• 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:
    The invention proposes to identify precursors to all phenomena that may have an impact on the commissioning of an aircraft. It relates to a system for prediction of faults in an aircraft, comprising a processor (3) configured to analyze a current behavior of at least one flight parameter of said aircraft to detect any deviation from said current behavior with respect to a behavior model (14). ) predetermined said parameter, said behavior model being determined from a plurality of learning data suites relating to said parameter collected during the flights of a set of aircraft.
    公开号:FR3052273A1
    申请号:FR1655040
    申请日:2016-06-02
    公开日:2017-12-08
    发明作者:Jean-Max Huet；Franck Duluc；Eric Benhamou；Bruno Maillard
    申请人:Airbus Operations SAS；
    IPC主号:
    专利说明:

    PREDICTION OF TROUBLES IN AN AIRCRAFT DOMAIN OF THE INVENTION
    The present invention relates to the field of detection of first failures in an aircraft. In particular, the invention relates to a method and a system for predicting failures to anticipate the maintenance operations of an aircraft.
    Each aircraft has a Built-in Test Equipment (BITE) to detect and isolate any equipment that has failed. In addition, critical flight controls are monitored by a Flight Warning System (FWS), which reports any failure or malfunction to the crew. The alert transmitted before the flight by the FWS indicates the impact of an event on the operability of the aircraft allowing the crew to determine the commissioning status of the aircraft.
    The data transmitted by the BITE and FWS systems are acquired and processed by the centralized supervision system CMS (Central Maintenance System) of the various components of the aircraft.
    The functions of the centralized CMS supervision system are hosted on two identical computers. The CMS aggregates and processes anomaly data from a chain of different components to generate the diagnosis of the anomaly.
    The results of the various diagnoses made by the CMS are saved in an on-board database and are also displayed on an on-board interface via Multipurpose Control and Display Units (MCDUs).
    In addition, alert or anomaly messages are transmitted to the ground stations via a message system called Aircraft Communication Addressing and Reporting System (ACARS) to be processed in the case of obvious anomalies and otherwise to be archived in data base. These messages are transmitted to the ground in the form of reports during the CFR (Current Flight Report) flight or at the end of the PFR (Post Flight Report) flight.
    The CMS monitoring system thus provides useful information to the maintenance and pilots to know if an aircraft can be used safely or whether to block the aircraft to perform maintenance before putting it into service.
    This monitoring system is very efficient and provides maximum security for the aircraft. However, in some cases, pilots may be caught short by an unexpected blackout alert prior to the flight that may result in flight delays or even cancellations.
    In addition, the monitoring system relies on predefined rules in advance to detect faults. However, there may be phenomena that are not currently suspected and that may have an impact. The object of the present invention is therefore to provide a method or a system for identifying precursors to all phenomena that may have an impact on the operability of an aircraft, thus making it possible to plan maintenance operations for a long period of time. in advance to avoid problems with flight delays or cancellations.
    OBJECT AND SUMMARY OF THE INVENTION
    The present invention relates to a method for prediction of faults in an aircraft, comprising an analysis of a current behavior of at least one flight parameter of said aircraft to detect any deviation from said current behavior with respect to a model of preconstructed behavior of said parameter, said behavior model being constructed during a prior learning phase from a plurality of training data suites relating to said parameter collected during the flights of a set of aircraft.
    This makes it possible to identify precursors of faults very early, thus making it possible not to be taken aback by an unexpected failure and consequently, to anticipate the maintenance and to plan well in advance while ensuring an optimal availability of the aircraft without flight delays. In addition, this method makes it possible to detect without any a priori atypical behavior or the premise of unsuspected failures that may have an impact on the availability of the aircraft.
    According to one embodiment of the invention, the detection of any deviation in the behavior of a flight parameter comprises the following steps: collecting on each current flight a current data series resulting from measurements relating to said parameter; segmenting said current data stream into a predetermined number of segments to transform it into a current segment chain, said number of segments being defined by the behavior model; aligning the segments of said current segment chain with respect to a reference segment chain defined by the behavior model; allocating predetermined characteristic measurements to said current segment chain to generate a current measurement vector; and calculating a distance between said current measurement vector and a reference measurement vector defined by the behavior model, the value of said distance being representative of a normal or deviated behavior of said flight parameter.
    This makes it possible to estimate the probability of occurrence of failure with great precision.
    Advantageously, the method further comprises calculating an abnormal behavior score of said flight parameter.
    This makes it possible to characterize the level of degradation and to estimate the delay between the detection of the first moments of failure and the actual occurrence of the failure.
    Advantageously, the method further comprises an identification of a set of flights of said aircraft exhibiting deviated behavior and maintenance actions performed during said set of flights.
