Method and device for predicting the potential service life of a heat sink of an aircraft

Abstract

The present invention relates to a method and to a device for predicting the potential service life of a heat sink of an aircraft, which method comprises: - obtaining characteristics of each airport; - obtaining environmental data for each airport; - forming (E500), for each airport, a descriptive overview of the airport from the characteristics of the airport and the environmental data; - determining (E510) for at least one known aircraft, a list of airports visited by the aircraft since the last time heat sinks were fitted on the aircraft; - forming (E520, E530, E540) a training dataset; - training and evaluating a model based on the training dataset; - predicting (E550) a potential service life of the heat sink of the aircraft from the descriptive overview and a list of airports visited by the aircraft.

Description

[0001] METHOD AND DEVICE FOR PREDICTING THE LIFE POTENTIAL OF A WELL

[0002] HEAT FROM AN AIRCRAFT

[0003] TECHNICAL FIELD

[0004] The present invention relates to a method and device for predicting the life potential of heat sinks in an aircraft.

[0005] STATE OF PRIOR ART

[0006] In the aeronautics industry, predicting equipment lifespan is a major challenge, and heat sinks are among the components for which this is particularly true. Indeed, there is a significant disparity in the lifespan of heat sinks from one aircraft to another. A heat sink is a series of discs located on the landing gear wheels; when these discs come into contact with each other, they allow the aircraft to brake. Predicting the potential lifespan of a heat sink is a complex task, and currently, no solution exists that can accurately make such predictions.

[0007] Indeed, while the use of data measured and / or stored by equipment within an aircraft often allows for the calculation of remaining service life, access to this type of information is not guaranteed for all aircraft. The lack of suitable sensors or the difficulty of accessing the data are, in fact, major obstacles.

[0008] Currently, most aircraft are not equipped with sensors that measure the wear and tear of a heat sink. Even if such sensors existed, accessing this data for predictive purposes for other aircraft is difficult due to the sheer volume of data to be processed and the need for authorization from the aircraft owner to process this data.

[0009] The present invention overcomes the problem of data access, enables personalized predictions of the number of cycles a heat sink can perform, and provides recommendations to airlines that will allow them to optimize the lifespan of their heat sinks and better manage their fleets. DESCRIPTION OF THE INVENTION

[0010] To this end, according to a first aspect, an embodiment proposes a method for predicting the heat sink lifespan potential of an aircraft, characterized in that the method comprises the following steps:

[0011] - obtaining characteristics of each airport, the characteristics of an airport being at least representative of the airport's runways and their characteristics,

[0012] - obtaining environmental data for each airport at different time periods within a year,

[0013] - creation, for each airport, of a descriptive set of the airport based on the airport's characteristics and environmental data for different time periods,

[0014] - determination, for at least one known aircraft, of a list of airports visited by the aircraft since the last heat sink installation on the aircraft, the determination of the airports being carried out from flight data of the aircraft since the last heat sink installation and the list of airports visited being obtained from flight tracking software,

[0015] - creation of a training game based on the descriptive set and the list of airports visited by the aircraft,

[0016] - training and evaluation of a model from the training set, the model having a property of invariance to the order of airports visited,

[0017] - Prediction of the aircraft's heat sink lifespan potential based on the descriptive dataset and a list of airports visited by the aircraft

[0018] One embodiment also relates to a device for predicting the heat sink lifespan potential of an aircraft, characterized in that the device comprises:

[0019] - means of obtaining characteristics of each airport, the characteristics of an airport being at least representative of the airport's runways and their characteristics,

[0020] - means of obtaining environmental data for each airport at different times within a year,

[0021] - means of creating, for each airport, a descriptive set of the airport based on the airport's characteristics and environmental data for different time periods; - means of determining, for at least one known aircraft, a list of airports visited by the aircraft since the last heat sink installation on the aircraft, the determination of the airports being carried out from flight data of the aircraft since the last heat sink installation and the list of airports visited being obtained from flight tracking software.

