Method for predicting runway friction information, and corresponding system

Abstract

Method for predicting friction information for a runway, and corresponding system. Method for predicting friction information for an aircraft runway comprising, for a given instant: obtaining (S100) runway context data, obtaining (S101) historical friction information data for the runway including data all older than the given instant, a prediction (S102) by a prediction model configured to receive as input at least the context data at the given instant and the historical friction information data, and to deliver predicted friction information (μ) for the runway at the given instant. Figure for the abstract: Fig. 1

Description

Title of the invention: Method for predicting friction information on a runway, and corresponding system. Technical field

[0001] The invention relates to the general field of aircraft runway management, and in particular to runway friction prediction.

[0002] Prior art Knowledge of friction, or more precisely of a friction coefficient, has become increasingly critical for traffic management within airports or more precisely for the management of airports and their runways.

[0003] Furthermore, knowledge of runway condition information, such as friction, makes it possible to optimize landings and thus reduce their carbon footprint. Characterizing runway conditions therefore improves the environmental performance of aircraft.

[0004] This friction information is also useful in aircraft for managing anti-skid braking. It therefore appears that knowledge of a friction coefficient or information relating to friction is also particularly critical for safety reasons.

[0005] As can be understood, runway weather conditions have an impact on friction. The presence of contaminants can also have an impact on friction. It should be noted that these contaminants can be either related to weather conditions (water, ice, snow, etc.) or independent of these conditions (pollution, particles, rubber, etc.).

[0006] While it is conceivable to calculate a friction coefficient within an aircraft using various measurements, this information is generally not immediately usable by air traffic / airport / runway managers. Indeed, this data is usually only used after landing, after the taxiing phase, and once data communication systems allow the transmission of this data to a remote server.

[0007] Real-time determination of friction for a track is therefore not feasible with prior art solutions.

[0008] The invention aims in particular to overcome these drawbacks. Description of the invention

[0009] To this end, the invention proposes a method for predicting friction information for an aircraft runway, the method being implemented by a computer system and comprising, for a given moment: obtaining context data from the track (for example, data prior to the given moment), obtaining historical friction information data for the runway including data all older than the given time, a prediction by a prediction model configured to receive as input at least the context data at the given time and the historical friction information data, and to deliver a predicted friction information of the runway at the given time.

[0010] It can be noted that the prediction can be considered as a temporal prediction.

[0011] Thus, the invention proposes using a predictive model that provides friction information for a specific instant for which historical friction data is not yet necessarily available. This makes it possible to obtain friction information before, for example, downloading data acquired during an aircraft's passage over the runway, and to implement real-time friction monitoring within the runway. Finally, since the friction information is accessible in real time, it can be used to implement runway management (i.e., management of said runway and possibly other runways at the same airport).

[0012] Historical friction information data and context data may be older than the given time, for example, older according to a chosen time step. For example, for data known at T, the friction information at t + dt is predicted if dt is the time step.

[0013] It can be noted that one can use friction information history data and context data which are included in a time window extending between f and ? - dt', dt' being either a time step equal to dt, or a different time step.

[0014] The method can be implemented by a computer system (one or more computers), for example a remote computer system. This is due in particular to the fact that the historical friction information data is older than the time for which the prediction is implemented: it is not necessary to have real-time information relating to the landing / taxiing of an aircraft on the runway, which is time-consuming to transfer.

[0015] The friction information can be a friction coefficient, for example a time average of a friction coefficient, and / or for example a friction coefficient representative of the track. Alternatively, the friction information can be a set of friction coefficients, for example each associated with a location within the track.

[0016] The friction information history data may be of the same type as the friction information to be determined or of a different type.

[0017] The runway context data can be selected from a data set including a representative value for precipitation, the number of landings on the runway during a given period, a representative value for wind, a representative value for humidity, etc. A person skilled in the art will be able to select this context data from this set, or even choose other context data, particularly if this other context data is available in real time, i.e., accessible / known for a prediction for the moment for which the process is implemented. The context data is chosen in particular if it has an impact on, or is related to, friction.

[0018] The prediction model can also be chosen by the person in the field according to the application.

