Method for enriching a structured database, method for assisting in piloting an aircraft, computer program product and associated electronic device

Abstract

Method for enriching a structured database, method for assisting in piloting an aircraft, computer program product and associated electronic device The present invention relates to a method (100) for enriching a structured database intended to assist in the decision-making of an aircraft pilot.The method comprises the following steps: obtaining (110) an initial structured database, called the initial structured database, comprising a plurality of objects, determining (120) at least two indicators from the group of indicators consisting of: a frequency indicator, an interconnection indicator, a relevance indicator, and a precision indicator, comparing (130) each determined indicator with at least one respective threshold, if the result of the comparison is negative, sending (140) a command to enrich the initial structured database to form an enriched structured database, called the enriched structured database, comprising a greater number of objects than the initial structured database. Figure for the abstract: Figure 6.

Description

Title of the invention: Method for enriching a structured database, method for assisting in piloting an aircraft, computer program product and associated electronic device

[0001] The present invention relates to a method for enriching a structured database. The present invention also relates to an electronic enrichment device suitable for implementing such a method.

[0002] The present invention also relates to a pilot assistance method and an electronic pilot assistance system.

[0003] The present invention also relates to computer program products suitable for implementing such methods.

[0004] The present invention relates to the field of pilot assistance systems.

[0005] In the field, it is known to use a flight computer to provide data to the pilot of the aircraft. As a general rule, the flight computer stores a database comprising a plurality of data that the pilot of the aircraft is able to consult.

[0006] Conventionally, the flight computer displays on displays information relating to the current and future state of the aircraft, so that the pilot of the aircraft can make the best decisions.

[0007] More sophisticated versions of flight computers include a human-machine interface with which the pilot can interact. For example, the pilot can query the databases stored in the flight computer to obtain richer and more contextualized data. This is called knowledge. Thanks to this knowledge, the pilot has a more detailed understanding of his environment and the situation he encounters. The decisions taken by the pilot are then of better quality.

[0008] It is clear that the quality of the decisions that the pilot makes then depends heavily on the quality of the knowledge that can be extracted from the database. In other words, the quality of these decisions intrinsically depends on the database itself.

[0009] However, it is difficult to assess whether the database is suitable for the intended use.

[0010] For example, if the database has too little data, it will not be able to provide an appropriate response to each situation. Conversely, if the database includes too much data, it will take a long time to search and may provide too many responses for the user to have time to analyze them all.

[0011] As another example, the way in which the database is organized is also of great importance. Indeed, if the database is not properly organized / structured, it may prove particularly complex to extract intelligible knowledge from it.

[0012] There is therefore a need for a tool to ensure that the database in question is sufficiently complete while remaining intelligible so that a user is able to use it.

[0013] To this end, the present invention relates to a method for enriching a structured database intended to assist the decision-making of an aircraft pilot, the method being implemented by an electronic enrichment device and comprising the following steps: - obtaining an initial structured database, called the initial structured database, comprising a plurality of objects including: • a plurality of classes, • a plurality of data, and • semantic links associating each piece of data with at least one class,

[0014] each data and / or each class comprising one or more words, - determination of at least two indicators from the group of indicators consisting of: • a frequency indicator quantifying the frequency of appearance of each word in the initial structured base, • an interconnection indicator representative of a distribution of semantic links in the initial structured base, • a relevance indicator quantifying the usefulness of the data in the initial structured database, • a precision indicator quantifying the polysemy of each word in the initial structured base, - comparison of each determined indicator with at least one respective threshold,

[0015] if the result of the comparison is negative, sending a command to enrich the initial structured base to form an enriched structured database, called an enriched structured base, comprising a greater number of objects than the initial structured base.

[0016] By evaluating the various indicators, it is possible to ensure that the initial structured base, or the enriched structured base, is sufficiently complete in terms of information while limiting the number of data it contains. Thus, the structured database resulting from the method according to the invention is optimized so that its use by a user is efficient.

[0017] According to other advantageous aspects of the invention, the enrichment method comprises one or more of the following characteristics, taken in isolation or in all technically possible combinations: - the indicator determination stage includes the determination of at least three indicators from the group of indicators,

[0018] the step of determining the indicators preferably comprising the determination of each indicator of the group of indicators; - one of the indicators determined during the determination step is the frequency indicator, the determination step comprising: • the calculation of the Shannon entropy associated with each word of the initial structured base, • the evaluation of the percentage of words whose Shannon entropy is greater than a respective value, • applying a statistical test to the calculated Shannon entropies to obtain a p-value and a percentage of similarity,

[0019] the information quantity indicator comprising: • the percentage of words whose Shannon entropy is greater than the predefined value, • the p-value, and • the percentage of similarity; • one of the indicators determined during the determination step is the interconnection indicator, the determination step comprising: • the calculation of the proportion of the number of classes in the initial structured base including an explanatory commentary on the content of the class, • the calculation of the number of semantic links connecting each class to data in the initial structured base, • the evaluation of the number of classes linked to data by a number of semantic links less than a first respective value or greater than a second respective value, • the calculation of a modularity value of the classes of the initial structured base,

[0020] the interconnection indicator comprising: • the number of classes in the initial structured base including an explanatory commentary on the content of the class, • the number of classes linked to data by a number of semantic links less than the first respective value or greater than the second respective value, and • the modularity value of the classes; • one of the indicators determined during the determination step is the precision indicator, the determination step comprising: • obtaining a dictionary comprising a plurality of words, and, for each word, at least one meaning, • the comparison of each word of each class of the initial structured base with the words of the dictionary, • the calculation of a percentage of words from the initial structured base appearing in the dictionary, • among the words appearing in the dictionary, the calculation of a percentage of words having a single meaning in the dictionary, • the transformation of the initial structured base into a graph, the graph comprising a central class, called the mother class, linked to a plurality of classes, called daughter classes, • the calculation of the Shannon entropy associated with each daughter class, • the evaluation of a percentage of daughter classes whose Shannon entropy is greater than a respective value,

