Method for detecting malfunctions of an integrated drive generator in an aircraft

Description

TECHNICAL FIELD

[0001] The present invention relates to the field of detecting malfunctions suffered by an integrated drive generator (IDG) in an aircraft. STATE OF PRIOR ART

[0002] In an aircraft, an IDG is a system associated with each propulsion engine (turbomachine) to convert the kinetic energy of said propulsion engine into electrical energy to provide electricity to the aircraft. The IDG potentially provides electrical power to all electrical loads on the aircraft.

[0003] The IDG consists of an alternator and a constant speed drive (CSD) mounted in a single housing. The CSD is a mechanical device that provides a constant rotation frequency at the alternator input despite variations in the propulsion engine's rotation speed. Thus, the IDG can deliver a fixed frequency (e.g., 400 Hz) to the aircraft's electrical network.

[0004] Oil is a vital component of the IDG, and matching the oil's characteristics, including its temperature, to specifications is critical to the longevity and efficiency of the IDG. Otherwise, the IDG is considered to be experiencing a malfunction, which should be detected.

[0005] The oil used to cool the IDG circulates in a closed circuit and is cooled, via an oil cooler, by the fuel supplying the propulsion engine to which the IDG is connected. Similarly, a heat exchanger is used to cool the oil used to lubricate the propulsion engine itself. The heat exchanges required to cool these oils therefore depend on the flow of fuel supplying the propulsion engine in question.

[0006] However, as during certain phases of flight, and more particularly during the descent phase, the propulsion engine's fuel requirement is lower, the fuel supplying the propulsion engine may not be sufficient to cool these oils, which reduces the longevity and performance of the IDG.

[0007] To address this, a Fuel Return Valve (FRV) is used to increase fuel flow to maintain cooling. The FRV controls the flow of fuel from a propulsion engine fuel delivery system to a recirculation system that returns the hot fuel to the aircraft's fuel tank. When the FRV is in the closed position, the recirculation system is inhibited.

[0008] It is clear from the above that when the FRV malfunctions, whether due to a failure of the FRV itself or a failure of another organ with which the IDG interacts, directly or indirectly, such as a sensor or the FRV, the IDG itself malfunctions, and the longevity and performance of the IDG may be reduced.

[0009] It is therefore desirable to overcome these disadvantages of the state of the art. In particular, it is desirable to provide a solution that allows the detection of a malfunction experienced by the IDG. It is also desirable to provide a solution that allows the identification of the cause of the detected malfunction. It is also desirable to provide a solution that is simple and inexpensive.

[0010] According to its abstract, document US2020382031 A1 discloses a method for predicting the health of an integrated drive generator (IDG) comprising determining an effective deviation between a plurality of output frequencies of the IDG for a given operating period of the IDG. The method comprises correlating the effective deviation to a capacity of the integrated drive generator to determine the health of the integrated drive generator. A system for predicting the health of an integrated drive generator (IDG) comprises an IDG and a generator control unit (GCU) operatively connected to the IDG to determine a plurality of output frequencies of the IDG for a given operating period of the IDG. The system comprises a central processing unit (CPU) operatively connected to the generator control unit to receive the generator output frequencies.The central processing unit is configured and adapted to determine an effective deviation across at least a portion of the plurality of output frequencies of the IDG for the given operating period of the IDG, and to correlate the effective deviation to a capacity of the IDG to determine the health status of the IDG.

[0011] According to its abstract, document EP3460611 A1 discloses an aircraft fault detection system comprising at least one aircraft data recording device configured to capture parametric flight data from at least one aircraft subsystem, and an aircraft controller coupled to the data recording device.

