Method and system for determining a faulty component of an aircraft

Abstract

The invention relates to a digital system that makes it possible to determine at least one faulty component of an aircraft. The system comprises: - a first module (MDS1) for determining at least a first symptom (S1 i) of the aircraft on the basis of state data (DEk) collected in the vicinity of the aircraft; - a second module (MDS2) for determining at least a second symptom (S2 i) of the aircraft on the basis of degradation data (DCk) representative of an abnormal variation in flight data; - a module (MDPD) for determining probabilities of failure in a plurality of components of the aircraft on the basis of the at least first and second symptoms, this module comprising a Bayesian network (BR) trained using a knowledge base (BC) comprising data produced by a natural language processing module (MNLP) configured to process aircraft maintenance reports; - a module (MSAC) configured to flag at least one faulty component from among the plurality of components according to the probabilities.

Description

Description Title of the invention: Method and system for determining a faulty component of an aircraft 5 Prior art

[0001] The present invention relates to the general field of aeronautics.

[0002] It relates more particularly to a method and a decision support system for identifying faulty components of an aircraft. io[OOO3] We can consider that an aircraft is a complex system made up of a certain number of hardware or software components (computer, sensors, etc.) and that these are likely to fail.

[0004] When a failure is suspected, it is common for an operator to move close to the aircraft (also called "under wing") to carry out tests in order to 15 to try to determine the faulty component(s) causing the breakdown.

[0005] In practice, the on-wing operator follows a diagnostic scenario comprising a linear list of instructions, for example presented in the form of a troubleshooting flowchart in a maintenance document in paper or digital form. These maintenance documents have a very long update cycle, and 20 are published approximately twice a year.

[0006] When the operator suspects that one or more components are faulty, he removes these components to send them to a repair shop for further analysis, using industrial testing equipment.

[0007] In order to intervene quickly or due to insufficient training, 25 it is common for the under-wing operator to remove an excessive number of components. Furthermore, it may turn out, during the workshop analysis phase, that the components supposedly faulty are not the real cause of the breakdown and that they were therefore wrongly removed.

[0008] For example, an operator on the wing may assume that a sensor is faulty, remove the sensor, and send it to the shop for repair. By analyzing the supposedly faulty sensor in the workshop, it can be determined that the sensor is functioning normally and that the real cause of the failure is in fact an electronic board configured to process the data measured by the sensor.*

[0009] A maintenance report is then drawn up, this report including by 5 example of alert messages or fault alarms, results of measurements carried out by the on-wing operator or by the repairer in the workshop, symptoms detected by the on-wing operator or by the repairer, the components determined to be faulty and the real cause of the fault determined after analysis. This maintenance report can for example be digital. io[OO1O] In the aeronautics industry in particular, there is a known reliability indicator NFF (in English No Fault Found) which represents a rate of unplanned and unjustified removals calculated from the ratio between the mean time between unplanned removals MTBUR (in English Mean Time Between Unscheduled Removal) and the net mean time between unplanned, justified removals MTBF (in 15 English Mean Time Between Failure). These indicators verify the following equality: MTBUR=(1 -NFF). MTBF.

[0011] It is known that the cost of these unplanned NFF removals is very high and that it has a significant impact on the operational costs of airlines.

[0012] The invention aims to overcome at least some of these drawbacks by proposing a failure determination system which allows the on-wing operator to better identify the real cause of a failure of an aircraft, consequently the component(s) to be removed, in particular so as to improve the reliability indicator NFF. Statement of the invention

[0013] The present invention responds in particular to this need by proposing a digital system for determining at least one faulty component of an aircraft, this system comprising: - an aircraft failure information module; - a first module for determining at least one first symptom of the aircraft from state data collected under the wing; 30 - a second module for determining at least one second symptom of the aircraft from degradation data representative of an abnormal variation in flight data; - a module for determining probabilities of failure of a plurality of components of the aircraft from said at least one first and second symptom, this module comprising a Bayesian network trained from a knowledge base comprising data produced by a natural language processing module configured to process aircraft maintenance reports, - a module configured to signal at least one faulty component among said plurality of components according to said probabilities.

