Device for estimating a physical quantity with three models, and associated system and method
The device addresses data drift and resource constraints in estimation models by using multiple modules to assess prediction consistency and issue alerts, maintaining model reliability and compliance with predefined domains.
Patent Information
- Application Number
- PCT/EP2026/063772
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-12
- Filing Date
- 2026-05-12
- Publication Date
- 2026-11-19
AI Technical Summary
Existing data-driven estimation models in critical systems face issues with data drift, concept drift, and outliers, leading to performance degradation and catastrophic forgetting due to frequent updates under resource constraints, which are unacceptable for secure operation.
A device with multiple estimation modules and an analysis module that assesses the consistency of predictions from different model versions, using confidence criteria and temporal persistence to determine reliable estimates, and issues alerts for inconsistencies.
Ensures secure and reliable estimation under resource constraints by minimizing development costs and maintaining model performance through incremental learning, while detecting drifts and ensuring compliance with predefined usage domains.
Smart Images

Figure EP2026063772_19112026_PF_FP_ABST
Abstract
Description
[0001] Device for estimating a physical quantity with three models, system and associated method
[0002] The present invention relates to a device for estimating a physical quantity. It also relates to an estimation system comprising such an estimation device and to an associated estimation method.
[0003] Artificial intelligence technologies are increasingly being used to develop data-driven machine learning models in critical systems.
[0004] These models are specifically estimation models designed to provide an estimated value of the physical quantity over a predefined domain of use.
[0005] The predefined usage domain is, by definition, the set of operating conditions in which the estimation model is specifically designed to operate, including, but not limited to, environmental, geographical, and temporal restrictions.
[0006] The predefined use domain is often referred to by the abbreviation ODD, which refers to the English term "operational design domain".
[0007] The estimation model is thus trained on data reflecting the domain of use, assuming that the data that the estimation model will process in operation will always be part of the domain of use.
[0008] However, it frequently happens that the situations (i.e. the data) encountered in practice by the estimation model are different from those used to train it.
[0009] This leads the estimation model to be confronted with the change over time of the distribution of data (more commonly referred to by the English term "data drift"), the change in the relationship between inputs and output(s) of the estimation model (more commonly referred to by the English term "concept drift") or even with data that is not included in the predefined domain of use (abnormal data or "outliers" in reference to the corresponding English term).
[0010] The performance of the estimation model is impacted, which is addressed by performing frequent updates to the estimation model, for example, quarterly, monthly, weekly, daily, or after each operational mission using the estimation model. These updates must be performed even with limited data access. Incremental learning and qualification techniques are known to be used for these updates. This involves retraining on small increments of datasets containing a selection of new and old data. This avoids the need for a complete but excessively large dataset to perform the estimation model update.
[0011] However, with frequent and rapid updates to an estimation model, it is common for the new version to contain problems (more often called bugs) due to insufficient design, verification, and validation activities for the estimation model, resulting from constraints imposed by incremental development and qualification. Examples of such constraints include a lack of time, insufficient new training and test data, or a lack of access to such data.
[0012] In particular, catastrophic forgetting problems can occur. This refers to a situation in which the estimation model loses previously acquired knowledge during incremental updates. This is unacceptable for a critical system that requires ensuring the secure maintenance of certain functionalities.
[0013] There is therefore a need for a device for estimating a physical quantity that allows an estimation model to be operated incrementally under constraints (reduced time, reduced data volume, reduced data access) with increased security, particularly for use of the estimation device in a critical system.
[0014] For this purpose, the description concerns a device for estimating a physical quantity, the estimation device comprising:
[0015] - a first estimation module, the first estimation module comprising:
[0016] - a first receiving unit designed to receive data acquired by a detection device, and
[0017] - a first computing unit capable of applying a first estimation model to the data received by the first receiving unit in order to obtain a first prediction of the physical quantity to be estimated,
[0018] - a second estimation module, the second estimation module comprising: - a second receiving unit capable of receiving data acquired by a detection device, and
[0019] - a second computing unit to apply a second estimation model to the data received by the second receiving unit to obtain a second prediction of the physical quantity to be estimated, the second estimation model being an update of the first estimation model, and - an analysis module, the analysis module being to analyze the consistency of values from the first prediction and the second prediction and to deduce from the consistency analysis an estimate of the physical quantity to be estimated for the estimation device.