[0021] the precision indicator comprising: • the percentage of words from the initial structured base appearing in the dictionary, • percentage of words in the initial structured base having a single meaning in the dictionary, • the percentage of daughter classes whose Shannon entropy is greater than the respective value; • the initial structured base comprises a plurality of scenarios each comprising a set of values ​​and an action taken by the pilot of the aircraft in the presence of this set of values, each value of the set of values ​​being a numerical value of a data item of the initial structured base,

[0022] one of the indicators determined during the determination step being the relevance indicator, the determination step comprising: • obtaining a trained artificial intelligence model to receive a scenario and determine a logical rule combining the values ​​of the set of values ​​resulting in the action of said scenario, and to quantify the importance of each value in the action taken, • application of said artificial intelligence model to each scenario of the initial structured base, to obtain, for each value of each set of values, a quantifier of importance in the action of said scenario, • calculation of the proportion of values ​​of the sets of values ​​for which the quanti-

[0023] importance indicator is below a respective threshold, the relevance indicator including: said proportion of values ​​of the sets of values ​​for which the importance quantifier is less than the respective threshold; following the step of sending an enrichment command, the steps of obtaining, determining, comparing and, where appropriate, sending the process are repeated as long as the result of the comparison is negative,

[0024] the structured base enriched by an iteration being the initial structured base of the following iteration,

[0025] the structured database obtained at the end of the iterations being called the validated structured database; and the method comprises, if the result of the comparison is positive, a step of importing the validated structured base into an electronic device for interaction with the pilot of the aircraft, for its use during an avionics mission.

[0026] The invention also relates to a method for assisting in piloting an aircraft from an initial structured database, called the initial structured database, the method comprising an enrichment phase and an exploitation phase,

[0027] the enrichment phase comprising an enrichment of the initial structured base by application of such an enrichment method,

[0028] the operating phase being implemented by an electronic device for interaction with the pilot of the aircraft, the electronic interaction device comprising the structured database resulting from the enrichment phase, called validated structured database, and comprising the following steps: receiving a request from the aircraft pilot, selection of objects, from the validated structured base, in response to the request received, transmission of a piloting instruction developed from the selected objects, intended for one of: the pilot of the aircraft and an actuator of the aircraft.

[0029] The invention also relates to a computer program product, a computer program product, comprising software instructions which, when executed by a computer, implement such a method.

[0030] The invention also relates to an electronic device for enriching a structured database intended to assist the decision-making of an aircraft pilot, the electronic enrichment device comprising: an obtaining module configured to obtain an initial structured database, called initial structured database, comprising a plurality of objects including: • a plurality of classes, • a plurality of data, and • semantic links associating each piece of data with at least one class,

[0031] each data and / or each class comprising one or more words, - a determination module configured to determine at least two indicators from the group of indicators consisting of: • a frequency indicator quantifying the frequency of appearance of each word in the initial structured base, • an interconnection indicator representative of a distribution of semantic links in the initial structured base, • a relevance indicator quantifying the usefulness of the data in the initial structured database, • a precision indicator quantifying the polysemy of each word in the initial structured base, - a comparison module configured to compare each determined indicator with at least one respective threshold, - a sending module configured to, if the result of the comparison is negative, send a command to enrich the initial structured database to form an enriched structured database, called an enriched structured database, comprising a greater number of objects than the initial structured database.

[0032] The invention also relates to an electronic device for enriching a structured database intended to assist the decision-making of an aircraft pilot, the electronic enrichment device comprising: - an obtaining module configured to obtain an initial structured database, called initial structured database, comprising a plurality of objects including: • a plurality of classes, • a plurality of data, and • semantic links associating each piece of data with at least one class,

[0033] each data and / or each class comprising one or more words, - a determination module configured to determine at least two indicators from the group of indicators consisting of: • a frequency indicator quantifying the frequency of appearance of each word in the initial structured base, • an interconnection indicator representative of a distribution of semantic links in the initial structured base, • a relevance indicator quantifying the usefulness of the data in the initial structured database, • a precision indicator quantifying the polysemy of each word in the initial structured base, - a comparison module configured to compare each determined indicator with at least one respective threshold, - a sending module configured to, if the result of the comparison is negative, send a command to enrich the initial structured database to form an enriched structured database, called an enriched structured database, comprising a greater number of objects than the initial structured database.

[0034] The invention also relates to an electronic system for assisting in piloting an aircraft from an initial structured database, called the initial structured database, the system comprising an electronic enrichment device comprising: - an obtaining module configured to obtain an initial structured database, called the initial structured database, comprising a plurality of objects among which: • a plurality of classes, • a plurality of data, and • semantic links associating each data with at least one class,

[0035] each data and / or each class comprising one or more words, - a determination module configured to determine at least two indicators from the group of indicators consisting of: • a frequency indicator quantifying the frequency of appearance of each word in the initial structured base, • an interconnection indicator representative of a distribution of semantic links in the initial structured base, • a relevance indicator quantifying the usefulness of the data in the initial structured database, • a precision indicator quantifying the polysemy of each word in the initial structured base, - a comparison module configured to compare each determined indicator with at least one respective threshold, - a sending module configured to, if the result of the comparison is negative, send a command to enrich the initial structured database to form an enriched structured database, called a structured database enriched, including a greater number of objects, than the initial structured base,

[0036] the system further comprising an electronic device for interaction with a pilot of the aircraft, the electronic interaction device comprising the structured database resulting from the enrichment module phase, called validated structured database, and comprising: - a reception module configured to receive a request from the aircraft pilot, - a selection module configured to select objects from the validated structured base, in response to the received request, - a transmission module configured to transmit a piloting instruction developed from the selected objects, to one of: the pilot of the aircraft and an actuator of the aircraft.

[0037] The invention will appear more clearly on reading the description which follows, given solely by way of non-limiting example, and made with reference to the drawings in which:

[0038] [Fig-1] [Fig.l] is a diagram of an electronic pilot assistance system according to the invention,

[0039] [Fig.2] [Fig.2] is a schematic representation of a structured base of data used by the electronic system according to the invention,

[0040] [Fig.3] [Fig.3] is a visual representation of a frequency indicator determined by an electronic device for enriching the electronic decision support system of [Fig.l],

[0041] [Fig.4] [Fig.4] is a visual representation of an element determined by a electronic device for enriching the electronic decision support system of [Fig.l],

[0042] [Fig.5] [Fig.5] is a visual representation of an element determined by a electronic device for enriching the electronic decision support system of [Fig.l],

[0043] [Fig.6] [Fig.6] is a flowchart of a pilot assistance method implemented by the decision support system of [Fig.l].