[0014] Correlatively, the invention relates to a method for determining at least one faulty component of an aircraft, this method implemented by computer comprising the following steps: - obtaining information about an aircraft failure; - determination of at least one first symptom of the aircraft from condition data collected under wing; - determination of at least one second symptom of the aircraft from degradation data representative of an abnormal variation in flight data; - determining probabilities of failure of a plurality of components of the aircraft from said at least one first and second symptom, this determination using a Bayesian network trained from a knowledge base comprising data produced by a natural language processing module configured to process aircraft maintenance reports, - reporting at least one faulty component among said plurality of components according to said probabilities.

[0015] Thus, and very advantageously, the digital system determines the faulty components using both "state data" and "degradation data".

[0016] Status data represents the state of the aircraft. This data can be collected by an operator (or a robot), for example with measuring equipment, operating close to the aircraft, i.e. "under the wing". It is, for example, fed back by a human-machine interface via alarms or fault messages. It can sometimes be directly determined or observed by the operator, when the latter detects a noise or odor characteristic of a malfunction.

[0017] On the contrary, degradation data, or behavioral data, is data that represents a trend, or an abnormal variation in flight data. It can be determined for example: - when flight data evolves and exceeds a threshold; - when a signature of a series of flight data is representative of a degradation in the operation of the aircraft; - by processing flight data continuously.

[0018] Unlike condition data, these degradation data are difficult for the operator to access. However, they are very rich in information and taking them into account makes it possible to overcome the limits of knowledge in terms of modeling the physical phenomena that cause failures.

[0019] Very advantageously, the status data as well as the behavioral data are used to determine symptoms (first and second symptoms within the meaning of the invention) which are processed in the same way by the module for determining the probabilities of failure of the components.

[0020] Furthermore, and very advantageously, the determination of the probabilities of failure of aircraft components uses a Bayesian network trained from a knowledge base comprising data produced by a natural language processing module configured to process aircraft maintenance reports.

[0021] Taking into account behavioral data to determine aircraft symptoms and feedback in the Bayesian network makes it possible to improve the modeling of physical phenomena and consequently the modeling of degradation phenomena.

[0022] The system and the determination method thus make it possible to improve the diagnosis leading to the identification of the components which are the real causes of failure and consequently to reduce the direct and indirect costs associated with poor fault-finding diagnosis during aircraft ground maintenance.

[0023] The invention thus makes it possible to improve the detectability of failures in systems and their equipment, to put the aircraft back into service more quickly and therefore to reduce the time of unplanned operational unavailability.

[0024] In one embodiment, said at least one second symptom is a signature determined from one or more indicators positioned when the second determination module detects behavioral data representative of a trend of degradation in the operation of at least one component of the aircraft.

[0025] For example, a signature can be determined from an indicator representative of a degradation of a first component and an indicator representative of a degradation of a second component, these components possibly being included in different subsystems of the aircraft. io[OO26] This embodiment advantageously makes it possible to model symptoms which result from interactions between subsystems of the aircraft, even when these interactions are difficult to understand or model. For example, the mechanical vibrations of a rotating assembly can generate vibrations and degrade equipment of another system which is not directly related to the rotating system.

[0027] The invention also relates to a method and a device for training a Bayesian network which can be used in a system according to the invention.

[0028] Thus, and according to a second aspect, the invention also relates to a method for training a Bayesian network intended to be implemented by a module 20 configured to determine probabilities of failure of a plurality of components of an aircraft from at least a first and at least a second symptom of the aircraft, said Bayesian network being trained from a knowledge base comprising data produced by a natural language processing module configured to process aircraft maintenance reports, in 25 which: - said at least one first symptom of the aircraft is determined from condition data collected under wing; and - said at least one second symptom of the aircraft is determined from degradation data representative of an abnormal variation in flight data. 30

[0029] Correlatively, the invention also relates to a device for training a Bayesian network intended to be implemented by a module configured to determine probabilities of failure of a plurality of components of a aircraft from at least a first and second aircraft symptom, said Bayesian network being trained from a knowledge base comprising data produced by a natural language processing module configured to process aircraft maintenance reports, wherein: - said at least one first aircraft symptom is determined from condition data collected under wing; and - said at least one second symptom of the aircraft is determined from degradation data representative of an abnormal variation in flight data.