[0020] Depending on other advantageous aspects, the estimation device includes one or more of the following characteristics, taken individually or in all technically possible combinations:
[0021] - If the values from the first prediction and the second prediction are inconsistent, the analysis module is capable of deducing that the value of the estimate of the physical quantity is the value from the first prediction. - If the values from the first prediction and the second prediction are consistent, the analysis module is capable of deducing that the value of the estimate of the physical quantity is the value from the second prediction.- the first estimation module includes a first unit for determining a first value representing the confidence in the prediction of the first calculation unit and the second estimation module includes a second unit for determining a second value representing the confidence in the prediction of the second calculation unit, the values from the first prediction and the second prediction whose consistency the analysis module verifies being obtained from the first prediction, the second prediction, the first value representing the confidence in the prediction of the first calculation unit and the second value representing the confidence in the prediction of the second calculation unit.
[0022] - The analysis module also includes:
[0023] - a first consolidation unit capable of obtaining a first consolidated prediction from the first prediction (PN) based on a first confidence criterion applied to the first representative value of the confidence in the prediction of the first calculation unit, and
[0024] - a second consolidation unit capable of obtaining a second consolidated prediction from the second prediction (PN) based on a second confidence criterion applied to the second value representing the confidence in the prediction of the second calculation unit,
[0025] the values from the first prediction and the second prediction whose consistency the analysis module checks are respectively the first consolidated prediction and the second consolidated prediction.- the first confidence criterion is a comparison of the first value representing the confidence in the prediction of the first calculation unit to a first predefined threshold and the second confidence criterion is a comparison of the second value representing the confidence in the prediction of the second calculation unit to a second predefined threshold.
[0026] - the estimation device also includes a warning module, the warning module being designed to issue an alert in the presence of multiple inconsistencies between the values from the first prediction and the second prediction whose consistency the analysis module checks.
[0027] - the warning module is designed to accumulate the results of consistency analyses and to issue the alert when the number of inconsistencies for a number of successive consistency analysis results performed by the analysis module is greater than or equal to a threshold.
[0028] - the data acquired by the detection device are images or audio signals.
[0029] - the physical quantity characterizes the environment of the detection device, the physical quantity being in particular the temperature or the presence and / or identification of an object in this environment.
[0030] - the physical quantity is a quantity relating to a means of transport including the detection device, in particular a quantity giving the trajectory of the means of transport.
[0031] - the estimation device includes a sending module to send the estimate of the physical quantity to be estimated to a control module, the control module being able to control a physical device according to the estimate of the physical quantity to be estimated, in particular to modify the operation of the physical device.
[0032] - the physical device is a means of transport, in particular an aircraft.
[0033] The description also relates to a device for estimating a physical quantity, the estimation device comprising:
[0034] - a first estimation module, the first estimation module comprising:
[0035] - a first receiving unit designed to receive data acquired by a detection device, and
[0036] - a first computing unit capable of applying a first estimation model to the data received by the first receiving unit in order to obtain a first prediction of the physical quantity to be estimated,
[0037] - a second estimation module, the second estimation module comprising: - a second receiving unit capable of receiving data acquired by a detection device, and
[0038] - a second computing unit capable of applying a second estimation model to the data received by the second receiving unit to obtain a second prediction of the physical quantity to be estimated, the second estimation model being a first update of the first estimation model, - a third estimation module, the third estimation module comprising: - a third receiving unit capable of receiving data acquired by a detection device, and
[0039] - a third computing unit capable of applying a third estimation model to the data received by the third receiving unit (66) to obtain a third prediction of the physical quantity to be estimated, the third estimation model being a second update of the first estimation model, - an analysis module, the analysis module being capable of receiving each prediction and obtaining an estimate of the physical quantity to be estimated for the estimation device from the predictions received, and
[0040] - an inhibition module, the inhibition module being capable of inhibiting the reception by the analysis module of the prediction of at least one estimation module among the first estimation module, the second estimation module and the third estimation module.