[0012] According to its abstract, the paper XP055706551 "Fault Diagnosis Via Univariate Frequency Analysis Monitoring: A Novel Technique Applied to a Simulated Integrated Drive Generator" by Brian Keith Bailey (https: / / trace.tennessee.edu / utk_gradthes / 1129) discloses a fault detection and diagnosis method that would detect and isolate seed faults in data generated from a simulated integrated drive generator. STATEMENT OF THE INVENTION

[0013] For this purpose, a method is proposed for detecting a malfunction suffered by an integrated drive generator, called an IDG to be monitored, in an aircraft, the method being implemented by a data analysis system comprising machine learning type artificial intelligence. The method comprises: collecting reference data relating to measurements carried out and recorded during previous flights of aircraft equipped with IDGs, called reference IDGs, the reference data comprising at least measurements corresponding to the following parameters, for each reference IDG: fuel flow rate for supplying a propulsion engine with which the reference IDG in question is associated; position of a fuel return valve associated with the reference IDG in question; temperature measured by a sensor at the inlet port of the reference IDG in question; temperature measured by a sensor at the outlet port of the reference IDG in question;frequency of electrical signal produced by the reference IDG in question; oil temperature supplying the propulsion engine with which the reference IDG in question is associated. The method further comprises: carrying out training, in supervised mode, of the machine learning type artificial intelligence using the reference data, with a first classification pass for training to detect possible malfunctions suffered by IDGs and a second classification pass for training to determine causes of malfunctions where applicable; putting the machine learning type artificial intelligence into production after validation of the training; collecting data to be analyzed, of the same type as the reference data, relating to measurements taken and recorded during flights of the aircraft comprising the IDG to be monitored;using the trained machine learning type artificial intelligence, and thus predicting a possible malfunction suffered by the IDG to be monitored and predicting, if applicable, a cause of the malfunctions, thanks respectively to said first and second passes, from the data to be analyzed. Thus, the IDG malfunction can be easily predicted and, if applicable, its cause can also be easily predicted.;

[0014] In a particular embodiment, the reference data further comprises measurements corresponding to at least one of the following parameters, for each reference IDG: duration during which the fuel return valve is open; duration during which the fuel return valve is closed; ratio between the duration during which the fuel return valve is closed and the duration during which the fuel return valve is open; duration taken by the fuel return valve to move from the closed position to the open position; temperature difference between the inlet port of the reference IDG in question and the outlet port of the reference IDG in question.

[0015] In a particular embodiment, the data analysis system applies processing to the reference data and to the data to be analyzed, the processing applied corresponding at least to the calculation of the following information, possibly in combination, for one or more of said parameters: minimum value; maximum value; standard deviation; average; sliding average; slope direction coefficient.

[0016] In a particular embodiment, the calculation of at least part of said information is carried out by grouping reference data, and respectively data to be analyzed, over several consecutive aircraft flights.

[0017] In a particular embodiment, the machine learning type artificial intelligence is trained to detect an IDG malfunction related to: a fuel return valve opening / closing control anomaly; an IDG oil clogging; an IDG outlet port temperature sensor failure; an IDG inlet port temperature sensor failure; an IDG oil leak; an IDG maintenance anomaly.

[0018] In a particular embodiment, the first classification pass is performed by applying a first algorithm and the second classification pass is performed by applying a second algorithm distinct from the first algorithm.

[0019] In a particular embodiment, the first algorithm is of the LightGBM type and the second algorithm is of the XGBooost type.

[0020] Also provided is a computer program product, which may be stored on a medium and / or downloaded from a communications network, for reading by a processor. This computer program comprises instructions for implementing the above-mentioned method in any of its embodiments, when said program is executed by the processor. Also provided is a non-transitory information storage medium on which such a computer program is stored.

[0021] In a particular embodiment, the machine learning type artificial intelligence is put into production in avionics of the aircraft equipped with the IDG to be monitored.

[0022] In a particular embodiment, the machine learning type artificial intelligence is put into production in ground computing equipment in a control center, and the data to be analyzed is transmitted by avionics of the aircraft equipped with the IDG to be monitored to the computing equipment of the control center.