[0030] In a particular embodiment, the natural language processing module used in the invention is a module of the training device.

[0031] In one embodiment, this natural language processing module takes as input texts extracted from aircraft maintenance reports and produces as output labels which constitute structured information used by an enrichment module to enrich the knowledge base. [OO32] In one embodiment of the invention, the knowledge base comprises records which associate symptoms with failures. The natural language processing module produces: (i) at least one first label representing at least one of flight data, status data, behavioral data and / or symptoms; and (ii) at least one second label representing a failure.

[0033] In one embodiment, the knowledge base enrichment module is configured to enrich the knowledge base with at least one record comprising: (i) at least one symptom identified from a first label, associated with (ii) a failure identified from a second label.

[0034] In a particular embodiment, the different steps of the method for determining faulty components and the main steps of the method for training the Bayesian network are determined by computer program instructions.

[0035] Consequently, the invention also relates to a computer program on an information medium, this program being capable of being implemented in a computer, this program comprising instructions adapted to the implementation of the steps of a method as described above.

[0036] This program can use any programming language, and be in the form of source code, object code, or intermediate code between code 5 source and object code, such as in partially compiled form, or in any other desirable form.

[0037] The invention also relates to a computer-readable information medium, and comprising instructions of a computer program as mentioned above. io[OO38] The information medium may be any entity or device capable of storing the program. For example, the medium may comprise a storage means, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, or a magnetic recording means, for example a hard disk. 15[OO39] On the other hand, the information carrier may be a transmissible carrier such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The program according to the invention may in particular be downloaded from a network such as the Internet.

[0040] Alternatively, the information carrier may be an integrated circuit in 20 in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the method in question. Brief description of the drawings

[0041] Other characteristics and advantages of the present invention will emerge from the description given below, with reference to the appended drawings which illustrate an exemplary embodiment thereof without any limiting character. In the figures: [Fig. 1] Figure 1 represents a system for determining faulty components in accordance with the invention; [Fig. 2] Figure 2 represents the hardware architecture of a terminal conforming to a particular embodiment of the invention; [Fig. 3] Figure 3 represents the hardware architecture of a training device according to a particular embodiment of the invention; [Fig. 4] Figure 4 represents, in the form of a flowchart, the main steps of a training method according to the invention; and [Fig. 5] Figure 5 represents, in the form of a flowchart, the main steps of a method for determining faulty components according to the invention. Detailed description of embodiments

[0042] Figure 1 represents a digital system SYS for determining faulty components in accordance with an embodiment of the invention.

[0043] This system includes a first MDSi module for determining initial symptoms.

[0044] When an aircraft problem or failure is suspected, an operator moves close to the aircraft (or "under wing") and performs a number of tests, to obtain DE status data k of the aircraft, and deduce from the first symptoms S 1 i using this first MDSi symptom determination module.

[0045] In practice, the first MDSi symptom determination module may be a module integrated into a TRM terminal of the OP operator. In the embodiment of FIG. 1, the TRM terminal is configured to interface with a BDT test scenario base, which comprises tests T1, T2, T3 to be performed for a predetermined fault detection SDP scenario.

[0046] In the embodiment described here, each test T1, T2, T3 allows the determination (for example by calculation or by measurement) of one or more state data DE k .

[0047] The first symptoms S 1 i can be obtained from a BDS database that associates at least one first symptom S 1 i to a combination of at least one state data DE k . In the example in Figure 1: - DE status data 1 ; DE2 are representative of a first symptom S 1 i ; - state data OF 1 ; DE3 are representative of a first symptom S 1 2; and - DE2 status data are representative of a first symptom S 1 3.

[0048] DE state data kare for example the result of observations by the operator (odor, part inspection, ...), results of measurements carried out by the operator under wing, or data obtained by the operator, for example data provided from an aircraft CALC computer to the TRM terminal via a 5 I / O input / output interfaces.

[0049] In the described embodiment, the SYS system comprises a second MDS2 module for determining symptoms S 2 i, from behavior data DCk representative of a trend of degradation of the operation of at least one component of the aircraft. io[OO5O] In the embodiment described here, this second module MDS2 is a module of the CALC computer of the aircraft.