[0041] Depending on other advantageous aspects, the estimation device includes one or more of the following characteristics, taken individually or in all technically possible combinations:
[0042] - the inhibition module includes an inhibition unit for each estimation module, the inhibition module of which is specific to inhibit the reception of the prediction by the analysis module.
[0043] - each inhibition unit has at least one switch having an open position in which the reception of the prediction by the analysis module is inhibited and a closed position in which the reception of the prediction by the analysis module is not inhibited.
[0044] - the inhibition module is specific to inhibiting only the reception of the prediction from the second estimation module and the third estimation module.
[0045] - the inhibition module has a first state in which only the reception of the prediction of the second estimation module is inhibited and a second state in which only the reception of the prediction of the third estimation module is inhibited, the estimation device also comprising a control module, the control module being suitable for controlling the state of the inhibition module.
[0046] - The control module is designed to trigger the implementation of a loading operation for the third estimation model on the third computing unit and to control the inhibition module so that the inhibition module is in the first state when the loading operation has not been implemented and in the second state when the loading operation has taken place
[0047] - the analysis module is designed to analyze the consistency of values from received predictions and to deduce from the consistency analysis an estimate of the physical quantity to be estimated for the estimation device.
[0048] - the predictions received by the analysis module are the first prediction and one prediction from the second prediction and the third prediction.
[0049] - the data acquired by the detection device are images or audio signals.
[0050] - the physical quantity characterizes the environment of the detection device, the physical quantity being in particular the temperature or the presence and / or identification of an object in this environment.
[0051] - the physical quantity is a quantity relating to a means of transport including the detection device, in particular a quantity giving the trajectory of the means of transport.
[0052] - the estimation device includes a sending module to send the estimate of the physical quantity to be estimated to a control module, the control module being able to control a physical device according to the estimate of the physical quantity to be estimated, in particular to modify the operation of the physical device.
[0053] - the physical device is a means of transport, in particular an aircraft.
[0054] The description also concerns a system for estimating a physical quantity, the estimation system comprising:
[0055] - several devices for estimating the physical quantity, each estimation device comprising an analysis module, each analysis module being suitable for analyzing the consistency of values from predictions obtained by applying at least two estimation models to data acquired by a detection device, an estimation model being an update of another estimation model, to obtain a value representative of the consistency between the predictions, and - a coordination system suitable for applying an aggregation function on the values representative of the consistency between the predictions to obtain an aggregated value and suitable for deducing an estimate of the physical quantity to be estimated from the aggregated value.
[0056] Depending on other advantageous aspects, the estimation system includes one or more of the following characteristics, taken individually or in all technically possible combinations:
[0057] - each device for estimating the physical quantity includes a communication system and the coordination system includes a communication unit, the communication unit of the coordination system being suitable for communicating the estimate deduced to each communication system of an estimation device, the estimate of the physical quantity by each estimation device being the estimate deduced by the coordination system.
[0058] - the aggregation function is a majority vote.
[0059] - for each analysis module, the values from each prediction are obtained from the value of the prediction and a value representing the confidence in the prediction;
[0060] - each estimation device includes, for each prediction, a consolidation unit capable of obtaining a consolidated prediction from the prediction according to a respective confidence criterion applied to the representative value of confidence in the prediction, the values from each prediction whose consistency the analysis module verifies being respectively the consolidated prediction.
[0061] - each estimation device also includes a warning module designed to obtain a value representative of the temporal evolution of the analysis results, the consistency, the values representative of the consistency between the predictions being the values representative of the temporal evolution.
[0062] - each estimation device also includes:
[0063] - a third estimation module, the third estimation module comprising:
[0064] - a third receiving unit specifically designed to receive data acquired by a detection device, and
[0065] - a third computing unit capable of applying a third estimation model to the data received by the third receiving unit to obtain a third prediction of the physical quantity to be estimated, the third estimation model being a second update of the first estimation model, and - an inhibition module, the inhibition module being capable of inhibiting the reception by the analysis module of the prediction of at least one estimation module among the first estimation module, the second estimation module and the third estimation module;
[0066] - the coordination system is part of one of the estimation devices;
[0067] - each estimation device is specific to receiving data acquired from a detection device.
[0068] - the data acquired by each detection device are images or audio signals.