[0019] According to a particular embodiment, after the last landing of an aircraft on the runway, the prediction is implemented for a given instant between an instant of said landing (for example its beginning) and an instant of obtaining real information relating to friction determined by the aircraft during said landing.

[0020] A person skilled in the art will be able to identify the moment when actual information relating to friction, determined by the aircraft during landing, is obtained. In particular, the actual information is not predicted and is obtained, for example, following a transfer of information determined by the aircraft during landing. This acquisition corresponds, for example, to acquisition by the computer system implementing the present method; in this sense, a person skilled in the art knows how to identify this actual, unpredicted information. This acquisition may correspond to the display of the information on a screen of the computer system.

[0021] For example, the time of acquisition may be subsequent to the following periods: aircraft taxiing (for example between 5 and 25 minutes), physical transfer of data (a few minutes via a communication network), and a frequency chosen to group and transfer the data (this frequency may be chosen by the airline).

[0022] In this particular embodiment, it is not necessary to wait for information from the last aircraft to land on the runway to estimate friction information (this information being obtained by the entity implementing the process remotely from the aircraft). Therefore, friction information can even be used for runway maintenance operations.

[0023] According to a particular embodiment, a first portion of the predicted friction information evolution curve of the runway is reconstructed step by step, between the moment of the last landing of an aircraft on the runway and the moment of obtaining actual information relating to the friction determined by the aircraft during said landing.

[0024] Step-by-step reconstruction can be implemented with a given time step, said time step being determined, for example, according to a sampling step for one of the data used as input by the prediction model. This time step can be that of the prediction (the prediction is made for a time interval spaced from the last date of the data in the time step).

[0025] According to a particular embodiment, after the last landing of an aircraft on the runway, the prediction for a time after the time of obtaining actual friction information determined by the aircraft during the landing is further implemented (thus, there is another prediction for this later time), the obtaining of runway context data including a prediction of runway context data.

[0026] In this particular embodiment, the runway context data is predicted. This can be done, for example, using weather forecasts. The context data obtained by prediction can be analogous to the context data described above. However, it is obtained from a predictive model, and therefore, a person skilled in the art will choose context data that can be predicted, for example, if predictive models are available. By way of example, data relating to wind and humidity are usually predicted.

[0027] It can be noted that a prediction may be necessary if the context data are not yet available for the given time, although that time is in the past, due to the time of transfer of the context data (typically a few minutes).

[0028] Also, in this embodiment, the prediction can be limited by using a time that is prior to the expiry of a given duration.

[0029] According to a particular embodiment, a second portion of the evolution curve of the predicted friction information of the runway is reconstructed step by step, from the moment of obtaining real information relating to friction determined by the aircraft during landing (and for example the moment of expiry of the given duration).

[0030] According to a particular implementation method, the predicted friction information is compared with a given threshold, and an alert signal is generated based on the result of the comparison.

[0031] This warning signal can be an audible or visual signal, for example configured to be returned to track management operators.

[0032] According to a particular embodiment, the predicted friction information includes a friction coefficient associated with the track, or includes a set of co friction efficiencies, each of the coefficients of said set being preferably associated with a location within the track.

[0033] It may be noted that the friction information history data may also include friction coefficients associated with the runway, or include sets of friction coefficients, each of the coefficients in said sets preferably being associated with a location within the runway.

[0034] According to a particular implementation method, the prediction model is a model trainable by machine learning.

[0035] Typically, the model can be an artificial neural network run by a computer system.

[0036] A person skilled in the art will be able to choose a type of model trainable by learning, and will be able, for example, to adapt the number of input neurons of the model according to the dimensions of the input data of the model.

[0037] Preferably, the prediction model is a model known by the English acronym LSTM ("Long Short-Term Memory"). Alternatively, a recurrent neural network model (usually designated in English by the acronym RNN: "Recurrent Neural Network") can be used.

[0038] According to a particular embodiment, the method includes a prior learning phase of the prediction model in which historical friction information data (typically so-called real data, obtained during aircraft landings with sensors) and previously obtained runway context data are used for a first instant, the learning being configured to increase a similarity between a model output and a prior friction information value, for a second instant spaced from the first instant by a given time step (the model output and the prior value each corresponding to the second instant).