[0044] In [Fig.l] an electronic system 10 for assisting in piloting an aircraft 15 is shown.

[0045] The system 10 comprises an electronic device 20 for enriching a structured database which will be described below, and an electronic device 25 for interaction with a pilot of the aircraft 15.

[0046] The enrichment device 20 is preferably remote from the aircraft 15.

[0047] The enrichment device 20 comprises a module 30 for obtaining a base initial data structure 32, called initial structured base 32, a module 35 for determining indicators which will be described below, a module 40 for comparing the determined indicators with respective thresholds, and a sending module 45.

[0048] The interaction device 25 is preferably on board the aircraft 15.

[0049] The interaction device 25 comprises a module 50 for receiving a request from a pilot of the aircraft 15, a selection module 55 and a transmission module 60.

[0050] In the example of [Fig. 1], the electronic enrichment device 20 and the electronic interaction device 25 respectively comprise a first and a second information processing unit.

[0051] The first information processing unit is for example formed of a first memory 70 and a first processor 75 associated with the first memory 70.

[0052] In the example of [Fig.l], the obtaining module 30, the determining module 35, the comparing module 40, and the sending module 45, are each produced in the form of software, or a software brick, executable by the first processor 75. The first memory 70 of the electronic enrichment device 20 is then capable of storing obtaining software, determining software, comparing software, and sending software. The first processor 75 is then capable of executing each of the software among the obtaining software, the determining software, the comparing software and the sending software.

[0053] Similarly, the second information processing unit is for example formed of a second memory 80 and a second processor 85 associated with the second memory 80.

[0054] In the example of [Fig.l], the reception module 50, the selection module 55, and the transmission module 60 are each produced in the form of software, or a software brick, executable by the second processor 85. The second memory 80 of the electronic interaction device 25 is then capable of storing reception software, selection software and transmission software. The second processor 85 is then capable of executing each of the software among the reception software, the selection software, and the transmission software.

[0055] In a variant not shown, the obtaining module 30, the determining module 35, the comparing module 40, and the sending module 45, as well as, as an optional addition, the receiving module 50, the selecting module 55 and the sending module 60, are each produced in the form of a programmable logic component, such as an FPGA (Field Programmable Gate Array) or an integrated circuit, such as an ASIC (Application Specific Integrated Circuit).

[0056] When the electronic enrichment device 20 and the electronic interaction device 25 are respectively produced in the form of one or more software programs, that is to say in the form of a computer program, also called a computer program product, they are furthermore capable of being recorded on a medium, not shown, readable by a computer. The computer-readable medium is for example a medium capable of storing electronic instructions and of being coupled to a bus of a computer system. For example, the readable medium is an optical disk, a magneto-optical disk, a ROM memory, a RAM memory, any type of non-volatile memory (for example FLASH or NVRAM) or a magnetic card. A computer program comprising software instructions is then stored on the readable medium.

[0057] Preferably, the interaction device 25 further comprises a display screen 65 and / or a keyboard 67 so that the pilot can communicate a request to the interaction device 25.

[0058] The obtaining module 30 is configured to obtain the initial structured database 32, called initial structured database 32.

[0059] In the present application, the term “structured database” means a database comprising a plurality of objects including: a plurality of classes, a plurality of data, also called class instances, and semantic links associating each data with at least one class.

[0060] Preferably, the initial structured base 32 is an ontology.

[0061] Each data / class instance, and each class, comprises one or more words. Preferably, each class and each class instance is a noun phrase.

[0062] In the present application, the term “class instance” means data intended to qualify or quantify an aspect of a class.

[0063] Preferably, certain semantic links also connect two classes together, to indicate a dependency link between these two classes.

[0064] Thanks to the semantic links, it is possible to represent the initial structured base 32 in the form of a graph.

[0065] [Fig.2] shows an example of an initial structured base 32 comprising the following classes: "aircraft", "physical properties", "air traffic control", "location", "approach category", "flight properties", "call sign", "controller", "weather", "air speed", "altitude", and "arrival time".

[0066] In this example, the different class instances are: "abort", "key terms in communication", "acknowledge", "calculated actual landing time", "affirmative", "aircraft type", "alert", "cleared", "commands", "crosswind component", "weather", "departure time", "dew point", "elevation", "airport", "error", "flight rules", "remaining fuel", "gate (boarding)", "terminal", "ground speed", "heading", "runway", "maintain your position", "identifier", "immediately", "latitude", "length", "longitude", "mayday mayday mayday", "negative", "pilot", "pitch", "pitch rate", "roll", "roll rate", "sky condition", "hold", "temperature", "turn rate", "visibility", "wake turbulence", "weight class", and "wind shear".

[0067] In [Fig.2], the sematic links are represented by continuous lines connecting a class to a class instance, or two classes to each other.

[0068] In a manner not shown, in the initial structured base 32, at least one class and preferably several classes, comprise an explanatory commentary on the content of the class. In the previous example, an explanatory commentary of the class is for example “An aircraft has a latitude and is located at an airport”.

[0069] Optionally, at least some data / class instances correspond to quantifiable physical parameters. In the preceding example, these data / class instances are for example the following: "air speed", "altitude", "location", "crosswind component", "dew point", "elevation", "ground speed", "heading", "latitude", "length", "longitude", "pitch", "roll", "roll rate", "temperature", "turn rate", and "wind shear".

[0070] The initial structured base 32 optionally comprises a plurality of scenarios each comprising a set of values ​​and an action taken by the pilot of the aircraft 15 in the presence of this set of values. Each value of the set of values ​​is a numerical value of a data item of the initial structured base 32. Preferably, each value of a set of values ​​is a numerical value of a class instance corresponding to a physical parameter. Thus, the class instance and the associated value form knowledge.