    This allows for accurate planning of future maintenance to properly manage the return to service of aircraft.
    Advantageously, each training data or current data stream includes time data from a flight recorder and / or time indexed fault messages from a centralized supervision system.
    This makes it possible to have several sources of information increasing the accuracy of the detection of first failures.
    According to one embodiment of the present invention, the construction of the behavior model relating to a given parameter comprises the following steps: - collecting at each of the flights of all the aircraft at least one suite of training data resulting from the measurements relating to said parameter thus forming, during the flights of all the aircraft, said plurality of training data sequences; -segmenting each sequence of training data into an optimal number of segments to transform it into a chain of learning segments, the optimal number of segments defining said predetermined number of segments associated with said parameter; transforming said set of training data suites into a corresponding set of training segment strings; selecting a central segment string from said set of training segment strings, the central segment string defining said reference segment string; aligning the segments of said set of training segment strings by locating themselves with respect to said reference segment string; generating a learning measure vector for each training segment string by assigning predetermined characteristic measures to each learning segment and / or set of segments of each training segment string; calculating a learning distance between each learning measurement vector and a reference measurement vector associated with said reference segment string; and partitioning the set of learning measure vectors into consecutive intervals according to the learning distances thus defining a normal behavioral interval and atypical behavior intervals.
    This makes it possible to use a large volume of information to generate an accurate, reliable, robust and easy-to-use behavior model to highlight any atypical behavior of a parameter relating to a new flight and therefore allows analysis to be performed. the causes of atypical behavior and to prevent future failures.
    Advantageously, said central segment chain corresponds to the medoid of said set of segment chains.
    Advantageously, the method further comprises the following steps: - saving said reference segment chain, the distances between the measurement vectors and the reference measurement vector and the partitioning of said measurement vectors, - displaying a graph representative of the model of reference behavior. The invention also relates to a system for predicting failures in an aircraft, comprising a processor configured to analyze a current behavior of at least one flight parameter of said aircraft to detect any deviation from said current behavior with respect to a predetermined behavior model ( for example, a theoretical model) of said parameter, said behavior model being determined from a plurality of training data sequences relating to said parameter collected during the flights of a set of aircraft. The invention also relates to a monitoring system comprising avionics maintenance and management systems and the fault prediction system according to the above characteristics.
    This makes it possible very early to make an accurate planning of the maintenance in order to properly manage the return to service of the aircraft.
    BRIEF DESCRIPTION OF THE DRAWINGS Other features and advantages of the device and the method according to the invention will emerge more clearly on reading the description given below, by way of indication but without limitation, with reference to the appended drawings in which: FIG. . 1 schematically illustrates a system for predicting failures in an aircraft, according to one embodiment of the invention; - Figs. 2 and 3 schematically illustrate a construction of a behavior model, according to an embodiment of the invention; FIG. 4 illustrates a display of a graph representative of the behavior model associated with a given parameter, according to one embodiment of the invention; FIG. 5 illustrates a method of fault prediction in an aircraft in relation to the system of FIG. 1, according to a preferred embodiment of the invention; FIG. 6 illustrates a graph representing a set of successive flights of a specific aircraft, according to one embodiment of the invention; FIG. 7 illustrates the feature of a specific parameter during a set of successive flights of a specific aircraft between two specific airports, according to another embodiment of the invention; and - FIG. 8 illustrates a monitoring system including the fault prediction system, according to one embodiment of the invention.
    DETAILED DESCRIPTION OF EMBODIMENTS
    The principle of the invention is to make the best use of the history of behavior of parameters of a very large number of flights of a plurality of aircraft in order to automatically and without priori detect the premise of failures on an aircraft.
    Fig. 1 schematically illustrates a system for predicting failures in an aircraft, according to one embodiment of the invention.
    The fault prediction system 1 comprises a data acquisition module 2, a processor 3, storage units 4 and an input and output interface 5.
    The acquisition module 2 is configured to retrieve data from measurements relating to flight parameters of the aircraft 6.
    Indeed, during each flight, the aircraft 6 proceeds to record information on different avionics parameters. An acquisition system 7 centralizes and formats all the data from the various sensors, on-board computers, or other instruments and transfers them to flight recorders 8 (FDR) (Flight Data Recorders) via dedicated links. The data may be discrete (eg logic detection states, indicators, switch or relay states, etc.), analog (eg potentiometer data), synchronization data, etc.
    In addition, a centralized supervision system 9 (CMS) processes the data transmitted by the fault detection (BITE) and monitoring (FWS) systems 10 and generates reports or fault messages.