[0051] In the embodiment described herein, at least one symptom S 2determined by the second MDS2 symptom determination module is constituted by a SIG fault signature calculated from the DC behavior data k . 15[OO52] More specifically, in the embodiment described here, the second symptom determination module MDS2 comprises at least one processing module MTi configured to process flight data DV made available by sensors or by electronic systems of the engines or the aircraft and to position indicators INDi when they detect data DC k of 20 behaviors representative of a trend of degradation in the functioning of at least one component.

[0053] Such an MTi processing module is, for example, configured to track a flight data trend over a given period or to detect a shape in a representation of this flight data. 25

[0054] For example, an MTi processing module sets a first INDi flag when it detects DC data k representative of an increase in pressure or temperature in a component of the aircraft, or more generally the exceeding of a normality threshold by an operating variable of a component of the aircraft.

[0055] For example, an MT processing module sets a second INDj flag when it detects DC data. k representative of a particular vibratory phenomenon.

[0056] In the embodiment described here, the second symptom determination module MDS2 generates a SIG signature from these IN Di indicators and this SIG signature is treated as a symptom S 2 i in the same way as the S symptoms 1 i determined by the first MDSi symptom determination module from DE status data k .

[0057] For example, the second MDS2 symptom determination module has LOG logic configured to generate: - a first SIG signature 2 I when the first indicator INDi is set but not the second indicator INDj; - a second SIG signature 2 2when the second indicator INDj is set but not the first indicator INDi; - a third SIG signature 2 3when both INDi and INDj indicators are set.

[0058] The SYS system includes an MDPD module for determining the probability of failure from: - first symptoms S 1 i determined from the DE state data k ; and - second symptoms S 2 determined from behavioral data DCk.

[0059] In the embodiment described herein, the MDPD failure probability determination module uses a Bayesian RB network.

[0060] Depending on the presence or absence of first and / or second symptoms, the Bayesian network RB model dynamically updates the probabilities associated with each branch of a decision tree, and thus directs the diagnosis towards identifying the root cause of the breakdown.

[0061] Thus, the symptoms and failures included in the knowledge base constitute the training data of the Bayesian network. The network is trained to predict, during network inference, failures from symptoms provided at the network input, for example determined under wing.

[0062] During system use, the Bayesian network probabilities are updated to represent the probability of symptoms causing a given failure, for each failure, using the associations between symptoms and failures in the knowledge base.

[0063] In other words, the Bayesian network learns as it is used and as the knowledge acquired during maintenance actions in search of faults increases.

[0064]

[0065] In the embodiment described herein, the MDPD failure probability determination module is integrated into the operator's TRM terminal.

[0066] This Bayesian network RB is configured to determine, from the first and second symptoms, the probability of a Dj failure. [OO67] For example, a Dj failure can be: - the malfunction of a sensor; or - a torn harness; or - a software problem.

[0068] In the embodiment described here, this Bayesian network RB is trained from a knowledge base BC which comprises records, each associating, with at least one symptom S, a failure D. This knowledge base BC is for example initialized from theoretical data from the AMDEC (Analysis of Failure Modes, their Effects and their Criticality) of the different components.

[0069] This BC knowledge base can also be enriched by feedback from the use of the SYS digital system. For example, if it is detected that combinations of first symptoms S 1 i and second symptoms S 2 i particulars determined by the CAL calculator from behavioral data DCk are representative of a failure D k, the knowledge base BC can be updated with this new teaching to retrain the Bayesian network RB.

[0070] In the embodiment described here, this knowledge base BC includes records from feedback determined by natural language analysis of documents dealing with real causes of failure.

[0071] In the embodiment described here, the various documents available 5 which deal with the real causes of aircraft failure are analyzed by a Natural Language Processing (NLP) MNLP module. In the embodiment described here, the SYS system includes a DENT training module which includes the Natural Language Processing (NLP) MNLP module. io[OO72] These documents include, for example, maintenance or troubleshooting reports provided by airlines. In practice, for a given airline, the number of these reports can be very high, for example several thousand per month. They are most often written in human language: the operator describes a situation and details the actions 15 fixes made.