[0022] - training resources, for each airport in a training game based on the descriptive set and the list of airports visited by the aircraft,

[0023] - means of training and evaluating a model from the training set, the model having a property of invariance to the order of airports visited,

[0024] - methods for predicting the potential lifespan of an aircraft's heat sink based on a descriptive dataset and a list of airports visited by the aircraft

[0025] Thus, the present invention makes it possible to predict the life potential of heat wells without requiring the use of sensors.

[0026] The present invention makes it possible to overcome the problem of accessing the intrinsic data of an aircraft.

[0027] The present invention also makes it possible to make a precise and personalized prediction of the number of cycles that a heat well will be able to perform depending on the airports on which it will operate.

[0028] The present invention also makes it possible to take into account the different factors influencing the wear of heat sinks and to provide recommendations to airlines to optimize the lifespan of their heat sinks and to better manage their fleet.

[0029] The invariance property of the model makes it possible to evaluate a heat sink life potential and to train the model only once on the training set in order to then predict the number of cycles that a heat sink will be able to perform.

[0030] According to a particular method, the process also includes a step of determining a severity score for each airport from the trained model, the descriptive set of the airport and a list containing only the airport.

[0031] The invariance property of the model allows us to extrapolate a severity score for individual airports and to train the model only once on the training set.

[0032] Thus, the model has two capabilities: predicting the number of cycles a heat sink can complete given its frequency of visits to different airports. Therefore, the present invention makes it possible to identify the airports responsible for significant wear on heat sinks.

[0033] In a particular way, the characteristics of an airport are the number of runways at each airport, the number of runways with a surface, an indicator of the condition of the surface of each runway, the length of the runways at each airport, the mechanical characteristics of the runway surface, the distances of the airport taxiways laid out for the movement of aircraft between parking points and runways, the cleanliness of the runways, the altitude of the airport.

[0034] Thus, the present invention makes it possible to identify the airports that cause significant wear and tear on heat sinks.

[0035] In a specific method, environmental data such as temperature, humidity, wind, visibility, and pollution levels are considered during a particular time of year. Thus, the present invention enables a precise and personalized prediction of the number of cycles a heat well can perform, based on the airports where it will operate and the seasons.

[0036] In a particular way, the process also includes a step of determining the frequency of visits to each airport.

[0037] The model takes as input a list of airports and an associated visit frequency list, and by training the model only once, it can obtain both a prediction of the number of cycles and a prediction of the airport's severity. In a specific method, the determination of a severity score and the prediction are performed using machine learning software.

[0038] In a particular way, the machine learning software determines the severity score and prediction from two levels of dense neural networks and between the two levels of dense neural networks the process includes a weighting step by frequency of visits to each airport.

[0039] A particular mode also relates to a computer program product. It includes instructions for implementing, by a piece of equipment, the process according to one of the preceding embodiments, when said program is executed by a processor of the device. A particular mode also relates to a storage medium. It stores a computer program comprising instructions for implementing, by a node device, the process according to one of the preceding embodiments, when said program is executed by a processor of the device. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The features of the invention mentioned above, as well as others, will become clearer upon reading the following description of an exemplary embodiment, said description being made in relation to the accompanying drawings, among which:

[0041] [Fig. 1] schematically illustrates an example of an algorithm for determining descriptive data of an airport for different periods of a year;

[0042] [Fig. 2] schematically illustrates an example of an algorithm for determining, for each aircraft, the airports visited by the aircraft since the last heat sink installation on the aircraft;

[0043] [Fig. 3] schematically illustrates an example of a model training and evaluation algorithm;

[0044] [Fig. 4] schematically illustrates an example of an algorithm for predicting the life potential of a heat sink;

[0045] [Fig. 5] schematically illustrates an example of an algorithm for predicting the severity score of an airport;

[0046] [Fig. 6] schematically illustrates an example of the architecture of a device for predicting the life potential of heat sinks of an aircraft.

[0047] DETAILED DESCRIPTION OF IMPLEMENTATION METHODS

[0048] Fig. 1 schematically illustrates an example of an algorithm for determining descriptive data of an airport for different periods of a year.