[0023] A data analysis system comprising machine learning-type artificial intelligence is also proposed, the system comprising electronic circuitry configured to perform malfunction detection of an integrated drive generator, called an IDG to be monitored, in an aircraft. The electronic circuitry is configured to collect reference data relating to measurements made and recorded during previous flights of aircraft equipped with IDGs, called reference IDGs, the reference data comprising at least measurements corresponding to the following parameters, for each reference IDG: fuel flow rate to supply a propulsion engine with which the reference IDG in question is associated; position of a fuel return valve associated with the reference IDG in question; temperature measured by a sensor at the input port of the reference IDG in question;temperature measured by a sensor at the output port of the reference IDG in question; frequency of the electrical signal produced by the reference IDG in question; temperature of the oil supplying the propulsion engine with which the reference IDG in question is associated. The electronic circuitry is further configured to: carry out training, in supervised mode, of the machine learning type artificial intelligence using the reference data, with a first classification pass to train to detect possible malfunctions suffered by IDGs and a second classification pass to train to determine the causes of the malfunctions if any; put the machine learning type artificial intelligence into production after validation of the training;collecting data to be analyzed, of the same type as the reference data, relating to measurements taken and recorded during flights of the aircraft comprising the IDG to be monitored; using machine learning type artificial intelligence which has been trained, and thus predicting a possible malfunction suffered by the IDG to be monitored and predicting, where appropriate, a cause of the malfunctions, using respectively said first and second passes, from the data to be analyzed.; BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above-mentioned and other features of the invention will become more clearly apparent from the following description of at least one exemplary embodiment, said description being given in relation to the attached drawings, among which: [ Fig. 1 ] schematically illustrates, in top view, an aircraft equipped with propulsion engines respectively provided with integrated drive generators IDG; [ Fig. 2A ] schematically illustrates an algorithm for obtaining data that can be used to train and validate machine learning-type artificial intelligence; [ Fig. 2B ] schematically illustrates a training and validation algorithm for machine learning-type artificial intelligence; [ Fig. 3A ] schematically illustrates an algorithm for obtaining data to be analyzed by machine learning type artificial intelligence, in order to detect possible malfunctions of an integrated drive generator and, in this case, to find the causes; [ Fig. 3B ] schematically illustrates a data analysis algorithm using machine learning-type artificial intelligence, in order to detect possible malfunctions of an integrated drive generator and, in this case, to find the causes; [ Fig. 4 ] schematically illustrates an algorithm for configuring and using machine learning-type artificial intelligence; and [ Fig. 5 ] schematically illustrates an example of a hardware platform configured to perform an implementation of machine learning-type artificial intelligence. DETAILED PRESENTATION OF IMPLEMENTATION METHODS

[0025] There Fig. 1 thus schematically illustrates, in a top view, an aircraft 10. The aircraft 10 comprises at least one propulsion engine 102, preferably of the turbomachine type. Each propulsion engine 102 is equipped with an integrated drive generator IDG 103. Each propulsion engine 102 is also preferably equipped with a generator control unit GCU (Generator Control Unit) whose main functions are to regulate and monitor the frequency and voltage of the alternator of the IDG 103 of the propulsion engine 102 in question, to monitor feeder faults, overloads and phase reversals, and to protect the electrical network of the aircraft 10 by controlling an associated generator line contactor GLC (Generator Line Contactor).Each GCU is connected to the avionics 101 of the aircraft 10, more particularly to a centralized electronic aircraft monitoring system ECAM (“Electronic Centralized Aircraft Monitoring” in English), so as to provide alert signals as needed, in particular in the cockpit of the aircraft 10.

[0026] Each IDG 103 is typically equipped with two oil temperature sensors: one oil temperature sensor at the inlet port and another oil temperature sensor at the outlet port. These sensors are typically thermistors. Using these sensors, the GCU associated with the IDG 103 in question monitors the oil temperature of the IDG 103 and provides an alert signal if a predefined temperature threshold is exceeded on one and / or the other of these two oil temperature sensors.