[0069] - the physical quantity characterizes the environment of each detection device, the physical quantity being in particular the temperature or the presence and / or identification of an object in this environment.
[0070] - the physical quantity is a quantity relative to a means of transport including each detection device, in particular a quantity giving the trajectory of the means of transport.
[0071] - the estimation device includes a sending module to send the estimate of the physical quantity to be estimated to a control module, the control module being able to control a physical device according to the estimate of the physical quantity to be estimated, in particular to modify the operation of the physical device.
[0072] - the physical device is a means of transport, in particular an aircraft.
[0073] The description also relates to a device for estimating a physical quantity, the estimation device comprising:
[0074] - a first estimation module, the first estimation module comprising:
[0075] - a first receiving unit designed to receive data acquired by a detection device, and
[0076] - a first computing unit capable of applying a first estimation model to the data received by the first receiving unit in order to obtain a first prediction of the physical quantity to be estimated,
[0077] - a second estimation module, the second estimation module comprising: - a second receiving unit capable of receiving data acquired by a detection device, and
[0078] - a second computing unit to apply a second estimation model to the data received by the second receiving unit to obtain a second prediction of the physical quantity to be estimated, the second estimation model being an update of the first estimation model, and - an analysis module, the analysis module being to analyze the consistency of values from the first prediction and the second prediction and to deduce from the consistency analysis an estimate of the physical quantity to be estimated for the estimation device.
[0079] The description also relates to a device for estimating a physical quantity, the estimating device comprising:
[0080] - a first estimation module, the first estimation module comprising:
[0081] - a first receiving unit designed to receive data acquired by a detection device, and
[0082] - a first computing unit capable of applying a first estimation model to the data received by the first receiving unit in order to obtain a first prediction of the physical quantity to be estimated,
[0083] - a second estimation module, the second estimation module comprising: - a second receiving unit capable of receiving data acquired by a detection device, and
[0084] - a second computing unit capable of applying a second estimation model to the data received by the second receiving unit to obtain a second prediction of the physical quantity to be estimated, the second estimation model being a first update of the first estimation model, - a third estimation module, the third estimation module comprising: - a third receiving unit capable of receiving data acquired by a detection device, and
[0085] - a third computing unit capable of applying a third estimation model to the data received by the third receiving unit to obtain a third prediction of the physical quantity to be estimated, the third estimation model being a second update of the first estimation model,
[0086] - an analysis module, the analysis module being designed to receive each prediction and to obtain an estimate of the physical quantity to be estimated for the estimation device based on the predictions received, and
[0087] - an inhibition module, the inhibition module being capable of inhibiting the reception by the analysis module of the prediction of at least one estimation module among the first estimation module, the second estimation module and the third estimation module.
[0088] - The data acquired by the detection device are images or audio signals. - The physical quantity characterizes the environment of the detection device, the physical quantity being in particular the temperature or the presence and / or identification of an object in this environment.
[0089] - the physical quantity is a quantity relating to a means of transport including the detection device, in particular a quantity giving the trajectory of the means of transport.
[0090] - the estimation device includes a sending module to send the estimate of the physical quantity to be estimated to a control module, the control module being able to control a physical device according to the estimate of the physical quantity to be estimated, in particular to modify the operation of the physical device.
[0091] - the physical device is a means of transport, in particular an aircraft.
[0092] The description also relates to a system for estimating a physical quantity, the estimation system comprising:
[0093] - several devices for estimating the physical quantity, each estimation device comprising an analysis module, each analysis module being specific to analyzing the consistency of values resulting from predictions obtained by applying at least two estimation models to data acquired by a detection device, an estimation model being an update of another estimation model, to obtain a value representative of the consistency between the predictions, and
[0094] - a coordination system capable of applying an aggregation function to the representative values of the consistency between the predictions to obtain an aggregate value and capable of deducing an estimate of the physical quantity to be estimated from the aggregate value.
[0095] The description also relates to a method for estimating a physical quantity, the estimation method being implemented by computer and comprising the steps of:
[0096] - reception of data acquired by a detection device,
[0097] - application of a first estimation model to the received data to obtain a first prediction of the physical quantity to be estimated,
[0098] - application of a second estimation model on the received data to obtain a second prediction of the physical quantity to be estimated, the second estimation model being an update of the first estimation model, and - analysis of the consistency of values from the first prediction and the second prediction, and - deduction from the consistency analysis of an estimate of the physical quantity to be estimated.