[0049] The device 60 for predicting the life potential of an aircraft heat sink obtains the characteristics of each airport at stage E100.

[0050] The characteristics of an airport include, for example, the number of runways at each airport, the number of runways with a surface, an indicator of the condition of the surface of each runway, the length of the runways at each airport, the mechanical characteristics of the runway surface, the distances of the airport taxiways designed for the movement of aircraft between parking points and runways, the cleanliness of the runways and the altitude of the airport.

[0051] By design, any characteristic that can be measured across all airports worldwide can be integrated. For example, satellite data can be used to associate each runway with taxiway length or proximity to residential areas. Indeed, runway length affects whether or not an aircraft pilot needs to reverse thrust during a short landing. Similarly, the proximity of a runway to residential areas can influence its use or help limit noise pollution.

[0052] The aircraft heat sink life potential prediction device 60 obtains environmental data from each airport at stage El 10.

[0053] Environmental data includes, for example, temperature, humidity, wind, visibility, pollution, etc.

[0054] At step E120, the device 60 for predicting the potential life of an aircraft heat sink forms predetermined airport / time period pairs.

[0055] Because environmental conditions can vary significantly throughout the year, environmental data is grouped into several predetermined periods, such as seasons. This seasonal representation provides a more accurate picture of airport conditions at each time of year.

[0056] In step E130, the device 60 for predicting the lifetime potential of an aircraft heat sink aggregates all the information obtained in step E120 to produce a statistical representation of the airport's environmental characteristics. Data from the same airport are aggregated using the following statistical tools: median, quantiles (Q10 and Q90), and mean deviation from the median.

[0057] The value below which 10% of the lowest values ​​are found is called the 10th quantile (Q10), and the value above which 10% of the highest values ​​are found is called the 90th quantile (Q90).

[0058] At step E140, the device 60 for predicting the life potential of an aircraft heat sink performs a preprocessing of the data obtained at steps El 00 and El 30.

[0059] Each pair (season, airport) is represented by a vector whose values ​​represent the airport's characteristics. To make these vectors usable by a neural network, we perform a 1-out-of-n encoding, which involves encoding an n-state variable using n bits for categorical variables and normalizing the data for numerical variables. Outliers are identified and removed. Missing values ​​are replaced by the mean in the case of numerical data or by the most frequent value in the case of categorical data.

[0060] The pre-processed data is thus available at step El 50 and forms a descriptive set of airport-season pairs. Fig. 2 schematically illustrates an example of an algorithm for determining, for each aircraft, the airports visited by the aircraft since the last heat sink installation on the aircraft.

[0061] At step E200, the device 60 for predicting the life potential of an aircraft heat sink obtains an identifier (tail number) for each aircraft

[0062] At step E210, the device 60 for predicting the life potential of an aircraft heat sink obtains, for each identified aircraft, the last date of shipment and receipt of the heat sink.

[0063] At step E220, the device 60 for predicting the life potential of an aircraft heat sink obtains, for each identified aircraft, the list of all airports visited by the aircraft since it was equipped with the received heat sinks.

[0064] The list of airports visited, according to the invention, is obtained from flight tracking software such as "FlightRadar24" and "FlightAware," available on the internet. This software indicates the position of aircraft in flight. From the positions of each identified aircraft, it is possible to deduce the list of airports visited by each identified aircraft, starting from the date it is equipped with the heat sinks. This solution avoids having to contact the operator of the identified aircraft.

[0065] At step E230, device 60 for predicting the life potential of an aircraft heat sink determines for each identified aircraft the frequency of visits, per season, of the identified aircraft to each airport.

[0066] At step E240, device 60 for predicting the life potential of an aircraft heat sink determines a training dataset at least from the frequency of visits and the descriptive set of airport season pairs.

[0067] The data obtained at steps E230 and E240 form a training dataset in the form of vectors of the representations of the airports visited and the frequencies of visits at step E250.

[0068] Fig. 3 schematically illustrates an example of a model training and evaluation algorithm.