[0071] As an example, a scenario is during a landing, an evaluation of a go-around of an aircraft, the proposed module evaluates the following class instances: air speed, altitude and roll of aircraft class; the crosswind component and wind shear from weather class; the heading of the airport runway.

[0072] For example, from an operational point of view, if the situation returns non-nominal physical parameters (and their interactions), the operator is asked to perform a go-around of the aircraft. More precisely, if the aircraft is not aligned by its roll with the runway, and / or if the aircraft has an air speed higher than the recommended speed for landing (higher by 20% compared to the regulations in force), and / or if the wind shear or the crosswind component is too high (higher by 15% compared to the regulations), the pilot is asked to perform a go-around.

[0073] Preferably, the obtaining module 30 is configured to receive the initial structured base 32 from a user of the enrichment device 20.

[0074] The determination module 35 is configured to determine at least two indicators from the group of indicators consisting of: a frequency indicator quantifying a frequency of appearance of each word in the initial structured base 32, an interconnection indicator representative of a distribution of the semantic links in the initial structured base 32, a relevance indicator, and a precision indicator quantifying the polysemy of each word of the initial structured base 32.

[0075] Preferably, the determination module 35 is configured to determine at least three indicators of said group of indicators.

[0076] More preferably, the determination module 35 is configured to determine each indicator of said group.

[0077] The frequency indicator makes it possible to quantify whether the words present in the initial structured base 32 are semantically sufficiently distinct from each other and thus to ensure that it does not have too many synonyms which could harm the intelligibility of the initial structured base 32.

[0078] To determine the frequency indicator, the determination module 35 is preferably configured to calculate the Shannon entropy associated with each word of the initial structured base 32.

[0079] For this purpose, the determination module 35 is for example configured to calculate for each word of the initial structured base 32, a similarity score with each other word of the structured database 32. The determination module 35 is for example configured to use a “sentence transformer” type model to associate a vector of numerical values ​​with each word, and to calculate the similarity score between two words as being a normalized scalar product between the vectors associated with each of the two words. The “transformed sentence” type model is for example derived from the Universal sentence encoder tool from Google®.

[0080] Preferably, the Shannon entropy is calculated according to the following formula:

[0081] [Math.l] «) P(m, n))

[0082] where m is a word present in the initial structured base 32,

[0083] ô(m) is the Shannon entropy of the word m,

[0084] Pim, n) is the similarity score between the wordIn and the wordn, and

[0085] log(.) is the logarithmic function in base 2.

[0086] Preferably again, the determination module 35 is configured to then evaluate the percentage of words, in the initial structured base 32, whose Shannon entropy is greater than a first predefined value. The first predefined value is for example equal to 5%.

[0087] The determination module 35 is configured to then apply a statistical test to the calculated Shannon entropies to obtain a p-value and a similarity percentage. The statistical test is, for example, a Student test known per se.

[0088] Preferably, the determination module 35 is configured to obtain, when applying the Student test, for each word of the initial structured base 32, a similarity score, for example between 0 and 1. The determination module 35 is then configured to calculate the similarity percentage as being the percentage of words of the initial structured base 32, the similarity score of which is greater than a second predefined value. The second predefined value is for example equal to 0.8.

[0089] The p-value quantifies whether the initial structured base 32 has enough classes.

[0090] The percentage of similarity quantifies, for its part, whether the initial structured base 32 does not contain too many classes.

[0091] The determination module 35 is then configured to determine the frequency indicator as comprising the percentage of words whose Shannon entropy is greater than the first predefined value, the p-value, and the percentage of similarity.

[0092] In the previous example, the percentage of words whose Shannon entropy is greater than the first predefined value is equal to 5%, the p-value is equal to 4.50418514748828 10 17, and the similarity percentage is equal to 0.4563605248146035.

[0093] [Fig.3] illustrates an example of a computer window representing these results.

[0094] In [Fig.3], “entropy value” corresponds to the sum of the calculated entropies. In [Fig.3], "max entropy value" corresponds to the said sum if all the similarity scores were equal to 0.5, i.e. if all the words were substantially similar. In [Fig.3], "ratio" corresponds to the ratio between "entropy value" and "max entropy value".

[0095] The interconnection indicator quantifies whether the class instances are well distributed among the classes of the initial structured base 32. Indeed, a structured base comprising a strongly majority class comprising almost all of the class instances would prove difficult to browse since it would require going through too many branches.

[0096] To determine the interconnection indicator, the determination module 35 is preferably configured to calculate the proportion of the number of classes of the initial structured base 32 comprising an explanatory commentary of the content of the class.

[0097] Preferably, the determination module 35 is further configured to calculate the number of semantic links connecting each class to a data / class instance, in the initial structured base 32. This number of semantic links quantifies the number of class instances associated with each class.

[0098] Optionally, the determination module 35 is further configured to evaluate the number of classes linked to data / class instances, by a number of semantic links less than a respective first value, called the third predefined value, or greater than a respective second value, called the fourth predefined value.

[0099] For this purpose, the determination module 35 is preferably configured to sort the classes according to the number of semantic links connecting said class to a class instance.

[0100] [Fig.4] represents a histogram of the classes thus sorted in the previous example of initial structured base 32. The abscissa of [Fig.4] represents the classes of the initial structured base 32. The ordinate of [Fig.4] represents the number of class instances associated with each class, i.e. the number of semantic links connecting said class and a class instance.

[0101] Still for this purpose, the determination module 35 is preferably configured to calculate a box plot on the number of said links per class.

[0102] The third value then corresponds to the value of the first quartile, and the fourth value corresponds to the value of the third quartile. In other words, the determination module 35 is preferably configured to determine the number of classes of the initial structured base 32, the number of instances of which is not between the first and third quartiles.

[0103] In the previous example, this number of classes is equal to 2, or 16.67% of the classes.

[0104] The determination module 35 is further configured to calculate the modularity of the classes of the initial structured base 32.

[0105] For this purpose, the determination module 35 is for example configured to apply the Louvain algorithm, known per se. The Louvain algorithm makes it possible to form clusters of classes and instances and provides a modularity value between -1 and 2.1.