[0099] The description also relates to a method for estimating a physical quantity comprising the steps of:
[0100] - reception of data acquired by a detection device,
[0101] - application of a first estimation model to the received data to obtain a first prediction of the physical quantity to be estimated,
[0102] - application of a second estimation model on the received data to obtain a second prediction of the physical quantity to be estimated, the second estimation model being a first update of the first estimation model, - application of a third estimation model on the received data to obtain a third prediction of the physical quantity to be estimated, the third estimation model being a second update of the first estimation model, - reception of each prediction, the reception of the prediction of at least one estimation module among the first estimation module, the second estimation module and the third estimation module being inhibited by an inhibition module, and - obtaining an estimate of the physical quantity to be estimated for the estimation device from the received predictions.
[0103] The description also relates to a method for estimating a physical quantity, the estimation method being implemented by computer and comprising the steps of:
[0104] - obtaining representative values of the consistency between the predictions of several devices for estimating the physical quantity by analyzing the consistency of values from predictions obtained by applying at least two estimation models to data acquired by a detection device, an estimation model being an update of another estimation model,
[0105] - applying an aggregation function to the values representing the consistency between the predictions to obtain an aggregated value, and
[0106] - deduction of an estimate of the physical quantity to be estimated from the aggregated value.
[0107] In this description, the expression "specific to" means interchangeably "suited for", "adapted to" or "configured for".
[0108] The invention will become clearer upon reading the following description, given solely by way of non-limiting example, and made with reference to the drawings in which: - Figure 1 is a schematic representation of an example of a system for estimating a physical quantity comprising a detection device and an estimation device,
[0109] - Figure 2 is a representation of the operation of the units of the estimation device shown in Figure 1,
[0110] - Figure 3 is a flowchart of an example of the implementation of an estimation process using the estimation system of Figure 1,
[0111] - Figure 4 is a schematic representation of another example of a system for estimating a physical quantity, comprising a detection device and an estimation device.
[0112] - Figure 5 is a representation of an example of the operation of the units of the estimation device shown in Figure 4,
[0113] - Figure 6 is a flowchart of an example of the implementation of an estimation process using the estimation system of Figure 4,
[0114] - Figure 7 is a schematic representation of an example of a system for estimating a physical quantity, including a coordination system; - Figure 8 is a flowchart of an example of the implementation of an estimation process using the estimation system of Figure 7.
[0115] - Figure 9 is a schematic representation of another example of a system for estimating a physical quantity including a coordination system, and - Figure 10 is a schematic representation of yet another example of a system for estimating a physical quantity including a coordination system.
[0116] A 10 estimation system is illustrated in Figure 1.
[0117] The estimation system 10 is suitable for estimating a physical quantity.
[0118] The physical quantity can be of any type, with the estimation system 10 serving as the measurement system.
[0119] According to a particular example, the estimation system 10 can be seen as an environmental observation system.
[0120] The estimation system 10 includes a detection device 12 and an estimation device 14.
[0121] The detection device 12 is designed to acquire data.
[0122] As such, the detection device 12 includes one or more sensors designed to collect data.
[0123] The data collected are therefore measurements. According to the example described, the detection device 12 is a camera, the camera providing images.
[0124] The data acquired by the detection device 12 are therefore images.
[0125] Alternatively or in addition, the detection device 12 may include any type of sensors, and in particular sensors designed to collect audio, video, time series, tabular data or any other type of signal.
[0126] The estimation device 14 is a device for estimating a physical quantity. The estimation device 14 is designed to process data acquired by the detection device 12 to obtain an estimated value, which corresponds to the value of the physical quantity being sought.
[0127] These processes involve applying an estimation model to data acquired by the detection device 12
[0128] The estimation model is a model derived from artificial intelligence, specifically a model obtained through machine learning.
[0129] As a non-limiting example, the estimation model is a neural network. The estimation model is designed to provide an estimated value of the physical quantity within the predefined ODD usage domain.