[0039] The invention also proposes a system for predicting friction information for an aircraft runway comprising, for a given instant: a runway context data acquisition module, configured to obtain runway context data at a given instant, a friction information history data acquisition module, configured to obtain friction information history data for the runway comprising data all older than the given instant, and a prediction module using a prediction model, configured to receive as input at least the context data at the given time and the historical friction information data, and to deliver predicted friction information of the track at the given time.

[0040] This system can be configured for the implementation of all the implementation modes of the process as described above.

[0041] The invention also proposes a computer program comprising instructions for the execution of the steps of a process as defined above when said program is executed by a computer.

[0042] Note that the computer programs mentioned in this exposition may use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0043] The invention also proposes a computer-readable recording medium on which is recorded a computer program comprising instructions for executing the steps of a process as defined above.

[0044] The recording (or information) media mentioned in this description can be any entity or device capable of storing the program. For example, the media can include 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 floppy disk or a hard disk drive.

[0045] On the other hand, the recording media may correspond to a transmissible medium such as an electrical or optical signal, which can be transmitted via an electrical or optical cable, by radio, or by other means. The program according to the invention can, in particular, be downloaded onto an Internet-type network.

[0046] Alternatively, the recording media may correspond to an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the process in question. Brief description of the drawings

[0047] Other features and advantages of the present invention will become apparent from the description below, with reference to the accompanying drawings, which illustrate an example of an embodiment without being limiting in any way. In the figures:

[0048] [Fig-1] Fig. 1 is a schematic representation of the steps in a process of predicting friction information based on an example.

[0049] [Fig.2] The [Fig.2] is a schematic representation of steps comprising the process of the [Fig.1].

[0050] [Fig.3] The [Fig.3] is an illustration of a machine learning phase of the friction model.

[0051] [Fig.4] The [Fig.4] is a friction coefficient curve obtained according to an example of implementation of the friction information prediction process.

[0052] [Fig. 5] Fig. 5 shows a system for predicting friction information for example. Description of the implementation methods

[0053] We will now describe a method for predicting friction information for an aircraft runway.

[0054] This method can be implemented at any time (for example, a past time, the present, or a future time). In particular, this method can be implemented when friction information from the last aircraft to land on the runway is not yet available, as this information is only transferred to remote servers at a later time, thus delaying its availability to runway operators. Here, the method can provide runway operators with friction information for a runway at any time, and they can then provide this information to air traffic controllers. It is in this sense that friction information is useful for optimized air traffic management.

[0055] In [Fig.1], the steps of a PI process for predicting friction information for a runway at a given instant are schematically represented: the friction information is a prediction of the friction at that instant.

[0056] This process can be implemented by a computer system, as will be described in more detail with reference to [Fig.5].

[0057] In a first step S100, runway context data is obtained. If the given time is in the past, this context data may result from observations, and if the given time is in the future, this context data may result from a prediction. The runway context data may be chosen from a data set including a representative value for precipitation, the number of landings on the runway during a given period, a representative value for wind, a representative value for humidity, etc.

[0058] In a second step S101, a method is implemented to obtain historical friction information data for the runway, including data all older than the given time. This data may result from the landing and taxiing of aircraft on the runway (measurements taken during these phases provide friction information). Alternatively, this historical data may result, for example, in part from a previous implementation of the PL method.

[0059] The friction information history data can be of the same type as the predicted friction information obtained by means of the PL process. By "history", it is understood that they relate to one or more moments earlier than the given moment targeted by the process.

[0060] Finally, in a step S102, a prediction is implemented by a prediction model configured to receive as input at least the context data and the historical friction information data, and to deliver a predicted friction information of the track at the given time.

[0061] The prediction model is for example of the LSTM type (“Long Short-Term Memory”, i.e. a long short-term memory network).

[0062] The model here delivers a predicted friction information value, denoted as representative of the friction within the track. For example, if the data from steps S100 and S101 were associated with a time t, the prediction can be associated with a time t + dt where dt is a chosen time step.

[0063] Fig. 2 shows how the PI process described with reference to Fig. 1 can be implemented, for example in a runway management or even air traffic context.

[0064] The first step S001 designates the phase during which aircraft landed on the runway referred to herein. During these landings followed by taxiing, friction data were acquired in a manner known per se.