[0027] Each IDG 103 is preferably associated with a fuel return valve FRV 104 which controls the flow of fuel, used in particular for cooling the oil of the IDG 103, from a fuel distribution system to the propulsion engine 102, with which the IDG 103 is associated, to a recirculation system returning hot fuel to a fuel tank of the aircraft in order to allow cooler fuel to be injected into the fuel distribution system from said aircraft fuel tank. The FRV 104 also controls said flow of fuel for cooling the oil of said propulsion engine 102. Thus, the oil of the IDG 103 in question and the oil of said propulsion engine 102 are better cooled. Note that the propulsion engines 102 are provided with oil temperature sensors supplying them.

[0028] A malfunction of these temperature sensors or of the FRV 104, or of a fuel level sensing control unit FLSCU (Fuel Level Sensing Control Unit), or of any other organ with which the IDG 103 in question interacts, directly or indirectly, may lead to a malfunction, sometimes long-term, of the IDG 103. To detect possible malfunctions of the IDG 103 in question and to find the causes thereof, a data analysis system is used. The data analysis system is based on machine learning type artificial intelligence, such as an artificial neural network or a support vector machine SVM (Support Vector Machine), as described below.

[0029] There Fig. 2A schematically illustrates an algorithm for obtaining data that can be used to train and validate machine learning-type artificial intelligence.

[0030] In a step 201, the data analysis system collects reference data relating to measurements taken and recorded during aircraft flights. The data includes at least measurements relating to the operation and configuration of the IDGs 103 of these aircraft and any other component with which these IDGs 103 interact, directly or indirectly. Preferably, the measurements correspond at least to the following parameters, for each IDG 103 concerned: Fuel flow rate to supply the propulsion engine 102 with which the IDG 103 in question is associated; Position (open / closed) of the FRV 104 associated with the IDG 103 in question; Temperature measured by the sensor at the input port of the IDG 103 in question; Temperature measured by the sensor at the output port of the IDG 103 in question; Frequency of the electrical signal produced by the IDG 103 in question; Temperature of the oil supplying the propulsion engine 102 with which the IDG 103 in question is associated.

[0031] In a particular embodiment, the above list of parameters is supplemented by the following parameters: Time the fuel return valve is open; Time the fuel return valve is closed; Ratio of time the fuel return valve is closed to time the fuel return valve is open; Time taken for the fuel return valve to move from closed to open position; Temperature difference between the IDG inlet port and the IDG outlet port.

[0032] In a step 202, the data analysis system preferably performs processing on the data collected in step 201. The processing comprises, for example, selection, transformation, aggregation and combination operations. Preferably, the processing applied corresponds at least to the calculation of the following information, possibly in combination, for one or more of the parameters listed above: Minimum value; Maximum value; Standard deviation; Mean; Moving average; Slope slope coefficient.

[0033] In a particular embodiment, the calculation of at least a portion of said information is carried out by grouping reference data, and respectively data to be analyzed, over several consecutive aircraft flights. For example, the data processing produces a sliding average, over a predefined quantity of consecutive aircraft flights, of the temperature difference between the IDG inlet port 103 and the IDG outlet port 103 (eg, temperature at the inlet port minus temperature at the outlet port). According to another example, the data processing produces a slope direction coefficient of the sliding average, over a predefined window of consecutive aircraft flights, of the average (per flight or per cruise flight segment) of the temperature difference between the IDG inlet port 103 and the IDG outlet port 103 (eg, temperature at the inlet port minus temperature at the outlet port).In another example, the data processing outputs a slope direction coefficient, over a predefined quantity of consecutive aircraft flights, of the maximum (or minimum, or average, or standard deviation) value of the temperature difference between the inlet port of IDG 103 and the outlet port of IDG 103 (eg, temperature at the inlet port minus temperature at the outlet port).

[0034] In a step 203, the data analysis system provides a data set, derived from the collected data and preferably derived from the processing of step 202, usable for training and validating machine learning type artificial intelligence.