[0130] The predefined usage domain is, by definition, the set of operating conditions in which the estimation model is specifically designed to operate, including, but not limited to, environmental, geographical, and temporal restrictions.
[0131] In the present context, the estimation model is a model intended to evolve over time.
[0132] This change could have several causes.
[0133] For example, the performance of the estimation model is insufficient, the estimation model must perform a different or additional task, or the evaluation criteria of the estimation model change (in a particular case due to regulatory change).
[0134] Such an evolution of the model is usually obtained by retraining on a new data increment and the associated constrained incremental qualification activities (i.e., incompletely).
[0135] This corresponds to incremental learning, implying that the estimation model will present several versions.
[0136] As will be shown later, the estimation device 14 allows here the use of two successive versions of the estimation model, a first version before update and a second version after update, having been qualified under constraint. To illustrate this well, a specific estimation model is considered which is a classifier.
[0137] The estimation device 14 comprises a first estimation module 16, a second estimation module 18, an analysis module 20 and a warning module 22.
[0138] The first estimation module 16 is designed to obtain a first prediction and a first representative value of the confidence in this prediction from data acquired by the detection device 12.
[0139] In Figure 2, the first prediction is noted PN while the first representative confidence value is noted CN.
[0140] The first estimation module 16 comprises a first reception unit 24, a first calculation unit 26 and a first determination unit 28.
[0141] The first receiving unit 24 is designed to receive the data acquired by the detection device 12.
[0142] The first computing unit 26 is used to apply a first estimation model to the data received by the first receiving unit 24 to obtain the first prediction P N .
[0143] The first estimation model is assumed here to be version N, where N is an integer greater than or equal to 1.
[0144] Therefore, the first estimation model is noted as MEN and thus corresponds here to the estimation model before the update (here the Nth).
[0145] To simplify the notation, the first estimation model will be noted simply as the MEN estimation model.
[0146] An estimation model is considered that would be able to classify the input data according to three output classes: a first class, a second class, and an anomaly class.
[0147] The anomaly class corresponds to abnormal data.
[0148] In such a case, it implies that the estimation model, taking an abnormal data point as input, must, under normal operating conditions, interpret such data as an anomaly. Typically, the estimation model will output an error code or classify the abnormal data point as, for example, unknown data.
[0149] More specifically, it is assumed that the ME estimation model N is a recognition model specifically designed to identify two elements in the observed environment, referred to hereafter as element A (first class) and element B (second class). Any other element is not recognized by the MEN (anomaly class) estimation model.
[0150] The first prediction PN is therefore element A, element B or anomaly. The first unit of determination 28 is suitable for determining a first value representing the confidence in the prediction of the first calculation unit 26.
[0151] For example, the first unit of determination 28 applies a function specific to give confidence knowing the estimation model MEN and the first prediction PN.
[0152] According to another example, the first unit of determination 28 uses the MEN estimation model. In such a case, the first calculation unit 26 and the first unit of determination 28 can be confused, the ME estimation model N being then able to provide the first prediction PN and the first representative value of the confidence CN in the prediction of the first unit of calculation 26.
[0153] The first representative value of confidence in the prediction of the first calculation unit 26 is notably expressed in the form of a probability, without this example being limiting.
[0154] The second estimation module 18 is designed to obtain a second prediction and a second value representing the confidence in this prediction from data acquired by the detection device 12.
[0155] In Figure 2, the second prediction is noted PN+I while the second representative confidence value is noted CN+I.
[0156] The second estimation module 18 comprises a second receiving unit 30, a second calculation unit 32 and a second determination unit 34.
[0157] The second receiving unit 30 has the same role as the first receiving unit 24.
[0158] In one embodiment, the first receiving unit 24 and the second receiving unit 30 are combined.
[0159] The second computing unit 32 is designed to apply a second estimation model to the data received by the second receiving unit 30 in order to obtain the second prediction P N +i.
[0160] The second estimation model corresponds to the update of the first MEN estimation model, here by implementing incremental learning.
[0161] Put another way, the second estimation model corresponds to version (or increment) N+1 of the estimation model.
[0162] Therefore, the second estimation model is denoted MEN+I and will thus be referred to hereafter as the ME estimation model N +I to simplify the notation.