[0073] The invention advantageously makes it possible to automatically process these reports to enrich the knowledge base BC of the Bayesian network RB.

[0074] In the embodiment described here, the natural language processing module MNLP is configured to receive unstructured DNS documents 20 as input and to provide relevant information structured in the form of LAB labels as output, used to feed the feedback data REX from the knowledge base BC used for training the Bayesian network RB.

[0075] In the embodiment described herein, the MNLP natural language processing module uses an NLP learning model. As is known, the 25 developing an NLP model essentially consists of: - build a learning base starting with the cleaning of the input data and leading to a labeling of the input data; - implement the NLP model; and - validate the NLP model from a statistical and business point of view. 30

[0076] In the context of the invention, the input data of the MNLP module are maintenance report texts and the LAB labels are, for each report: - at least one first label representing at least one piece of information among flight data, status data, behavioral data and / or symptoms; and - at least a second label representing a failure.

[0077] In the embodiment described here, the DENT training device of the SYS system comprises a MEBC module for enriching the knowledge base with at least one record comprising: (i) at least one symptom identified from a first label, associated with: (ii) a failure identified from a second label.

[0078] In the embodiment described here, the MEBC knowledge base enrichment module uses: - the BDS symptom database to identify first symptoms from first LAB labels, representative of state data; - the second MDS2 module for determining second symptoms to identify second symptoms from first LABj labels representing flight or behavior data.

[0079] For example, with reference to Figure 1, if the MNLP natural language processing module produces a first label “DE2” (representative of state data) and a second label “D5” (representative of a failure) from the unstructured information included in the maintenance reports, the MEBC knowledge base enrichment module: - determines from the third record of the BDS database that the status data DE2 is representative of a symptom S3; and - enriches the BC knowledge base with a record that associates the symptom S3 and the failure D5.

[0080] The development of an NLP model includes for example: - the construction of a learning base which includes the cleaning of input data (management of punctuation, accents, synonyms and antonyms, removal of multiple spaces, transformation of uppercase letters into lowercase letters, etc.); - the extraction of information (motifs or patterns) corresponding to regular expressions; - lemmatization of extracted information; - cutting (or in English tokenization); - spelling correction; - removal of non-discriminatory words (in English stop words).

[0081] The SYS system comprises an MSAC module configured to signal at least one faulty component among the plurality of components of the aircraft based on the failure probabilities calculated by the MDPD module, and to propose an appropriate corrective action.

[0082] The operator can carry out a corrective action Cj proposed by the digital system to be carried out under wing, for example the removal of a component to be sent to the workshop for analysis and repair

[0083] In the embodiment described herein, the MSAC component failure reporting module is integrated into the operator's TRM terminal.

[0084] For example, if it is determined by the MDPD failure probability determination module that the most likely failure is sensor wear, the MSAC module can signal to the operator that the sensor has failed and suggest a corrective action Ci consisting of replacing the sensor.

[0085] In one embodiment, the system for determining at least one faulty aircraft component is implemented in a TRM terminal.

[0086] In the embodiment described here, and as shown in Figure 2, this TRM terminal has the hardware architecture of a computer. It comprises a processor 10A, a random access memory 10B, a read only memory 10C, a non-volatile flash memory 10D, input / output means 10E as well as communication means 10F.

[0087] These means of communication 10F allow in particular the terminal TRM 10 to obtain status data DE k from the aircraft's CALC computer, or from externally configured measuring devices to perform under-wing tests.

[0088] In one embodiment, these under-wing tests can be carried out by dedicated modules integrated into the TRM terminal.

[0089] In one embodiment, the second module MDS2 for determining at least one second symptom S 2 i is integrated into the TRM terminal. This module is configured to obtain DV flight data from the aircraft, and to determine the less one SIG fault signature from DC degradation data k representative of an abnormal variation in these flight data.

[0090] The read-only memory 10C constitutes a recording medium in accordance with the invention, readable by the processor 10A and on which a computer program PROG is recorded. T according to the invention, comprising instructions for executing the steps of a method for determining at least one faulty aircraft component according to the invention and which will be described with reference to Figure 5.