[0106] [Fig.5] represents an example of the clusters of classes and instances resulting from the Louvain algorithm applied to the previous example. As visible in [Fig.5], the most important classes of each cluster, i.e. those connected by a greater number of semantic links in each cluster, are represented in bold in a larger font than the other classes and class instances. In [Fig.5], the central classes are: “aircraft”, “flight properties”, “controls”, “weather”, “airport”, “key terms in communication”, and “air traffic control”.

[0107] The modularity value quantifies the randomness of semantic links between classes and class instances, or between pairs of classes. The higher the value of mo The higher the modularity, the less random the semantic links are, i.e. the more well organized they are. In the previous example and with reference to [Fig.5], the modularity value is for example equal to 0.7.

[0108] The determination module 35 is preferably configured to determine the interconnection indicator as comprising: the proportion of the number of classes of the initial structured base 32 comprising an explanatory commentary of the content of the class, the number of classes linked to data by a number of semantic links less than the third predefined value or greater than the fourth predefined value (preferably the result of the box plot), and the modularity value of the classes.

[0109] The accuracy indicator quantifies the extent to which the initial structured base data 32 is unambiguously understandable.

[0110] To determine the precision indicator, the determination module 35 is preferably configured to obtain a dictionary comprising a plurality of words, and, for each word, at least one meaning. The dictionary, not shown, is for example stored in the first memory 70. The dictionary is for example the World Net dictionary known per se.

[0111] The determination module 35 is configured to then compare each word of each class of the initial structured base 32 to the words of the dictionary, and to calculate a percentage of words of the initial structured base 32 appearing in the dictionary. Optionally, the determination module 35 is configured to compare each word of the initial structured base 32, i.e. the class instances included, to the words of the dictionary.

[0112] The determination module 35 is preferably configured to calculate, among the words appearing in the dictionary, a percentage of words having a single meaning in the dictionary.

[0113] The determination module 35 is further optionally configured to transform the initial structured base 32 into a graph, for example as shown in [Fig.2]. The graph comprises at least one central class, called the parent class, linked to a plurality of classes, called daughter classes. In the previous example, the parent classes are “aircraft”, “weather”, “key terms in communication”, and “airport”, as shown in [Fig.5].

[0114] The determination module 35 is optionally configured to then calculate the Shannon entropy associated with each daughter class as described previously by replacing each word with the name of a daughter class.

[0115] The determination module 35 is then configured to evaluate a percentage of daughter classes whose Shannon entropy is greater than a respective predefined value, i.e. fifth predefined value.

[0116] The determination module 35 is preferentially configured to determine the precision indicator as comprising the percentage of words from the initial structured base 32 appearing in the dictionary, the percentage of words from the initial structured base 32 having a single meaning in the dictionary, and the percentage of daughter classes whose Shannon entropy is greater than the fifth predefined value.

[0117] The relevance indicator quantifies how important each data / class instance is for decision-making. If the initial structured base 32 includes too many data / class instances whose usefulness is low for decision-making, it will be particularly cumbersome to handle.

[0118] To determine the relevance indicator, the determination module 35 is configured to obtain an artificial intelligence model trained to: receive a scenario, determine a logical rule combining the values ​​of the set of values ​​resulting in the action of said scenario, and to quantify the importance of each value in the action taken.

[0119] In particular, said model is preferably configured to quantify the importance of each knowledge, i.e. of the class instance having the corresponding value of the set of values.

[0120] Preferably, said model is a “Logic Tensor Networks” type model, also called LTN model, known per se and notably described in the article “Logic Tensor Networks: Deep Learning and Logical reasoning from Data and Knowledge” by Serafini et al.

[0121] Such a model is for example a neural network trained on a plurality of historical scenarios. The logical rules that the LTN model is able to determine include for example only elementary operators such as “greater than”, “less than”, “AND”, “OR”, “NOT” and “IMPLIES”.

[0122] The LTN model is preferentially stored in the first memory 70.

[0123] The determination module 35 is configured to apply said LTN model to each scenario of the initial structured base, to obtain, for each value of each set of values, i.e. for each knowledge, an importance quantifier in the action of said scenario. For example, each importance quantifier is a value between 0 and 1. An importance quantifier close to 0 corresponds to unimportant knowledge, while an importance quantifier close to 1 corresponds to very important knowledge.

[0124] The determination module 35 is configured to calculate the proportion of values ​​of the sets of values ​​for which the importance quantifier is less than a respective threshold, i.e. the sixth predefined threshold. The sixth predefined threshold is for example equal to 0.2.

[0125] The relevance indicator comprises said proportion of values ​​of the value sets for which the importance quantifier is lower than the respective threshold. In other words, the relevance indicator comprises the proportion of knowledge, from the scenarios, for which the importance quantifier is lower than the sixth predefined threshold.

[0126] The comparison module 40 is configured to compare each determined indicator with at least one respective threshold.

[0127] When the or each determined indicator comprises several values, said comparison is considered to be the comparison of each value of said indicator to a respective threshold value.

[0128] In particular, for the frequency indicator, the respective threshold of the percentage of words whose Shannon entropy is greater than the first predefined value is for example equal to 0.002. Still for the frequency indicator, the respective threshold of the p-value is for example equal to 0.05, i.e. the result of this comparison is positive if the p-value is less than 0.05, and negative otherwise. Still for the frequency indicator, the respective threshold of the similarity percentage is for example 0.8, i.e. the result of this comparison is positive if the p-value is greater than 0.8.

[0129] For the interconnection indicator, the respective threshold of the proportion of the number of classes comprising an explanatory commentary on the content of the class is, for example, equal to 80%, i.e. the result of this comparison is positive if the proportion of the number of said classes is greater than this threshold, and negative otherwise. For the interconnection indicator still, the respective threshold of the number of classes connected to data by a number of semantic links less than the third predefined value or greater than the fourth predefined value is preferentially defined by the first and third quartiles of the numbers of connections between the different classes, i.e. the result of this comparison is positive if the number of said classes is between these quartiles, and negative otherwise.For example, if the class "airport" has three connections, we check whether this class is between the first and third quartile of the data set which corresponds to all the connections of each class.