[0163] In the example described, it is assumed that the update aims to obtain a MEN+I estimation model capable of identifying two elements in the observed environment, hereafter referred to as element C (first class) and element B (second class). Any other element is not recognized by the MEN+I estimation model (anomaly class). Element C is different from element A, which the MEN estimation model is specifically designed to recognize.
[0164] The second unit of determination 34 is suitable for determining a second value representing the confidence CN+I in the prediction of the second unit of calculation 32.
[0165] Remarks similar to those concerning the first unit of determination 28 are also valid here and are not repeated for the sake of brevity.
[0166] The analysis module 20 is designed to receive the outputs from the estimation modules 16 and 18 and to select the best estimate of the physical quantity to be estimated.
[0167] According to the example described, the analysis module 20 is designed to analyze the outputs of the estimation modules 16 and 18 to obtain the best prediction as output.
[0168] More specifically, according to the example illustrated to obtain the best prediction, the analysis module 20 takes as input the different predictions PN and PN+I and the confidence scores CN and CN+I and outputs a predicted value for the warning device 22. This predicted value is specified later.
[0169] As shown in Figure 2, the analysis module 20 comprises a first consolidation unit 36, a second consolidation unit 40 and a verification unit 42.
[0170] The first consolidation unit 36 is suitable for consolidating the first prediction PN of the first estimation module 16.
[0171] For this, the first consolidation unit 36 is suitable for obtaining such a first consolidated prediction PCN as a function of a confidence criterion dependent on the first confidence value CN.
[0172] According to the example described, the performance criterion is a comparison with a threshold. The first consolidation unit 36 compares the first confidence value CN to a predefined threshold SN.
[0173] The predefined SN threshold is, for example, provided with the MEN estimation model; this value is determined during the training of the ME estimation model. N based on its performance on a test dataset.
[0174] When the first confidence value CN is greater than or equal to the threshold, the first consolidation unit 36 outputs a first consolidated prediction PCN which is equal to the first prediction PN of the first estimation module 16.
[0175] Conversely, when the first confidence value CN is strictly less than the threshold, the first consolidated prediction PCN is the anomaly class (regardless of the value of the first prediction PN). The second consolidation unit 40 functions similarly to the first consolidation unit 36 for the second prediction P N +i.
[0176] Thus, when the second confidence value CN+I is greater than or equal to a predefined threshold SN+I, the second consolidation unit 40 outputs a second consolidated prediction PCN+I which is equal to the second prediction PN+I of the second estimation module 18.
[0177] Conversely, when the second confidence value CN+I is strictly less than the predefined threshold SN+I, the second consolidated prediction PCN+I is the anomaly class (regardless of the value of the second prediction PN+I).
[0178] The verification unit 42 is a unit for verifying the consistency between the consolidated PCN and PCN+I predictions of the two consolidation units 36 and 40.
[0179] The verification unit 42 is suitable for comparing the values of the two consolidation units 36 and 40 and for deducing a verified prediction PV based on the result of the comparison.
[0180] According to one embodiment, the verification unit 42 uses a truth table to perform the comparison and deduction actions.
[0181] The truth table is a function that takes as input the value of the first consolidated prediction PCN and the value of the second consolidated prediction PCN+I and gives as output the verified prediction PV.
[0182] Table 1 below gives the truth table used by the verification unit 42 for the example described with the two estimation models MEN and MEN+I.
[0183] First Second
[0184] Verified Prediction Result: Prediction
[0185] comparison of consolidated PV PCN consolidated PCN+I
[0186] MEN Model Output Element A Element C Disagreement
[0187] (Element A) MEN Model Output Element A Element B Disagreement
[0188] (element A)
[0189] Not recognized. MEN+I model output. Element A. Agreement.
[0190] (neither C nor B) (neither C nor B)
[0191] MEN Model Output Element B Element C Disagreement
[0192] (element B)
[0193]
[0194] First Second
[0195] Verified Prediction Result: Prediction
[0196] comparison of consolidated PV PCN consolidated PCN+I
[0197] MEN+I Model Output Element B Element B Accord
[0198] (element B)
[0199] Not recognized. MEN model output. Element B. Disagreement.
[0200] (neither C nor B) (element B)
[0201] Not recognized. MEN+I model output. Element C. Accord.