[0035] There Fig. 2B schematically illustrates a machine learning type artificial intelligence training and validation algorithm.

[0036] In a step 211, the data analysis system injects a training data set as input to the machine learning type artificial intelligence. The data set provided in step 203 is separated into two data sets: the training data set and a validation data set.

[0037] Machine learning artificial intelligence training is performed in a supervised manner. In other words, a human operator annotates the data in the dataset in question, in order to indicate to the machine learning artificial intelligence the expected classification for the injected input data. The machine learning artificial intelligence configures itself taking into account the injected input data and the provided annotation. The data is classified with regard to malfunctions that may subsequently be observed on the IDG 103 concerned by the collected data.

[0038] Preferably, the data is classified with an indication of the time elapsed or number of flights made between the measurements in question and the detection of the malfunction. This allows artificial intelligence to learn to predict the occurrence of malfunctions impacting the IDG 103.

[0039] The training is carried out according to two classification axes. A first classification axis concerns the detection or prediction of IDG 103 malfunction. A second classification axis concerns, in the event of detection or prediction of IDG 103 malfunction, the prediction of the cause of the IDG 103 malfunction.

[0040] To do this, in a step 212, the data analysis system performs a first classification pass on the injected data set. The classification trained for this first pass concerns the detection or prediction of malfunction of IDG 103. For example, the first pass is performed using a first decision-making algorithm. In a particular embodiment, the first pass is performed with a LightGBM (“Light Gradient Boosting Machine” in English) type algorithm. LightGBM is well suited to the first pass, due to its speed of execution and its reliability with a substantial input data set (availability of failure reports from previous flights).

[0041] Thus, in a step 213, depending on the first pass, the data analysis system obtains a prediction as to the malfunction of the IDG 103 concerned, for the injected data, in view of the annotation provided for training.

[0042] To be able to learn the cause of the malfunctions, if any, the data analysis system performs a second classification pass on a selection (step 214) from the data set, the selection corresponding to the injected data which have been annotated as representative of a malfunction of the IDG 103. For example, the second pass is performed using a second decision-making algorithm different from the first decision-making algorithm. In a particular embodiment, the second pass is performed with an XGBoost type algorithm. XGBoost is well suited to the second pass due to its greater reliability with a more restricted input data set (due to the selection following the first pass).

[0043] Thus, in a step 216, depending on the second pass, the data analysis system obtains a prediction as to the cause of the malfunction of the IDG 103 concerned, for the injected data, in view of the annotation provided for training.

[0044] In order to verify that the machine learning type artificial intelligence has been adequately configured by the training performed, a new cycle is performed, in which the validation data set is injected in step 211. As part of the validation, the injected data do not undergo annotation, and the machine learning type artificial intelligence performs the classification expected by the first and second passes by itself. If the machine learning type artificial intelligence performs the classification as expected, then the data analysis system can be put into production, i.e. used to detect possible malfunctions of an integrated training generator and, in this case, to find the causes thereof. ; otherwise, new training must be performed.

[0045] Thus, thanks to suitable training, machine learning-type artificial intelligence can be easily configured, based on failure reports from previous flights, to detect IDG 103 malfunctions linked to: An FRV 104 open / close control fault; An IDG 103 oil clogging; An IDG 103 outlet port temperature sensor fault; An IDG 103 inlet port temperature sensor fault; An IDG 103 oil leak; An IDG 103 maintenance fault.

[0046] Machine learning-type artificial intelligence can be trained to detect other types of causes of IDG 103 malfunctions.

[0047] There Fig. 3A schematically illustrates an algorithm for obtaining data to be analyzed by machine learning type artificial intelligence.

[0048] Once the data analysis system is in production, the purpose of its use is to detect or predict the occurrence of an IDG 103 malfunction, and if so, to establish the cause.