[0091] This PROGT computer program defines functional and software modules configured here to implement the steps of the determination method according to the invention. These functional modules rely on or control the hardware elements 10A to 10D of the TRM terminal. They include in particular: - a first MDSi module for determining at least one first symptom S 1 i of the aircraft from DE state data k collected under wing; - the second MDS2 module for determining at least one second symptom S 2 i of the aircraft; - an MDPD module for determining probabilities of failure of a plurality of components of the aircraft from at least a first and second symptom, this module comprising a Bayesian network RB trained from a knowledge base BC comprising data produced by a natural language processing module configured to process aircraft maintenance reports, and - an MSAC module configured to signal at least one faulty component among said plurality of components according to said probabilities.

[0092] Figure 3 represents the hardware architecture of a DENT drive device according to the invention. This device has the hardware architecture of a computer and comprises a processor 20A, a random access memory 20B, a read only memory 20C, a non-volatile flash memory 20D, input / output means 20E as well as communication means 20F.

[0093] The read-only memory 20C of the drive device constitutes a recording medium in accordance with the invention, readable by the processor 20A and on which a computer program PROG is recorded. E according to the invention, comprising instructions for carrying out the steps of a training method according to the invention and which will be described with reference to figure 4.

[0094] This DENT device is configured to enrich the knowledge base BC used for training the Bayesian network RB used by the MDPD module to determine the failure probabilities of a plurality of aircraft components.

[0095] The PROGE computer program defines functional and software modules configured here to implement the steps of the drive method according to the invention. These functional modules rely on or control the hardware elements 20A to 20E of the DENT drive device.

[0096] They include in particular a natural language processing module MNLP configured to process aircraft maintenance reports DNS and produce relevant structured information IS which feeds the feedback data REX from the knowledge base BC used to train the Bayesian network RB.

[0097] In one embodiment of the invention, this structured information consists of: (i) first labels representing information from among flight data, state data, behavior data and / or symptoms; and (ii) second labels representing failures.

[0098] In the embodiment described here, the functional modules of the DENT device comprise an enrichment MEBC module configured to enrich the BC knowledge base with at least one record comprising: (i) at least one symptom identified from a first label; associated with (ii) a failure identified from a second label.

[0099] In the embodiment described here, the MEBC module determines at least some of these symptoms by using the BDS symptom database to identify first symptoms from first LAB labels, representative of state data and / or the second MDS2 second symptom determination module to identify second symptoms from first LAB labels, representing flight or behavior data.

[0100] Figure 4 represents the main steps of a training method according to the invention.

[0101] It includes a step E10 of analyzing aircraft maintenance reports using a natural language processing method to produce structured data in the form of LABj labels.

[0102] In the embodiment described here, these LAB labels represent: - information among flight data, state data, behavior data and / or symptoms; and - failures.

[0103] In the embodiment described here, the training method comprises a step E20 for enriching the knowledge base BC of the Bayesian network RB from: - first symptoms determined from LAB labels, representing state data and the BDS symptom database; - second symptoms determined from LAB labels, representing flight or behavior data and the second MDS2 module for determining the second symptom; - LAB labels, representative of failures.

[0104] In the embodiment described here, the training method comprises a step E30 of training the Bayesian network RB from this knowledge base so as to update the probabilities of this network to improve the diagnosis of identification of the root cause of the breakdown thanks to this structured information obtained by NLP processing of the unstructured information contained in the maintenance and troubleshooting reports.

[0105] Figure 5 represents the main steps of a method for determining faulty components in accordance with the invention.

[0106] It includes: - a step F10 of determining at least one first symptom S 1 i of the aircraft from DE state data k collected under wing; - a step F20 of determining at least one second symptom S 2 i of the aircraft from DC degradation data krepresentative of an abnormal variation in flight data; - a step F30 of determining probabilities of failure of a plurality of components of the aircraft from at least a first and second symptom, this determination step using a Bayesian network RB trained from a knowledge base BC comprising data produced by a natural language processing module configured to process aircraft maintenance reports, - a step F40 of reporting at least one faulty component among said plurality of components according to said probabilities.

[0107] This reporting allows the operator to take corrective action with a high probability that it is actually appropriate to the real cause of the aircraft failure.