    Thus, the acquisition module 2 is configured to recover the ordered data (usually time-indexed) relating to the flight parameters from at least one flight recorder 8 and / or the messages of failures indexed in an orderly manner from the system. Centralized Supervision System 9 (CMS).
    The processor 3 is configured to analyze the behavior of at least one flight parameter of the aircraft 6 to detect any deviation from the behavior of the parameter with respect to a preconstructed reference behavior model 14 saved in the storage units 4.
    Furthermore, the interface 5 is configured to display the results of the analysis illustrating any difference in behavior of each parameter with respect to each corresponding model of behavior 14.
    In addition, the storage units 4 are configured to save the data acquired during the flight, the results of the analysis and the reference behavior model 14.
    It will be noted that the behavior model 14 of a parameter is constructed during a preliminary learning phase from a plurality of learning data suites relating to said parameter collected during the flights of a plurality of aircraft. As a result of data is meant data having an ordered temporal type indexing.
    The present invention thus proposes to analyze ordered data in a global and / or local manner according to different perspectives in order to detect data that are significantly different from a set of other data labeled as being normal.
    Fig. 2 schematically illustrates a construction system of the behavior model, according to one embodiment of the invention.
    In addition, FIG. 3 illustrates in connection with FIG. 2, a method of constructing the behavior model associated with a parameter, according to one embodiment of the invention.
    The construction of the behavior model 14 consists in selecting, for each given parameter, a group of flights of a set of aircraft 16 having a normal behavior (at least, with respect to said parameter) and recording their ordered data. to extract specific characteristics to generate the model of behavior 14. This model 14, once constructed can then be used for each new flight to determine if the parameter associated with the new flight has an abnormal behavior compared to the model of behavior 14.
    In a similar manner to the prediction 1 system, the construction system 21 of the behavior model 14 comprises a learning acquisition module 22, a learning processor 23, learning storage units 24 and an interface learning 25.
    In the following, we will expose the construction of a behavior model 14 associated with a specific parameter but of course, the process is the same for the construction of each of the flight parameters. In step E1, the learning acquisition module 22 is configured to collect at each flight of all the aircraft 16 at least one sequence of training data from the measurements relating to the parameter. Thus, during the flights of all the aircraft 16, the learning acquisition module 22 collects a plurality of training data sequences.
    Each training data suite includes time data from a flight recorder of a corresponding aircraft and / or time indexed failure messages from the centralized supervision system of the corresponding aircraft.
    Note that each flight may have a different duration, a different route and flight phases different from other flights. Thus, the ordered data from the different flights do not necessarily have the same lengths or the same characteristics at the same time thus complicating the comparison between these different data.
    In addition, data from flights of a large population of aircraft 16 represents a huge mass of information that can not be analyzed directly.
    Indeed, in step E2 and in order to reduce the size of the data, the learning processor 23 is configured to segment each series of training data into an optimal number of segments sl ... sk having a good approximation of the sequence of starting data.
    Advantageously, the learning processor 23 is first configured to calculate for each flight independently, a minimum number of segments associated with the parameter. The minimum number of segments for each flight is calculated according to an iterative process by initially setting an initial number of segments and analyzing the convergence of the segments to the data suite at each iteration. By way of example, the convergence criterion corresponds to a minimum difference between the curve formed by the data sequence and that formed by the segments. This difference can be determined by simply calculating the area between the two curves.
    In order to verify that we have reached the minimum number of segments, we can apply a strategy of gains to zero result (non-positive gain, in English) which consists of removing segments that do not (or almost no) provide additional precision to convergence.
    Once a minimum number of segments is calculated for each flight, the learning processor 23 is configured to analyze the distribution of all the minimum numbers of segments of the different flights to determine a single optimal number n of segments sl. .sn acceptable for all flights. In other words, for a given parameter, the learning processor 23 calculates a single optimum number of segments n (which will be called in the following predetermined number of segments) valid for all the flights of all the aircraft 16. Each predetermined number n The segment associated with each parameter is stored in the training storage units 14 so that it can be used during the detection of first failures at each new flight.
    Thus, the segmentation transforms each sequence of learning data into a curve consisting of a chain of learning segments, each segment being defined only by two points that can be further shared by the neighboring segments. For example, if a data sequence indexed by 3000 points is divided into six segments, then a subsequence of only 7 points instead of 3000 is obtained. In step E3, the training processor 23 is configured to use the unique predetermined number n (ie, the optimal number) of segments associated with a given parameter for further segmenting each sequence of training data.
    Thus, the training processor 23 transforms each set of training data suites into a corresponding set of training segment strings cl ... cp which is then stored in the learning storage units 24. At each In step E4, the training processor 23 is configured to select a central segment chain M1 from the set of learning segment strings determined in the previous step.
