenir N valeurs y1,...,yN est déterminée ;

    - l'écart moyen absolu des valeurs y1,...,yN défini par mad = mediane {|yi - med|} avec med = médiane {y1,...,yN} est déterminé ;

    - le paramètre de largeur de bande σ2 est déterminé à partir de l'écart moyen absolu des valeurs y1,...,yN par la fonctionnelle suivante


     
    7. Méthode d'analyse selon l'une des revendications précédentes, dans laquelle les données d'un vol i sont regroupées en une matrice Fi de dimension T x P avec T le nombre de données enregistrées pendant le vol i et P le nombre de paramètres enregistrés, le vecteur Xi ayant pour composantes les colonnes de la matrice Fi bout à bout, le vecteur Xi étant alors de dimension d = T x P et est défini par


     
    8. Méthode d'analyse selon l'une des revendications 1 à 5, dans laquelle les données d'un vol i sont regroupées en une matrice Fi de dimension T x P avec T le nombre de données enregistrées pendant le vol i et P le nombre de paramètres enregistrés, le vecteur Xi ayant pour composantes les colonnes de la matrice Fi échantillonnée pour sélectionner n < T enregistrements

    de paramètres, le vecteur Xi étant alors de dimension d = n x P et est défini par


     
    9. Méthode d'analyse selon l'une des revendications 1 à 6, dans laquelle les données d'un vol i sont regroupées en une matrice Fi de dimension T x P avec T le nombre de données enregistrées pendant le vol i et P le nombre de paramètres enregistrés, le vecteur Xi ayant pour composantes la moyenne, la variance, la valeur minimale, la valeur maximale d'un paramètre parmi tous les enregistrements de ce paramètre, le vecteur Xi étant alors défini par :


     




    Drawing











    Cited references

    REFERENCES CITED IN THE DESCRIPTION



    This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

    Patent documents cited in the description




    Non-patent literature cited in the description