Claims

Claims

1. Digital system for determining at least one faulty component of an aircraft, this system comprising: 5 - a first module (MDSi) for determining at least one first symptom (S 1 !) of the aircraft from state data (DE k ) collected under wing; - a second module (MDS2) for determining at least one second symptom (S 2 ) of the aircraft from degradation data (DC k) representative of an abnormal variation in flight data; 0 - a module (MDPD) for determining probabilities of failure of a plurality of components of the aircraft, this module comprising a Bayesian network (RB) configured to carry out said determination of probabilities from said at least one first and second symptom, said Bayesian network (RB) being trained from a knowledge base (BC) comprising data 5 produced by a natural language processing module (MNLP) configured to process aircraft maintenance reports, - a module (MSAC) configured to signal at least one faulty component among said plurality of components according to said probabilities. o

2. Determination system according to claim 1, characterized in that said natural language processing module (MNLP) takes as input texts extracted from said aircraft maintenance reports and produces as output labels (LABj) representing: - information from flight data, status data, behavioral data and / or symptoms; and - failures.

3. Determination system according to claim 1 or 2, characterized in that said at least one second symptom (S 2 ) is a signature0 determined from at least one indicator (INDj) set when the second determination module (MDS2) detects data (DC k ) of behavior representative of a trend of deterioration in the operation of at least one component of the aircraft.

4. Method for determining at least one faulty component of an aircraft, this method implemented by computer comprising the following steps: - determination (F10) of at least one first symptom (S 1 !) of the aircraft from state data (DE k ) collected under wing; - determination (F20) of at least a second symptom (S 2 ) of the aircraft from degradation data (DC k ) representative of an abnormal variation in flight data; - determination (F30) of probabilities of failure of a plurality of components of the aircraft from said at least one first and second symptom, this determination using a Bayesian network (RB) trained from a knowledge base (BC) comprising data produced by a natural language processing module (MNLP) configured to process aircraft maintenance reports, - reporting (F40) at least one faulty component among said plurality of components according to said probabilities.

5. Method for training a Bayesian network (RB) intended to be implemented by a module (MDPD) configured to determine probabilities of failure of a plurality of components of an aircraft, said determination being carried out by said Bayesian network (RB) from at least a first and at least a second symptoms of the aircraft, said Bayesian network (RB) being trained (E30) from a knowledge base (BC) comprising data produced by a natural language processing module (MNLP) configured to process (E10, E20) aircraft maintenance reports, in which: - said at least one first symptom (S 1 !) of the aircraft is determined from state data (DE k ) collected under wing; and - said at least one second symptom (S 2 ) of the aircraft is determined from degradation data (DCk) representative of an abnormal variation in flight data.

6. Device (DENT) for training a Bayesian network (RB) intended to be implemented by a module (MDPD) configured to determine probabilities of failure of a plurality of components of an aircraft from at least a first and at least a second symptom of the aircraft, said determination being carried out by said Bayesian network (RB) from at least a first and at least a second symptom of the aircraft, said Bayesian network (RB) being trained from a knowledge base (BC) comprising data produced by a natural language processing module (MNLP) configured to process aircraft maintenance reports, in which: - said at least a first symptom (S*) of the aircraft is determined from state data (DE k ) collected under wing; and - said at least one second symptom (S 2 (i) of the aircraft is determined from degradation data (DC k) representative of an abnormal variation in flight data.

7. Training device (DENT) according to claim 6 characterized in that said natural language processing module (MNLP) takes as input texts extracted from said aircraft maintenance reports and produces as output: (i) at least one first label (LABj) representing one of flight data, state data, behavior data and / or symptoms; and (ii) at least a second label (LABj) representing a failure, the training device (DENT) comprising a module (MEBC) for enriching the knowledge base (BC) with: (i) at least one symptom (S 1 !, S 2 ) identified from a said first label, associated with: (ii) a failure identified from a said second label.

8. Computer program (PROGT) comprising instructions for executing the steps of the method for determining at least one faulty component of an aircraft according to claim 4 when said program is executed by a computer.

9. Computer program (PROGE) comprising instructions for executing the steps of the training method according to claim 5 when said program is executed by a computer.

10. Medium (10C, 20C) comprising a computer program according to claim 8 or according to claim 9.

Images