    Advantageously, the central segment chain M1 is determined by calculating the medoid of the set of training segment strings. The medoid is a chain of segments belonging to the set of segment strings and represents a minimal similarity gap with all other segment strings. It roughly corresponds to the middle segment chain of the set of segment strings. The central segment chain M1, which will be referred to in the following reference segment chain M1, thus defines a reference with respect to which the other segment strings are aligned. This chain of reference segments M1 is saved in the training storage units 24 so that it can be used during the detection of first failures at each new flight.
    Indeed, in step E5, the learning processor 23 is configured to align the segments of the set of training segment chains cl ... cp by taking as an alignment mark, the chain of segments of reference Ml. The latter plays the role of a model for the set of segments chains cl ... cp. By way of example, the learning processor 23 is configured to apply a dynamic time warping (DTW) algorithm with respect to the reference segment chain M1 (DTW-Medoid). Alignment facilitates the task of comparison between the different segments. Note also that the alignment may possibly modify the initial segmentation of the training data suites.
    However, some characteristics of the initial data may be lost through segmentation. Then, in order to recover some of this lost information, metrics extracted from the initial ordered data are assigned to each segment or chain of learning segments cl ... cp.
    Indeed, in step E6, the learning processor 23 is configured to generate a vector of learning measurements VI ... Vp for each chain of learning segments cl ... cp by assigning predetermined characteristic measures individually to each learning segment and / or globally to the set of segments of each training segment chain. The predetermined characteristic measurements include for example measurements of slopes, averages, variances, standard deviations, minimums, maximums, angular velocities, frequency parameters, etc.
    The vectors of learning measurements VI ... Vp can be considered as multivariate time series increasing somewhat the dimension of the chains of learning segments. It will be noted that each learning measurement vector VI ... Vp associated with a given parameter represents a theft.
    This step (step E6) then corresponds to a meta-segmentation process consisting of calculating on each flight a few explanatory variables locally (i.e. by segment) or globally (i.e. for the entire segment chain). Each variable brings a new dimension which makes it possible to increase the precision of construction of the model of behavior 14.
    Advantageously, for each flight and each parameter, the learning processor 23 is configured to calculate percentiles Q 0, Q 1, Q 2, Q 3, 0.4 on the learning measure vectors knowing that the normal behavior is defined in the interval [ Qo, Q4]. Percentile values are saved in the training storage units 24.
    Advantageously, in step E7, the learning processor 23 is configured to normalize the learning measure vectors associated with the set of training segment strings cl ... cp. The standardized learning measurement vectors are saved in the learning storage units 24. In the step E8, the learning processor 23 is configured to calculate a learning distance dl ... dp between each vector of learning. learning measurements VI ... Vp (possibly normalized) and a vector of reference measurements Vm associated with the reference segment chain Ml.
    Advantageously, the learning processor 23 uses a measurement of Euclidean distance. As a variant, in the case where the measurement vectors are linearly independent, a Mahalanobis-type distance can be used which then takes into account the correlation between the different vectors. The learning distances (Euclidean and / or Mahalanobis) are stored in the learning storage units. In step E9, the learning processor 23 is configured to partition the set of learning measure vectors at consecutive intervals as a function of the learning distances. This partitioning defines a range of normal behavior In and one or more interval (s) of atypical behavior.
    Advantageously, it is possible to use the percentiles Qo, Qi, Q2, Q3, Q4 calculated previously on the measurement vectors associated with the set of segment chains to define the normal behavior interval knowing that outside this range, the behavior will be considered abnormal.
    Thus, all the analyzes and data (chain of reference segments M1, distances between the measurement vectors d1 ... dp, reference measurement vector Vm, partitioning of the measurement vectors) saved in the storage units of learning 24 is a precise, reliable and robust model of behavior.
    Advantageously, a representative graph of the behavior model is displayed on the learning interface 25.
    Indeed, FIG. 4 illustrates a display of a graph representative of the behavior model associated with a given parameter, according to one embodiment of the invention.
    This graph shows the different learning flights and their distances from a reference flight. Each flight is represented by a point corresponding to a learning measurement vector. The abscissa axis indicates the dates of the various flights and the ordinate axis indicates the distance of each flight relative to the reference flight. The ordinate axis is divided into two intervals In, la: the first In defining a flight population having a normal behavior for the parameter in question and the second defining it the flights having atypical behavior. This graph illustrates the positioning of each flight relative to the others and especially compared to the normal flight population. The further away a flight is from the normal population, the more abnormal it is.