[0049] Measurements are carried out during aircraft flights in order to collect, in a step 301, data making it possible to monitor each IDG 103 of said aircraft. The measurements in question provide data of the same type as those collected in step 201 to initiate the training of the machine learning type artificial intelligence.

[0050] In a step 302, the data analysis system performs processing on the data collected in step 301. The processing applied in step 302 is identical to that applied in step 202, in order to obtain in a step 303 data to be analyzed of the same type and nature as those obtained in step 203.

[0051] There Fig. 3B schematically illustrates a data analysis algorithm using machine learning-type artificial intelligence.

[0052] In a step 311, the data analysis system injects data to be analyzed, as obtained in step 303, as input to the machine learning type artificial intelligence. Unlike the training phase, it is not known here whether the data are indicative of a malfunction suffered by the IDG 103 concerned or not.

[0053] In a step 312, the data analysis system performs a first classification pass on the injected data. As for training, this first pass concerns the detection or prediction of malfunction of the IDG 103 concerned by the injected data. The first pass is carried out using the same algorithm as for training (first algorithm), with the difference that in production, the machine learning type artificial intelligence has been trained to detect or predict by itself a malfunction of IDG 103 from the injected data (data to be analyzed).

[0054] Thus, in a step 313, depending on the first pass, the data analysis system obtains a prediction as to the malfunction of the IDG 103 concerned, for the injected data, in view of the training carried out.

[0055] When the first pass reveals that the injected data are indicative of a malfunction suffered by the IDG 103, the corresponding injected data are selected in a step 314 in order to undergo a second pass in a step 315. The second pass is carried out using the same algorithm as for training (second algorithm), with the difference that in production, the machine learning type artificial intelligence has been trained to find the cause of the malfunction itself from the injected data (data to be analyzed).

[0056] Thus, in a step 316, depending on the second pass, the data analysis system obtains a prediction as to the cause of the malfunction of the IDG 103 concerned, for the injected data, in view of the training carried out.

[0057] Thus, the data analysis system is able to determine, from data collected in flight and from machine learning type artificial intelligence which has been trained as previously described, whether a malfunction of the IDG 103 has occurred or is about to occur, and in this case, what the cause is. Maintenance around the IDG 103 is then greatly facilitated.

[0058] There Fig. 4 schematically illustrates an algorithm for configuring and using machine learning-type artificial intelligence within the data analysis system.

[0059] In a step 401, the machine learning type artificial intelligence is trained and validated as previously described in relation to the Figs. 2A et 2B .

[0060] In a step 402, following training and validation, a trained and validated model is obtained, allowing production of the machine learning type artificial intelligence in a step 403.

[0061] In a step 404, a monitoring phase is performed. The monitoring phase consists of collecting data during aircraft flights in order to monitor the proper functioning of each IDG 103 of said aircraft, as well as analyzing the collected data. The data collection and the analysis of the collected data are performed as previously described in relation to the Figs. 3A et 3B .

[0062] Thus, the data analysis system can equip each aircraft concerned, and preferably be integrated into the avionics 101 of said aircraft. Alternatively, the data analysis system is computer equipment on the ground in a control center, for example in the premises of an airline. Then, the data to be collected are transmitted by the avionics 101 to the computer equipment of the control center, for example using ground-to-air communications, or by using an electronic flight bag (EFB) as an intermediary between the avionics 101 and the computer equipment of the control center. The electronic flight bag (EFB) then uses wired (for example, USB type) or wireless (for example, Bluetooth or Wi-Fi type) communications to exchange with the computer equipment of the control center and the avionics 101.

[0063] In a step 405, the data analysis system determines whether the analysis of the collected data shows a malfunction of IDG 103. If so, a step 406 is performed; otherwise, the monitoring is continued in step 404.

[0064] In step 406, the data analysis system generates an alarm representative of the detected malfunction of IDG 103 (detection using the first pass). The alarm is accompanied by information indicating the supposed cause of the detected malfunction (identification using the second pass). Maintenance can then be scheduled, and monitoring is continued in step 404.