[0130] Still for the interconnection indicator, the respective threshold of the modularity value of the classes is for example equal to 0, i.e. the result of this comparison is positive if the modularity value is greater than this threshold, and negative otherwise.

[0131] For the precision indicator, the respective threshold of the percentage of words of the initial structured base 32 appearing in the dictionary is for example equal to 80%, i.e. the result of this comparison is positive if said percentage is greater than this threshold, and negative otherwise. Still for the precision indicator, the respective threshold of the percentage of words of the initial structured base 32 having a single meaning in the dictionary is for example equal to 80%, i.e. the result of this comparison is positive if said percentage is greater than this threshold, and negative otherwise. Still concerning the precision indicator, the respective threshold of the percentage of daughter classes whose Shannon entropy is lower than the fifth predefined value, is for example 80%, i.e. the result of this comparison is positive if said percentage is higher than this threshold, and negative otherwise.

[0132] For the relevance indicator, the respective threshold of the proportion of knowledge whose quantifier is lower than the sixth predefined threshold and for example equal to 66% i.e. the result of this comparison is positive if said proportion is higher than this threshold, and negative otherwise.

[0133] The comparison module 40 is preferably configured to return a negative result if the comparison of at least one of the indicators, or of at least one value of at least one indicator, with the respective threshold, returns a negative result. It is clear that the comparison module 40 is then configured to return a positive result only if the comparison of each indicator, i.e. of each value of each indicator, with the respective threshold, returns a positive result.

[0134] The sending module 45 is configured to, if the result of the comparison is negative, send a command to enrich the initial structured base 32 to form an enriched structured database, called an enriched structured base, comprising a greater number of objects than the initial structured base 32.

[0135] The sending module 45 is for example configured to send this enrichment command to a user of the enrichment device 20 so that he manually enriches the initial structured base 35, for example by adding one or more classes, class instances and / or semantic links. Alternatively, the sending module 45 is configured to send the enrichment command to an automatic enrichment system not shown. Such an automatic system is for example configured to concatenate the initial structured base 32 with another structured database, for example found on the internet, to form the enriched structured base.

[0136] In the previous example, an enrichment optionally includes the addition of the data “remaining fuel”, “aircraft”, and “airport”, “communication” and semantic links connecting these data to the “alert” data which then becomes a class of the enriched structured base.

[0137] The enrichment device 20 is preferably configured to repeat the actions of its modules 30, 35, 40, 45 from the enriched structured base, rather than from the initial database 32.

[0138] The sending module 45 is configured to, if during an iteration, the result of the comparison is positive, send the initial structured base 32 to the interaction device 25, preferably for its storage in the second memory 80. It is clear that if the current iteration is another iteration than the first, the base The initial structured database 32 considered is the enriched structured database resulting from the previous iteration. Thus, the structured database that the sending module 45 is configured to send will subsequently be named validated structured database 95.

[0139] The interaction device 25 will now be described.

[0140] As indicated previously, the interaction device 25 stores the validated structured base 95 obtained from the sending module 45.

[0141] The reception module 50 is configured to receive a request from the pilot of the aircraft 15. The request is for example a sentence such as “Search for the list of destination airports according to my remaining fuel and the weather at the destination”.

[0142] For this purpose, the pilot enters his request for example via the keyboard 67. Alternatively, if the interaction module 25 includes a microphone, the pilot sends his request to the interaction device 25 by stating it vocally.

[0143] The selection module 55 is configured to select objects from the validated structured base 95 in response to the received request. Preferably, the selection module 55 is configured to select, from the validated structured base 95, the class(es), class instances and possibly values ​​of class instances concerned in a manner known per se. The selection module 55 is configured to develop a control instruction from the identified objects.

[0144] The transmission module 60 is configured to transmit the piloting instruction developed from the selected objects, to one of: the pilot of the aircraft 15, and an actuator of the aircraft 15.

[0145] The operation of the pilot assistance system 10 will now be described with reference to [Fig.6] illustrating a flowchart of a pilot assistance method according to the invention.

[0146] The pilot assistance method comprises an enrichment phase 100 and an exploitation phase 200.

[0147] Preferably, the enrichment phase 100 is implemented prior to the flight of the aircraft 15. The enrichment phase 100 is for example implemented in a design office.

[0148] The enrichment phase 100 is preferably implemented by the enrichment device 20 and comprises the implementation of an enrichment method 100.

[0149] The enrichment method 100 comprises a step 110 of obtaining the initial structured base 32.

[0150] The enrichment method 100 further comprises a step 120 of determining at least two indicators from the group of indicators consisting of: the frequency indicator, the interconnection indicator, the relevance indicator, and the precision indicator.

[0151] Preferably, step 120 of determining the indicators comprises the determination determination of at least three indicators from the group of indicators. More preferably, step 120 of determining the indicators comprises the determination of each indicator of the group of indicators.

[0152] If, during the determination step 120, one of the determined indicators is the frequency indicator, the determination step 120 then comprises the calculation of the Shannon entropy associated with each word of the initial structured base 32. The determination step 120 further comprises the evaluation of the percentage of words whose Shannon entropy is greater than the respective value (first predefined value). The determination step 120 further comprises the application of the statistical test to the calculated Shannon entropies to obtain the p-value and the similarity percentage. These actions implemented during the determination step 120 are preferably as explained previously with reference to the determination module 35.

[0153] The information quantity indicator then includes the percentage of words whose Shannon entropy is greater than the respective value (i.e. first predefined value), the p-value, and the percentage of similarity.

[0154] If, during the determination step 120, one of the determined indicators is the interconnection indicator, the determination step comprises calculating the proportion of the number of classes of the initial structured base 32 comprising an explanatory commentary of the content of the class. The determination step 120 further comprises calculating the number of semantic links connecting each class to a data / class instance in the initial structured base 32. The determination step 120 further comprises evaluating the number of classes connected to data / class instance by a number of semantic links less than a respective first value (third predefined value) or greater than a respective second value (fourth respective value). The determination step 120 further comprises calculating the modularity of the classes of the initial structured base 32.These actions implemented during the determination step 120 are preferably as explained previously with reference to the determination module 35.