    It should also be noted that this graph is divided into several columns (distinguished by the vertical lines) representing several corresponding aircraft. According to this example, there are eight columns representing eight aircraft and each column indicates the successive flights of each corresponding aircraft. Indeed, the flights of each aircraft are ordered in time along the abscissa axis within the corresponding column.
    This graph thus makes it possible to highlight in a very simple way any atypical behavior of a parameter relative to a new flight compared to normal flights and thus makes it possible to analyze the causes of the atypical behavior and especially to prevent future failures.
    Advantageously, before carrying out tests for new flights, the learning processor 23 is configured to perform statistical analyzes on the different learning flights on the basis of the dispersion of the percentiles Qo, Qi, Q2, Q3, Q4. We remove the points considered aberrant corresponding to flights that have very atypical behavior compared to the majority of other flights while avoiding eliminating extreme flights but valid.
    Fig. 5 illustrates a method of fault prediction in an aircraft in relation to the system of FIG. 1, according to a preferred embodiment of the invention.
    The fault prediction method includes steps similar to the method of constructing the behavior model. However, the fault prediction method applies the behavior model 14 to the data of a new flight to detect any behavioral deviation of a flight parameter from the normal flights of the behavior model 14. In the following, discloses the fault prediction method for a given parameter, but of course the method is applicable for each flight parameter.
    Initially, the behavior model 14 (including the predetermined number of segments, the reference segment chain, the predetermined characteristic measurements, etc.) is saved in the storage units 4. In step E21, the acquisition module 2 is configured to collect at each current flight a current data series from the measurements relating to the corresponding parameter.
    Each current data sequence includes time data from a flight recorder 8 of the aircraft and / or time indexed fault messages from a centralized supervision system 9 of the aircraft. In step E22, the processor 3 is configured to segment the current data stream according to the predetermined number of segments defined by the behavior model 14. This segmentation transforms the current data sequence into a current segment curve or chain S. In step E25, the processor 3 is configured to align the segments of the current segment chain C (by applying for example the DTW technique) with respect to the reference segment chain M1. It is recalled that the latter was defined during the construction of the behavior model 14 and serves as a reference for segment alignment. In step E26, the processor is configured to assign predetermined characteristic or explanatory measurements to the current segment chain C to generate a current measurement vector V. The predetermined characteristic measurements are the same as those used to construct the Behavior model 14. Characteristic measures (slope, mean, variance, standard deviation, minimum, maximum, angular velocity, frequency) are assigned to each segment and / or set of segments in the current segment chain. In step E28, the processor 3 is configured to calculate a d (Euclidean and / or Mahalanobis) distance between the current measurement vector V and the reference measurement vector Vm defined by the behavior model. The value of this distance is representative of a normal or abnormal behavior of the flight parameter.
    The value of the distance d between the current measurement vector V and the reference measurement vector Vm makes it possible to represent the current flight by a point on the graph representative of the behavior model 14 that can be displayed on the interface 5.
    Advantageously, the processor 3 is configured to calculate an abnormal behavior score of the flight parameter.
    Indeed, using the percentiles Qo, Qi, Ch, Ch, Ch, the processor 3 calculates the score K of a parameter of a flight represented by a point v according to the following algorithm: K = 0 for v <Qo then K = (v - Qi) / (Cb-Qi) otherwise if v> Q4 then K = (v- Q3) / (Cb-Qi).
    The K score makes it possible to quantify the level of anomaly and consequently to estimate the delay between the detection of the anomaly and the actual occurrence of the failure. The higher the value of the K score, the greater the anomaly.
    Advantageously, different graphs can be used to represent the scores of different flights of an aircraft.
    Fig. 6 illustrates a graph representing a set of successive flights of a specific aircraft, according to one embodiment of the invention. The y-axis represents the value of the K score (or the distance d). This graph identifies the Vi, Vj, Vk flights whose parameter exhibits atypical behavior as well as the actions and maintenance dates performed during these flights.
    Fig. 7 illustrates the behavior of a specific parameter during a set of successive flights of a specific aircraft between two specific airports, according to another embodiment of the invention.
    Successive flights are represented here by lines on a horizontal axis. The light or white line represents a flight during which no atypical behavior has been detected. The dark line tb represents a flight during which atypical behavior has been identified. Finally, the gray line tg represents a flight during which atypical behavior was detected but with a low K score compared to black lines.
    These clear lines tw, gray tg and dark tb define reliable markers of the degradation state of a system of the aircraft associated with the specific parameter. In particular, the gray line tg can be considered as a precursor of a degradation. In addition, by consulting the maintenance history, it is possible to identify the maintenance actions that have produced the best repairs, thus making it easier for maintenance operators to find any new fault finding.
    Fig. 8 illustrates a monitoring system including the fault prediction system, according to one embodiment of the invention.