[0065] There Fig. 5schematically illustrates an example of a hardware platform 500 of the data analysis system, in the form of electronic circuitry, which is adapted and configured to implement the algorithms described above. The hardware platform 500 is thus adapted and configured to implement the machine learning type artificial intelligence disclosed herein.

[0066] The hardware platform 500 then comprises, connected by a communication bus 510: a processor or CPU (Central Processing Unit) 501; a RAM (Read-Only Memory) 502; a read-only memory 503, for example of the ROM (Read Only Memory) or EEPROM (Electrically-Erasable Programmable ROM) type; a storage unit 504, such as a hard disk drive (HDD) or a storage media reader, such as an SD (Secure Digital) card reader; and an I / f interface manager 505.

[0067] The I / f interface manager 505 makes it possible to interact with one or more equipment of the aircraft 10, such as the aforementioned temperature sensors, and / or possibly with a communication network. For example, the I / f interface manager 505 is adapted and configured to make it possible to interact with the avionics 101.

[0068] The processor 501 is capable of executing instructions loaded into the RAM 502 from the ROM 503, an external memory, a storage medium (such as an SD card), or a communications network. When the hardware platform 500 is powered on, the processor 501 is capable of reading instructions from the RAM 502 and executing them. These instructions form a computer program causing the processor 501 to implement some or all of the steps and operations described herein.

[0069] All or part of the steps and operations described herein may thus be implemented in software form by executing a set of instructions by a programmable machine, for example a DSP (Digital Signal Processor) type processor or a microcontroller, or be implemented in hardware form by a machine or a dedicated electronic component (chip) or a dedicated set of electronic components (chipset), for example an FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit) component. Generally speaking, the hardware platform 500 comprises electronic circuitry adapted and configured to implement the operations and steps described herein.

Claims

1. Method for detecting a malfunction suffered by an integrated drive generator, referred to as IDG (103) to be monitored, in an aircraft (10), the method being implemented by a data analysis system comprising an artificial intelligence of the machine learning type, characterized in that the method comprises the following steps: - collecting (201) reference data relating to measurements made and recorded during previous flights of aircraft equipped with IDGs (103), called reference IDGs, the reference data comprising at least measurements corresponding to the following parameters, for each reference IDG (103): - Flow of fuel supplying a jet engine (102) with which the reference IDG (103) in question is associated; - Position of a fuel return valve (104) associated with the reference IDG (103) in question; - Temperature measured by a sensor on the inlet port of the reference IDG (103) in question; - Temperature measured by a sensor on the outlet port of the reference IDG (103) in question; - Frequency of electrical signal produced by the reference IDG (103) in question; - Temperature of oil supplying the jet engine (102) with which the reference IDG (103) in question is associated; - carrying out a training (401), in supervisor mode, of the artificial intelligence of the machine learning type by virtue of the reference data, with a first classification run (212) for training to detect potential malfunctions suffered by IDGs (103) and a second classification run (215) for training to determine causes of the malfunctions where relevant; - using in production (403) the artificial intelligence of the machine learning type after validation of the training; - collecting (301) data to be analysed, of the same type as the reference data, relating to measurements made and recorded during flights of the aircraft comprising the IDG (103) to be monitored; - using the artificial intelligence of the machine learning type that has been trained, and thus predicting (313) a potential malfunction suffered by the IDG (103) to be monitored and predicting (316), where relevant, a cause of the malfunctions by virtue, respectively, of the said first and second runs, using the data to be analysed.

2. Method according to Claim 1, in which the reference data furthermore comprise measurements corresponding to at least one of the following parameters, for each reference IDG (103): - Time during which the fuel return valve (104) is open; - Time during which the fuel return valve (104) is closed; - Ratio between the time during which the fuel return valve (104) is closed and the time during which the fuel return valve is open; - Time taken by the fuel return valve (104) to go from the closed position to the open position; - Difference in temperature between the inlet port of the reference IDG (103) in question and outlet port of the reference IDG (103) in question.