[0065] In a second step S002, friction information is obtained from the data acquired in step S001. It should be noted that for each aircraft, obtaining the actual (i.e., non-predicted) friction information from acquired data (e.g., from sensors) occurs after the entity implementing the process described herein (e.g., a computer system) has obtained actual aircraft information. It should be noted that the information transferred from the aircraft can be either friction information itself or friction-related data from which friction information can be obtained.

[0066] It may be noted that the determination of friction in an aircraft is known per se. For example, US patent 10202204 describes the obtaining of information relating to friction within an aircraft.

[0067] Next, a step-by-step reconstruction of a runway friction information evolution curve can be implemented between the last aircraft landing on the runway and the time at which actual information is obtained (the time Tl since the last landing has not yet expired). This is implemented by repeating the steps of the PI process shown in Figure 1, the repetition being designated by the reference RP1. To obtain a curve for the period Tl with a time step dt, n repetitions are performed. In each repetition, historical friction information data, partly estimated by the previous repetition (in other words, this is friction information predicted or obtained from landings preceding the one targeted here), and context data at the current time of the repetition are used.

[0068] It can be noted that Tl can typically be between 30 minutes and 6 hours, and depends, as indicated above, on the duration of the landing, taxiing, and the duration of one or more data transfers to a server (possibly with preprocessing), until the data is obtained by the entity implementing the process.

[0069] After reconstructing the curve for the entire period T1, it is possible to proceed to another reconstruction step designated by the reference RP1', in which, for a given duration T2, a first part of the friction information curve is reconstructed. For example, for a process that is implemented at the expiration of period T1, for which the RP1 repetitions have been implemented up to this expiration, observed context information was used during these RP1 repetitions. For the RP1' repetitions, predicted (and therefore unobserved) context information is used as input. To obtain a second part of the curve for period T2 with a time step dt, m repetitions are implemented.

[0070] It can be noted that T2 can typically be between 30 minutes and 2 hours. For example, T2 can be chosen according to the application by runway management operators. A short duration T2, for example, allows for more reliable information to be predicted.

[0071] In the figure, we note C the reconstructed friction information curve after implementation of steps RP1 and RP T.

[0072] During step S003, the predicted friction information obtained at a given time, for example illustrated on curve C, is compared with a given alert threshold visible in the figure. If the predictive friction information reaches the threshold, then an alert signal is generated during step S004, which can be relayed to runway managers.

[0073] As an indication, several thresholds can be used, each associated with the generation of an alert signal linked to a specific maintenance action to be carried out on the runway (inspection, snow removal, spreading of de-icing agent).

[0074] The threshold(s) are preferably predetermined and fixed. They can be set during an observation phase of the operations carried out for a track based on, for example, real friction information values.

[0075] As indicated above, the prediction model used can be of the LSTM type, and therefore be a model trainable by machine learning.

[0076] Figure 3 illustrates a machine learning phase of the friction information prediction model, which uses a previously obtained real friction information curve (for example, over a period on the order of one year). During an individual step of this learning phase, a time t is considered, and the model output for time t + dt, i.e., a second time, is obtained. The comparison between the model output and the actual value of the friction information at t + dt enables the implementation of the learning process.

[0077] Here, the model is configured to predict, for friction information of a single instant (either a single piece of friction information or a time window preceding t that includes several pieces of friction information), the friction information at time t + dt. The duration dt may be equal to the duration of the time window of the data used, or dt may differ from this duration.

[0078] Any suitable cost function and backpropagation may be used to then train the model using a distance between the model output and the real value at t + dt.

[0079] It can be noted that it is possible, during the model learning phase, to implement a model validation step by comparing real and predicted values.

[0080] Figure 4 shows in more detail the curve obtained after the implementation of repetitions RP1 and RP1', i.e., after implementations of the prediction process. The curve in Figure 4 is obtained after the implementation of the learning phase described with reference to Figure 3.

[0081] It is understood at this stage that it is possible to continuously establish friction information curves for a runway. Optionally, moving average processing can be implemented to track changes in friction information, for example, over windows of several flights (number of flights between, for example, 3 and 20) or over time windows (from 1 to 4 hours). Here, the curve remains above a given alert threshold, visible in the figure, compared with the predictive friction information. Thus, in this example, no alert signal is triggered.