[0155] The interconnection indicator comprises the number of classes of the initial structured base 32 comprising an explanatory commentary of the content of the class, the number of classes linked to data / class instances by a number of semantic links less than the respective first value (third predefined value) or greater than the respective second value (fourth predefined value), and the modularity value of the classes.

[0156] If, during the determination step 120, one of the determined indicators is the precision indicator, the determination step 120 comprises obtaining the dictionary. The determination step 120 further comprises comparing each word of each class of the initial structured base 32 to the words of the dictionary. The determination step 120 further comprises the calculation of the percentage of words of the initial structured base 32 appearing in the dictionary. The determination step 120 further comprises the calculation, among the words appearing in the dictionary, of the percentage of words having a single meaning in the dictionary. The determination step 120 further comprises the transformation of the initial structured base into a graph. The determination step 120 further comprises the calculation of the Shannon entropy associated with each daughter class of the graph. The determination step 120 further comprises the evaluation of the percentage of daughter classes whose Shannon entropy is greater than a respective value (fifth predefined value). These actions implemented during the determination step 120 are preferably as explained previously with reference to the determination module 35.

[0157] The precision indicator includes the percentage of words from the initial structured base 32 appearing in the dictionary, the percentage of words from the initial structured base 32 having a single meaning in the dictionary, and the percentage of daughter classes whose Shannon entropy is greater than the respective value (fifth predefined value).

[0158] If, during the determination step 120, one of the determined indicators is the relevance indicator, the determination step 120 comprises obtaining the trained artificial intelligence model to: receive a scenario, determine a logical rule combining the values ​​of the set of values ​​resulting in the action of said scenario, and to quantify the importance of each value in the action taken. The artificial intelligence model is preferably an LTN model. The determination step 120 further comprises applying said artificial intelligence model to each scenario of the initial structured base 32, to obtain, for each value of each set of values, an importance quantifier in the action of said scenario. Preferably, each importance quantifier is associated with a respective knowledge from the scenario, i.e. a class instance and corresponding value pair.The determination step 120 further comprises the calculation of the proportion of values ​​of the sets of values ​​for which the importance quantifier is lower than a respective threshold, i.e. the sixth predefined threshold. These actions implemented during the determination step 120 are preferably as explained previously with reference to the determination module 35.

[0159] The relevance indicator then comprises said proportion of values ​​of the sets of values ​​for which the importance quantifier is lower than the respective threshold, i.e. the sixth predefined threshold.

[0160] The enrichment method 100 further comprises a comparison step 130, during which each determined indicator is compared with at least one threshold respective, as explained previously.

[0161] If the result of the comparison is negative, i.e. if the comparison of at least one indicator with the respective threshold returns a negative result, the enrichment method comprises a step 140 of sending a command to enrich the initial structured base 32 to form the enriched structured database, called enriched structured database, comprising a greater number of objects than the initial structured base 32.

[0162] The steps of obtaining 110, determining 120, comparing 130, and where appropriate sending 140 are preferably repeated as long as the result of the comparison is negative. The structured base enriched by an iteration being the initial structured base 32 of the following iteration. The structured database obtained at the end of the iterations, that is to say after the first iteration for which the result of the comparison is positive, is called the validated structured base 95.

[0163] Preferably, when the result of the comparison is positive, then the initial structured base 32 is validated and becomes the validated structured base. The enrichment method 100 comprises an import step 150 during which the sending module 45 sends the validated structured base 95 to the interaction device 25.

[0164] The enrichment phase 100 then preferably ends.

[0165] The operating phase 200 is for example implemented by the device of interaction 25.

[0166] Preferably, the operating phase 200 is implemented during a flight of the aircraft 15, the validated structured base 95 being for example stored in the second memory 80.

[0167] At a moment in the flight of the aircraft 15, the pilot of the aircraft 15 wishes to obtain information from the validated structured base 95.

[0168] The operating phase 200 comprises a step 210 of receiving the request from the pilot of the aircraft 15, for example via the keyboard 67 or any other means allowing the pilot to send a request to the interaction device 25.

[0169] The operating phase 200 further comprises a step 220 of selecting objects, from the validated structured base, in response to the request received, preferably as described previously.

[0170] Optionally, during the selection phase 220, the interaction device 25 develops the piloting instruction in response to the request from the selected objects.

[0171] The operating phase 200 further comprises a step 230 of transmitting the piloting instruction developed from the selected objects, to one of: the pilot of the aircraft 15 and an actuator of the aircraft 15.

[0172] For example, if the piloting instruction is issued to the pilot, it is displayed on the display screen 65. The pilot then decides whether or not to implement it. the piloting instruction. Conversely, if the piloting instruction is issued to an actuator of the aircraft 15, said instruction takes the form of a piloting command which is directly implemented by said actuator. The actuator is for example an engine of the aircraft 15, a control surface or any other type of actuator.

[0173] Preferably, the operating phase 200 is repeated based on a new request from the pilot.

[0174] The present invention therefore makes it possible to guarantee that the structured database used as a piloting aid is suitable for use by supervising its enrichment, i.e. its construction, by different indicators, also called metrics.

Claims

Claims

1. Method (100) for enriching a structured database intended to assist the decision-making of a pilot of an aircraft (15), the method being implemented by an electronic enrichment device (20) and comprising the following steps: - obtaining (110) an initial structured database (32), called initial structured database (32), comprising a plurality of objects among which: • a plurality of classes, • a plurality of data, and • semantic links associating each data with at least one class, each data and / or each class comprising one or more words, - determination (120) of at least two indicators from the group of indicators consisting of: • a frequency indicator quantifying the frequency of appearance of each word in the initial structured base (32), • an interconnection indicator representative of a distribution of semantic links in the initial structured base (32), • a relevance indicator quantifying the usefulness of the data in the initial structured database (32), • a precision indicator quantifying the polysemy of each word of the initial structured base (32), - comparison (130) of each determined indicator with at least one respective threshold, - if the result of the comparison is negative, sending (140) a command to enrich the initial structured base (32) to form an enriched structured database, called an enriched structured base, comprising a greater number of objects than the initial structured base (32).