    The monitoring system 31 includes the fault prediction system 1 as well as existing maintenance and management avionics systems 33 including a flight scheduling system, a maintenance information system ), a Trouble Shooting Manual (TSM), an Aircraft Maintenance Manual (AMM), and a minimum equipment list (MEL). This monitoring system 31 combines the information generated by the fault prediction system with that from the avionics maintenance and management systems. This combination of information allows maintenance engineers to identify the flight plans of a specific aircraft on which atypical behavior has been detected allowing them to take the right action at the right time to minimize service interruption. flights.
    For example, in the case of detection of atypical behavior, an aircraft may be subject to an MEL after the next two flights. The maintenance operators thus have two flights to be able to perform anticipatory maintenance actions or to select the most appropriate maintenance center while controlling the appropriate equipment. Precise maintenance planning can then be performed optimizing the management of the aircraft return to service.
    权利要求:
    Claims (10)
    [1" id="c-fr-0001]
    1. A method for predicting failures in an aircraft, characterized in that it comprises an analysis of a current behavior of at least one flight parameter of said aircraft (6) to detect any deviation from said current behavior with respect to a model preconstructed behavior (14) of said parameter, said behavior model being constructed during a prior learning phase from a plurality of learning data suites relating to said parameter collected during the flights of a set of aircraft (16).
    [2" id="c-fr-0002]
    2. Method according to claim 1, characterized in that the detection of any behavioral deviation of a flight parameter comprises the following steps: - collect at each current flight a current series of data from the measurements relating to said parameter; segmenting said current stream of data into a predetermined number of segments to transform it into a current segment chain (S), said number of segments being defined by the behavior model (14); aligning the segments of said current segment chain (S) with a reference segment chain (M1) defined by the behavior model (14); allocating predetermined characteristic measurements to said current segment chain to generate a current measurement vector (V); and calculating a distance (d) between said current measurement vector and a reference measurement vector defined by the behavior model, the value of said distance being representative of a normal or deviated behavior of said flight parameter.
    [3" id="c-fr-0003]
    3. Method according to claim 2, characterized in that it further comprises calculating a score (K) abnormal behavior of said flight parameter.
    [4" id="c-fr-0004]
    4. Method according to claim 2 or 3, characterized in that it further comprises an identification of a set of flights (Vi, Vj, Vk) said aircraft exhibiting deviated behavior and maintenance actions performed during said set of flights.
    [5" id="c-fr-0005]
    5. Method according to claim 1 or 2, characterized in that each sequence of training data or current includes time data from a flight recorder and / or messages indexed temporally failures from a supervision system centralized.
    [6" id="c-fr-0006]
    6. Method according to any one of claims 2 to 5, characterized in that the construction of the behavior model relating to a given parameter comprises the following steps: - collect at each of the flights of all the aircraft at least one suite learning data derived from the measurements relating to said parameter thus forming during flights of all the aircraft (16) said plurality of training data sequences; segmenting each sequence of training data into an optimal number of segments to transform it into a chain of training segments, the optimal number of segments defining said predetermined number of segments associated with said parameter; transforming said set of training data suites into a corresponding set of training segment strings; selecting a central segment string (M1) from said set of training segment strings, the central segment string defining said reference segment string; aligning the segments of said set of training segment strings (cl, ..., cp) with respect to said reference segment string; generating a learning measure vector for each training segment string by assigning predetermined characteristic measures to each learning segment and / or set of segments of each training segment string; calculating a learning distance between each learning measurement vector and a reference measurement vector associated with said reference segment string; and partitioning the set of learning measure vectors into consecutive intervals as a function of the learning distances thus defining a normal behavior interval (In) and atypical behavior intervals (la).
    [7" id="c-fr-0007]
    The method of claim 6, characterized in that said central segment chain corresponds to the medoid of said set of segment chains.
    [8" id="c-fr-0008]
    8. Method according to claim 6 or 7, characterized in that it further comprises the following steps: -save said reference segment chain, the distances between the measurement vectors and the reference measurement vector and the partitioning of said measurement vectors, - display a representative graph of the behavior model.
    [9" id="c-fr-0009]
    9. System for predicting faults in an aircraft, characterized in that it comprises a processor (3) configured to analyze a current behavior of at least one flight parameter of said aircraft to detect any deviation from said current behavior with respect to a a predetermined behavior model (14) of said parameter, said behavior model being determined from a plurality of training data sequences relating to said parameter collected during the flights of a set of aircraft.
    [10" id="c-fr-0010]
    10. Monitoring system comprising avionics maintenance and management systems characterized in that it further comprises the fault prediction system according to claim 9.