[0082] This information can also be used to allow an airport to anticipate runway operations. For example, different thresholds can be used to trigger different actions (inspection, snow removal, de-icing).

[0083] [Fig.5] is a schematic representation of a system 100 which is a computer system configured to implement the process described above with reference to [Fig.1].

[0084] This computer system 100 comprises a processor 101 and a non-volatile memory 102. In the non-volatile memory, a computer program 103 is stored. The computer program 103 contains instructions for executing the steps of a process as described above.

[0085] Thus, the processor 101 and the computer program form, during program execution: a module for obtaining track context data at a given time, a module for obtaining historical friction information data for the track including data all older than the given time, a prediction module by a prediction model configured to receive as input at least the context data at the given time and the historical friction information data, and to deliver a prediction of the track friction information at the given time.

[0086] The embodiments described above make it possible to improve the management of the runway(s), in particular in terms of monitoring the condition of a runway.

[0087] In addition, the embodiments described above make it possible to obtain knowledge about friction (or adhesion) as perceived by an aircraft, with its level, its variability, its temporal evolution.

[0088] In particular, the method makes it possible to obtain friction information continuously, in order to optimize interventions on the track.

[0089] For example, the process can even make it possible to measure the effectiveness of chemicals deposited on runways, or to compare friction levels between airports, and the variability of friction.

[0090] This process finds application in airport management, in runway management, and can be useful for airlines.

Claims

Demands

1. A method for predicting friction information for an aircraft runway implemented by a computer system, said method comprising a plurality (n, ni) of iterations of the following steps: a) obtaining (S 100) context data for the runway for a current time t, b) obtaining (S 101) historical friction information data for the runway, comprising data all older than the current time t, c) prediction (S 102) by a prediction model configured to receive as input at least said context data and said historical friction information data, and to deliver predicted friction information (^) for the runway for a given time t+dt;the process further comprising: a step-by-step reconstruction (RP1) of a first portion of the evolution curve of the predicted friction information (^) of the runway, between a moment of a last landing of an aircraft on the runway and a moment of obtaining real information relating to a friction determined by the aircraft during said landing, each point of the curve corresponding to a predicted friction information ( / z) during one of the iterations of steps a) to c).;

2. A method according to claim 1, wherein after the last landing of an aircraft on the runway, the prediction (S102) is further implemented for a time after the time of obtaining actual information relating to friction determined by the aircraft during said landing, the obtaining (S100) of runway context data comprising a prediction of runway context data.

3. A method according to claim 2, wherein a second portion of the runway friction information evolution curve is reconstructed step by step (RPT), from the moment of obtaining actual friction information determined by the aircraft during landing.

4. A method according to any one of claims 1 to 3, wherein the predicted friction information is compared with a given threshold and an alert signal (S004) is generated based on the result of the comparison. comparison.

5. A method according to any one of claims 1 to 4, wherein the predicted friction information includes a friction coefficient associated with the track, or includes a set of friction coefficients, each of the coefficients of said set preferably being associated with a location within the track.

6. A method according to any one of claims 1 to 5, wherein the prediction model is a machine-learning trainable model, the method comprising a prior training phase of the prediction model in which historical friction information data and previously obtained track context data are used for a first instant, the training being configured to increase a similarity between a model output and a prior friction information value, for a second instant spaced from the first instant by a given time step.

7. System for predicting friction information for an aircraft runway, the system comprising: a runway context data acquisition module, configured to obtain runway context data for a current time t, a friction information history data acquisition module, configured to obtain friction information history data for the runway comprising all data older than the current time t, a prediction module using a prediction model, configured to receive as input at least said context data and said friction information history data, and to deliver predicted runway friction information for a given time t+dt, the system further comprising: a step-by-step reconstruction module (RP1) of a first portion of the evolution curve of the predicted friction information (^) of the runway,between the moment of the last landing of an aircraft on the runway and the moment of obtaining actual information relating to friction determined by the aircraft during said landing, each point on the curve corresponds to a piece of friction information predicted ( / z) by said prediction module at different times.

8. A computer program comprising instructions for carrying out the steps of a process according to any one of claims 1 to 6 when The program is executed by a processor.

Images