2. The method (100) of claim 1, wherein the step (120) of determining the indicators comprises determining at least three indicators from among the group of indicators, the step (120) of determining the indicators preferably comprising the determination of each indicator of the group of indicators.

3. The method (100) of claim 1 or 2, wherein one of the indicators determined during the determining step (120) is the frequency indicator, the determining step (120) comprising: - calculating the Shannon entropy associated with each word of the initial structured base (32), - evaluating the percentage of words whose Shannon entropy is greater than a respective value, - applying a statistical test to the calculated Shannon entropies to obtain a p-value and a similarity percentage, the information quantity indicator comprising: - the percentage of words whose Shannon entropy is greater than the predefined value, - the p-value, and - the similarity percentage.

4. Method (100) according to any one of the preceding claims, wherein one of the indicators determined during the determining step (120) is the interconnection indicator, the determining step (120) comprising: - calculating the proportion of the number of classes of the initial structured base (32) comprising an explanatory commentary of the content of the class, - calculating the number of semantic links connecting each class to a data item in the initial structured base (32), - evaluating the number of classes connected to data by a number of semantic links less than a respective first value or greater than a respective second value, - calculating a modularity value of the classes of the initial structured base (32), the interconnection indicator including: - the number of classes in the initial structured base (32) including an explanatory commentary on the content of the class, - the number of classes linked to data by a number of semantic links less than the first respective value or greater than the second respective value, and - the modularity value of the classes.

5. Method (100) according to any one of the preceding claims, wherein one of the indicators determined during the determining step (120) is the precision indicator, the determining step (120) comprising: - obtaining a dictionary comprising a plurality of words, and, for each word, at least one meaning, - the comparison of each word of each class of the initial structured base (32) with the words of the dictionary, - the calculation of a percentage of words from the initial structured base (32) appearing in the dictionary, - among the words appearing in the dictionary, the calculation of a percentage of words having a single meaning in the dictionary, - the transformation of the initial structured base (32) into a graph, the graph comprising a central class, called the mother class, linked to a plurality of classes, called daughter classes, - the calculation of the Shannon entropy associated with each daughter class, - the evaluation of a percentage of daughter classes whose Shannon entropy is greater than a respective value, the precision indicator including: - the percentage of words from the initial structured base (32) appearing in the dictionary, - percentage of words in the initial structured base (32) having a single meaning in the dictionary, - the percentage of daughter classes whose Shannon entropy is greater than the respective value.

6. Method (100) according to any one of the preceding claims, in which the initial structured base (32) comprises a plurality of scenarios each comprising a set of values ​​and an action taken by the pilot of the aircraft (15) in the presence of this set of values, each value of the set of values ​​being a numerical value of a data item of the initial structured base (32), one of the indicators determined during the determination step (120) being the relevance indicator, the determination step (120) comprising: - obtaining a trained artificial intelligence model to receive a scenario and determine a logical rule combining the values ​​of the set of values ​​resulting in the action of said scenario, and to quantify the importance of each value in the action taken, - applying said artificial intelligence model to each scenario of the initial structured base (32), to obtain, for each value of each set of values,an importance quantifier in the action of said scenario, - calculation of the proportion of values ​​of the sets of values ​​for which the importance quantifier is lower than a respective threshold, the relevance indicator comprising: - said proportion of values ​​of the sets of values ​​for which the importance quantifier is lower than the respective threshold.,

7. Method (100) according to any one of the preceding claims, in which, following the step of sending an enrichment command, the steps of obtaining (110), determining (120), comparing (130) and, where appropriate, sending (140) of the method are repeated as long as the result of the comparison is negative, the enriched structured base of an iteration being the initial structured base (32) of the following iteration, the structured database obtained at the end of the iterations being called the validated structured base (95).

8. Method (100) according to the preceding claim, wherein the method includes, if the result of the comparison is positive, a step (150) of importing the validated structured base (95), into an electronic device (25) for interaction with the pilot of the aircraft (15), for its use during an avionics mission.

9. Method for assisting in piloting an aircraft (15) from an initial structured database (32), called initial structured database (32), the method comprising an enrichment phase (100) and an exploitation phase (200), the enrichment phase (100) comprising an enrichment of the initial structured database (32) by applying an enrichment method (100) according to any one of the preceding claims, the exploitation phase (200) being implemented by an electronic device (25) for interaction with the pilot of the aircraft (15), the electronic interaction device (25) comprising the structured database resulting from the enrichment phase (100), called validated structured database (95), and comprising the following steps: - reception (210) of a request from the pilot of the aircraft (15), - selection (220) of objects, the validated structured base (95), in response to the received request, - transmission (230),of a piloting instruction developed from the selected objects, intended for one of: the pilot of the aircraft (15) and an actuator of the aircraft (15).,

10. A computer program product, comprising software instructions which, when executed by a computer, implement a method according to any one of the preceding claims.

11. Electronic device (20) for enriching a structured database intended to assist the decision-making of a pilot of an aircraft (15), the electronic enrichment device (20) comprising: - an obtaining module (30) configured to obtain an initial structured database (32), called initial structured database (32), comprising a plurality of objects among which: • a plurality of classes, • a plurality of data, and • semantic links associating each data item with at least one class, each data and / or each class comprising one or more words, - a determination module (35) configured to determine at least two indicators from the group of indicators constituted • a frequency indicator quantifying the frequency of appearance of each word in the initial structured base (32), • an interconnection indicator representative of a distribution of semantic links in the initial structured base (32), • a relevance indicator quantifying the usefulness of the data in the initial structured database (32), • a precision indicator quantifying the polysemy of each word of the initial structured base (32), - a comparison module (40) configured to compare each determined indicator with at least one respective threshold, - a sending module (45) configured to, if the result of the comparison is negative, send a command to enrich the initial structured base (32) to form an enriched structured database, called an enriched structured database, comprising a greater number of objects than the initial structured base (32).

Images