3. Method according to Claim 1 or 2, in which the data analysis system applies a processing to the reference data and to the data to be analysed, the processing applied corresponding at least to the calculation of the following information, potentially in combination, for one or more of the said parameters: - Minimum value; - Maximum value; - Standard deviation; - Average; - Moving average; - Gradient coefficient.

4. Method according to Claim 3, in which the calculation of at least a part of the said information is carried out by grouping reference data and, respectively, data to be analysed, over several consecutive aircraft flights.

5. Method according to any one of Claims 1 to 4, in which the artificial intelligence of the machine learning type is trained to detect an IDG (103) malfunction linked to: - An anomaly in the opening / closing controls of a fuel return valve (104); - A clogging of IDG (103) oil; - A failure of a temperature sensor on the IDG (103) outlet port; - A failure of a temperature sensor on the IDG (103) inlet port; - An oil leak in an IDG (103); - An anomaly of IDG (103) maintenance.

6. Method according to any one of Claims 1 to 5, in which the first classification run is carried out by applying a first algorithm and the second classification run is carried out by applying a second algorithm distinct from the first algorithm.

7. Method according to Claim 6, in which the first algorithm is of the LightGBM type and the second algorithm is of the XGBooost type.

8. Method according to any one of Claims 1 to 7, in which the artificial intelligence of the machine learning type is used in production in an avionics system (101) of the aircraft (10) equipped with the IDG (103) to be monitored.

9. Method according to any one of Claims 1 to 7, in which the artificial intelligence of the machine learning type is used in production in a computer system on the ground in a control centre, and in which the data to be analysed are transmitted by an avionics system (101) of the aircraft (10), equipped with the IDG (103) to be monitored, to the computer system in the control centre.

10. Computer program product, comprising instructions leading to the execution, by a processor (501), of the method according to any one of Claims 1 to 9, when the said instructions are executed by the processor (501).

11. Storage medium, storing a computer program comprising instructions leading to the execution, by a processor (501), of the method according to any one of Claims 1 to 9, when the said instructions are read and executed by the processor (501).

12. Data analysis system comprising an artificial intelligence of the machine learning type, the system comprising an electronic circuitry configured to carry out a detection of a malfunction suffered by an integrated drive generator, referred to as IDG (103) to be monitored, in an aircraft (10), characterized in that the electronic circuitry is configured for: - collecting (201) reference data relating to measurements made and recorded during previous flights of aircraft equipped with IDGs (103), called reference IDGs, the reference data comprising at least measurements corresponding to the following parameters, for each reference IDG (103): - Flow of fuel supplying a jet engine (102) with which the reference IDG (103) in question is associated; - Position of a fuel return valve (104) associated with the reference IDG (103) in question; - Temperature measured by a sensor on the inlet port of the reference IDG (103) in question; - Temperature measured by a sensor on the outlet port of the reference IDG (103) in question; - Frequency of electrical signal produced by the reference IDG (103) in question; - Temperature of oil supplying the jet engine (102) with which the reference IDG (103) in question is associated; - carrying out a training (401), in supervisor mode, of the artificial intelligence of the machine learning type by virtue of the reference data, with a first classification run (212) for training to detect potential malfunctions suffered by IDGs (103) and a second classification run (215) for training to determine causes of the malfunctions, where relevant; - using in production (403) the artificial intelligence of the machine learning type after validation of the training; - collecting (301) data to be analysed, of the same type as the reference data, relating to measurements made and recorded during flights of the aircraft comprising the IDG (103) to be monitored; - using the artificial intelligence of the machine learning type that has been trained, and thus predicting (313) a potential malfunction suffered by the IDG (103) to be monitored and predicting (316), where relevant, a cause of the malfunctions by virtue, respectively, of the said first and second runs, using the data to be analysed.