    类似技术:
    公开号 | 公开日 | 专利标题
    FR3052273A1|2017-12-08|PREDICTION OF TROUBLES IN AN AIRCRAFT
    EP2599039B1|2018-05-16|Method and system for flight data anlysis
    EP2376989B1|2019-10-16|Identification of defects in an aircraft engine
    EP2912526B1|2017-12-06|System for monitoring a set of components of a device
    CA2813556C|2019-01-15|System for monitoring an engine test bed
    AU2016287383A1|2018-01-18|Method for detecting anomalies in a water distribution system
    FR2970358A1|2012-07-13|PROGNOSTIC OF DURATION BEFORE MAINTENANCE BY FUSION BETWEEN MODELING AND SIMULATION, FOR ELECTRONIC EQUIPMENTS ON BOARD IN AN AIRCRAFT
    EP3097455B1|2019-02-20|Method for predicting an operational malfunction in the equipment of an aircraft or aircraft fleet
    EP2500267A1|2012-09-19|Monitoring of a flight control actuator of an aircraft
    EP3250974B1|2021-09-08|Method, system and computer program for learning phase of an acoustic or vibratory analysis of a machine
    FR3011105A1|2015-03-27|METHOD FOR IDENTIFYING FAILURE EQUIPMENT IN AN AIRCRAFT
    FR2933789A1|2010-01-15|METHODS OF IDENTIFYING FLIGHT PROFILES IN AIRCRAFT MAINTENANCE OPERATIONS
    CA2914819A1|2014-12-18|Methods of creating a database and of formulating a map of operating states of aircraft and a method of monitoring the operation of an associated aircraft
    US20130246860A1|2013-09-19|System monitoring
    EP3215903B1|2019-12-18|Tool for validating a system for monitoring an aircraft engine
    FR3037679A1|2016-12-23|
    EP3588387A1|2020-01-01|Method for testing an electronic air traffic control system, associated electronic device and platform
    FR2957170A1|2011-09-09|Equipment monitoring system designing tool for engine of aircraft, involves processing unit evaluating quantification of monitoring system based on quality value of result corresponding to output quality value associated with output module
    FR2986879A1|2013-08-16|METHOD AND SYSTEM FOR DETECTING ANOMALIES SOLVING IN AN AIRCRAFT
    WO2021122397A1|2021-06-24|Method and device for supervising a traffic control system
    FR3068493A1|2019-01-04|PREDICTION OF TROUBLES IN AN AIRCRAFT
    FR3098938A1|2021-01-22|Method and device for determining an anomaly prediction performance index value in an IT infrastructure from performance indicator values
    EP3443425A1|2019-02-20|Method for testing the integrity of the avionics of an aircraft, associated device and computer program product
    FR3084046A1|2020-01-24|METHOD FOR MONITORING THE CONDITION OF AN OVERHEAD CONTACT LINE AND ASSOCIATED PROGRAM DEVICE
    FR3076632A1|2019-07-12|METHOD FOR ANALYZING MAINTENANCE MESSAGES GENERATED BY AT LEAST ONE PLATFORM, AND ASSOCIATED SYSTEM
    同族专利:
    公开号 | 公开日
    CN107463161A|2017-12-12|
    CN107463161B|2021-10-12|
    FR3052273B1|2018-07-06|
    US20170352204A1|2017-12-07|
    US10360741B2|2019-07-23|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US20090249129A1|2007-10-12|2009-10-01|David Femia|Systems and Methods for Managing Multi-Component Systems in an Infrastructure|
    US20090254240A1|2008-04-07|2009-10-08|United Parcel Service Of America, Inc.|Vehicle maintenance systems and methods|
    FR2939924A1|2008-12-15|2010-06-18|Snecma|IDENTIFICATION OF FAILURES IN AN AIRCRAFT ENGINE|
    US20110172874A1|2010-01-13|2011-07-14|Gm Global Technology Operations, Inv.|Fault prediction framework using temporal data mining|
    FR2999318A1|2012-12-12|2014-06-13|Thales Sa|METHOD FOR EVALUATING THE OPERATING SAFETY OF A COMPLEX SYSTEM|WO2020115440A1|2018-12-07|2020-06-11|Safran Aircraft Engines|Computing environment system for monitoring aircraft engines|
    FR3098942A1|2019-07-17|2021-01-22|Safran Aircraft Engines|Method, device, computer program for predicting aircraft failure / anomaly in starting an aircraft|US8112368B2|2008-03-10|2012-02-07|The Boeing Company|Method, apparatus and computer program product for predicting a fault utilizing multi-resolution classifier fusion|