[0069] According to the invention, a learning model as described in the publication entitled "Deep Sets" whose authors are Manzil Zaheer, Satwik Kottur, and Siamak Ravanbhakhsh is used in step E320.

[0070] The model adheres to the conditions outlined in the DeepSet paper to ensure invariance to the order of visited airports. The model takes as input (E300) the vectors representing the visited airports and the visit frequencies obtained in step (E250). A first sequence of dense neural network layers represents these vectors in a latent space to perform a prediction (E330) of the number of cycles completed and subsequently penalize the error (E340). Training is performed in a supervised manner using gradient descent (E350) and weight adjustment (E360).

[0071] Several combinations of E370 hyperparameters are tested using a technique such as GridSearch to parameterize the model in the best possible way (number of neurons, layers, etc.) in order to obtain a trained E310 model which after several iterations is evaluated as E380.

[0072] Fig. 4 schematically illustrates an example of an algorithm for predicting the severity score of an airport.

[0073] This algorithm is performed for each airport.

[0074] To generate an airport's severity score, simply input E410 to the trained model a list containing, for each season, only the airport whose severity you wish to predict, and assign it a visit frequency of 1. The resulting prediction corresponds to the total number of cycles theoretically expected if only that airport were visited. This value can therefore be interpreted as the airport's severity for the model.

[0075] Since the model is used to predict the severity of each airport visited individually, it is preferable for the model to be independent of the order in which the airports are visited. Indeed, the present invention aims to obtain absolute severity scores, independent of the conditions encountered at other airports.

[0076] Within the scope of the present invention, the order of airport visits has no impact on wear. This property is expressed as follows:

[0077] With 7i being any permutation, x^i) the representation of airport i(i) and f our model. Such a function is said to be invariant. According to the theorem stated in the Deepset paper, it can also be decomposed as follows: fi (X ex M®)) In our case (p is the function that associates to an airport its representation by the model, weighted by its visit frequency, and p is the function that, from its weighted representations, predicts wear and tear.

[0078] The prediction is made from the descriptive vectors of the airport season pairs E400 obtained in step El 50 and the data from step E410.

[0079] A first sequence of dense neuronal layers (E420) represents these vectors in a latent space to obtain a model of the airport (E430). A weighted sum of these representations is performed using the airport visit frequencies (E410) as the weighting. The result of this weighted sum is fed into a second sequence of neuronal layers (E440), which predicts, at output (E450), the total number of cycles the heat sink can perform if only the airport is visited. This number corresponds to a severity score for the visited airport.

[0080] Fig. 5 schematically illustrates an example of an algorithm for predicting the life potential of a heat sink.

[0081] The user provides a list of visited airports and their frequency. Our solution retrieves the corresponding vector representations for each of these airports and provides them as input to the model along with the associated visit frequencies. The model then predicts the total number of cycles expected for the heat sink.

[0082] The prediction of the life potential of a heat well is carried out from the descriptive vectors of the airport season pairs E500 obtained in step El 50 and the anticipated visit frequencies for each airport / season pair.

[0083] A first sequence of dense neuronal layers (E520) creates a representation of these vectors in a latent space to obtain a model representing the airport models (E530). A weighted sum of these representations is performed using the frequencies (E510) of anticipated visits for each airport / season pair as weighting. The result of this weighted sum is given as input to a second sequence of neuronal layers (E540), which predicts the total number of cycles that the heat sink (E550) can perform.

[0084] Fig. 6 schematically illustrates an example of the architecture of a device for predicting the life potential of heat sinks of an aircraft.

[0085] According to the hardware architecture example shown in Fig. 6, the aircraft heat sink life potential prediction device 60 comprises, connected by a communication bus 600: a processor or CPU (Central Processing Unit) 601; a RAM (Random Access Memory) 602; a ROM (Read Only Memory) 603; a storage unit such as a hard disk drive (or a storage media reader, such as an SD card reader) 604; at least one communication interface 605 enabling the aircraft heat sink life potential prediction device 60 to access different databases or different flight tracking software.