    US8306778B2|2008-12-23|2012-11-06|Embraer S.A.|Prognostics and health monitoring for electro-mechanical systems and components|
    US8370045B2|2009-08-14|2013-02-05|Lockheed Martin Corporation|Starter control valve failure prediction machine to predict and trend starter control valve failures in gas turbine engines using a starter control valve health prognostic, program product and related methods|
    CN101909206A|2010-08-02|2010-12-08|复旦大学|Video-based intelligent flight vehicle tracking system|
    FR2987483B1|2012-02-29|2014-03-07|Sagem Defense Securite|METHOD OF ANALYSIS OF FLIGHT DATA|
    GB2510596B|2013-02-08|2015-02-18|Ge Aviat Systems Ltd|Method for predicting a trailing edge flap fault|
    GB2513132B|2013-04-16|2015-05-27|Ge Aviat Systems Ltd|Method for predicting a bleed air system fault|
    CN103499921B|2013-09-11|2015-12-02|西安交通大学|Structure changes fuzzy system sensor fault diagnosis method|
    GB2518893B|2013-10-07|2018-11-21|Ge Aviat Systems Ltd|Method for predicting an auxiliary power unit fault|
    CN103675525B|2013-11-14|2017-01-18|南京航空航天大学|DC-DC converter health monitoring and fault prediction method based on multiple SVDD models|
    EP3108313B1|2014-02-21|2020-07-22|Taleris Global LLP|Method for predicting a fault in an air-conditioning pack of an aircraft|
    US20170052836A1|2014-02-21|2017-02-23|Taleris Global Llp|Method for predicting a fault in a cabin temperature control system of an aircraft|
    EP3140637B1|2014-05-09|2019-12-18|Sikorsky Aircraft Corporation|Method and apparatus for condition based maintenance of fiber networks on vehicles|
    KR20160025664A|2014-08-27|2016-03-09|삼성에스디에스 주식회사|Apparatus and method for early detection of abnormality|
    US9435661B2|2014-10-08|2016-09-06|Honeywell International Inc.|Systems and methods for attitude fault detection based on air data and aircraft control settings|
    US10417251B2|2014-10-31|2019-09-17|The Boeing Company|System and method for storage and analysis of time-based data|
    US9550583B2|2015-03-03|2017-01-24|Honeywell International Inc.|Aircraft LRU data collection and reliability prediction|
    CN105427674B|2015-11-02|2017-12-12|国网山东省电力公司电力科学研究院|A kind of unmanned plane during flying state assesses early warning system and method in real time|
    US20170261406A1|2016-03-10|2017-09-14|Simmonds Precision Products, Inc.|Physical component fault diagnostics|
    US20170293517A1|2016-04-11|2017-10-12|Simmonds Precision Products, Inc.|Physical component predicted remaining useful life|US10587635B2|2017-03-31|2020-03-10|The Boeing Company|On-board networked anomaly detectionmodules|
    US10372389B2|2017-09-22|2019-08-06|Datamax-O'neil Corporation|Systems and methods for printer maintenance operations|
    CN107992077A|2017-12-13|2018-05-04|北京小米移动软件有限公司|Aircraft fault rescue method and device|
    CN108957173A|2018-06-08|2018-12-07|山东超越数控电子股份有限公司|A kind of detection method for avionics system state|
    US20200151967A1|2018-11-14|2020-05-14|The Boeing Company|Maintenance of an aircraft|
    US20200380295A1|2019-05-29|2020-12-03|Nec Laboratories America, Inc.|Failure prediction using gradient-based sensor identification|
    FR3101669A1|2019-10-07|2021-04-09|Safran|Aircraft engine monitoring device, method and computer program|
    法律状态:
    2017-06-21| PLFP| Fee payment|Year of fee payment: 2 |
    2017-12-08| PLSC| Search report ready|Effective date: 20171208 |
    2018-06-26| PLFP| Fee payment|Year of fee payment: 3 |
    2020-06-19| PLFP| Fee payment|Year of fee payment: 5 |
    2021-06-22| PLFP| Fee payment|Year of fee payment: 6 |
    优先权:
    申请号 | 申请日 | 专利标题
    FR1655040A|FR3052273B1|2016-06-02|2016-06-02|PREDICTION OF TROUBLES IN AN AIRCRAFT|
    FR1655040|2016-06-02|FR1655040A| FR3052273B1|2016-06-02|2016-06-02|PREDICTION OF TROUBLES IN AN AIRCRAFT|
    US15/606,917| US10360741B2|2016-06-02|2017-05-26|Predicting failures in an aircraft|
    CN201710405964.8A| CN107463161B|2016-06-02|2017-05-26|Method and system for predicting a fault in an aircraft and monitoring system|
    [返回顶部]