[0086] The 601 processor is capable of executing instructions loaded into RAM 602 from ROM 603, external memory (not shown), storage media (such as an SD card), or a communication network. When the Cont controller device is powered on, the 601 processor can read instructions from RAM 602 and execute them. These instructions form a computer program that causes the 601 processor to implement all or part of the process described in relation to Figs. 1 to 5.

[0087] The method described below in relation to Figs. 1 to 5 can be implemented in software form by executing a set of instructions by a programmable machine, for example a DSP (Digital Signal Processor) or a microcontroller, or in hardware form by a dedicated machine or component, for example an FPGA (Field-Programmable Gate Array) or an ASIC (Application-Specific Integrated Circuit). In general, the device 60 for predicting the heat sink lifetime potential of an aircraft includes electronic circuitry configured to implement the methods described in relation to Figs. 1 to 5.

Claims

DEMANDS 1. A method for predicting the heat sink lifetime potential of an aircraft, the method being implemented by electronic circuitry of a prediction device, characterized in that the method comprises the steps of: - obtaining characteristics of each airport, the characteristics of an airport being at least representative of the airport's runways and their characteristics, - obtaining environmental data for each airport at different time periods within a year, - training (E500), for each airport, of a descriptive set of the airport based on the airport's characteristics and environmental data for different time periods, - determination (E510) for at least one known aircraft, of a list of airports visited by the aircraft since the last heat sink installation on the aircraft, the determination of the airports being carried out from flight data of the aircraft since the last heat sink installation and the list of airports visited being obtained from flight tracking software, - training (E520, E530, E540) of a training game based on the descriptive set and the list of airports visited by the aircraft, - training and evaluation of a model from the training set, the model having a property of invariance to the order of airports visited, - prediction (E550) of the aircraft heat sink life potential from the descriptive set and a list of airports visited by the aircraft, the prediction being made to optimize the heat sink life, the process further comprising a step of determining a severity score of each airport from the trained model, the airport descriptive set and a list containing only the airport.

2. A method according to claim 1, characterized in that the characteristics of an airport are the number of runways at each airport, the number of runways with a surface, an indicator of the condition of the surface of each runway, the length of the runways at each airport, the mechanical characteristics of the runway surface, and the distances airport taxiways designed for the movement of aircraft between parking points and runways, runway cleanliness, airport altitude.

3. A method according to any one of claims 1 or 2, characterized in that the environmental data comprise one or more quantities among temperature, humidity, wind, visibility, pollution.

4. A method according to any one of claims 1 to 3, characterized in that the method further comprises a step of determining the frequency of visits to each airport.

5. Method according to claim 1, characterized in that the determination of a severity score and the prediction are carried out using machine learning software.

6. Method according to claim 5, characterized in that the machine learning software determines the severity score and prediction from two levels of dense neural networks and in that, between the two levels of dense neural networks, the method includes a weighting step by frequency of visits to each airport.

7. A device for predicting the lifetime potential of an aircraft heat sink, characterized in that the device comprises: - means of obtaining characteristics of each airport, the characteristics of an airport being at least representative of the airport's runways and their characteristics, - means of obtaining environmental data for each airport at different times within a year, - means of creating, for each airport, a descriptive set of the airport based on the airport's characteristics and environmental data for different time periods, - means of determining, for at least one known aircraft, a list of airports visited by the aircraft since the last heat sink installation on the aircraft, the determination of the airports being carried out from flight data of the aircraft since The latest heat well installation and the list of airports visited were obtained from flight tracking software. - means of creating a training set from the descriptive set and the list of airports visited by the aircraft, - means of training and evaluating a model from the training set, the model having a property of invariance to the order of the airports visited, - means of predicting the potential lifespan of the aircraft heat sink from the descriptive set and a list of airports visited by the aircraft, the prediction being made to optimize the lifespan of the heat sink, - means of determining a severity score for each airport from the trained model, the descriptive set of the airport and a list containing only the airport.

Images