[0202] (neither A nor B) (element C)
[0203] Not recognized. MEN model output. Element B. Disagreement.
[0204] (neither A nor B) (neither A nor B)
[0205] Not recognized Not recognized MEN+I model output Agreement
[0206] (neither A nor B) (neither C nor B) (neither C nor B)
[0207]
[0208] It appears from the analysis of the previous table that, when the first consolidated PCN prediction and the second consolidated PCN+I prediction are inconsistent, the analysis module 20 is able to deduce that the value of the estimate of the physical quantity is the first consolidated PCN prediction.
[0209] Conversely, when the first consolidated PCN prediction and the second consolidated PCN+I prediction are consistent, the analysis module 20 is able to deduce that the value of the estimate of the physical quantity is the second consolidated PCN+I prediction.
[0210] In this embodiment, the lack of consistency between the two estimation models MEN and MEN+I is thus a sign that the MEN+I estimation model is not sufficiently reliable.
[0211] This is how analysis module 20 analyzes the consistency of values from the first prediction P N and the second prediction P N+i and deduces an estimate of the physical quantity to be estimated.
[0212] The warning module 22 is designed to issue an alert in the event of a temporal persistence of a disagreement in output of 42 (disagreement between the predictions of the two estimation models MEN and MEN+I).
[0213] To assess the existence of temporal persistence, the warning module 22 analyzes the number of disagreements on a set of comparisons performed by the verification unit 42.
[0214] As an example, the warning module 22 will accumulate the results of the comparisons and apply a persistence criterion to at least some of these accumulated results to determine whether or not to issue a warning. In a simple example, when the number of results exceeds a threshold K, the warning module 22 implements, for instance, a synthesis technique as described below. K here is a natural number strictly greater than 1.
[0215] The K threshold is configurable, so the warning model summarizes information over a configurable sliding window.
[0216] More specifically, for comparisons k to kK (k being an integer greater than or equal to K), the number of disagreements is compared to a threshold, exceeding the threshold triggering an alarm from the warning module 22.
[0217] Alternatively, it is possible to consider the percentage of disagreements instead of the number of disagreements.
[0218] It is also possible to consider checking only consolidated results.
[0219] In operation, the estimation system 10 thus implements a method for estimating a physical quantity, for example like the one described with reference to figure 3.
[0220] The process includes a reception step E50, a first application step E52, a second application step E54, an analysis step E56 and a deduction step E58.
[0221] During the reception stage E50, the data acquired by a detection device 12 are received by the first estimation module 16 and the second estimation module 18.
[0222] During the first application step E52, the first computing unit 26 applies the MEN estimation model to the received data to obtain the first prediction PN of the physical quantity to be estimated.
[0223] During the second application step E54, the second computing unit 32 applies the MEN+I estimation model to the received data to obtain the second prediction P N +i of the physical quantity to be estimated.
[0224] The two application steps E52 and E54 can be implemented in parallel, with potentially synchronization constraints depending on the applications.
[0225] During the E56 analysis step, the analysis module 20 analyzes the consistency of values from the first prediction PN and the second prediction P N +i.
[0226] In this case, the values resulting from the first prediction P Nand from the second prediction PN+I are the consolidated values PCN and PCN+I.
[0227] During the deduction step E58, the analysis module 20 deduces from the consistency analysis an estimate of the physical quantity to be estimated. For this, the analysis module 20 applies, for example, the previous truth table.
[0228] The estimation device 14 is thus suitable for monitoring the update of an ME estimation model by comparing the predictions between the MEN estimation model (before the update) and the MEN+I estimation model (after the update).
[0229] More specifically, the estimation device 14 inferentially monitors the new MEN+I estimation model using predictions from the MEN estimation model. This monitoring takes the form of comparing their inputs and / or internal states and / or outputs in real time, and deducing a consolidated prediction from the data seen in inference.
[0230] This comparison is carried out here by verification unit 42 on the basis of a truth table.
[0231] The table of truth here is to be understood in the sense of a logic of comparison and consolidation.
[0232] In this sense, the estimation device 14 corresponds to a two-way processing architecture allowing both the implementation of two estimation models (here the MEN and MEN+I models) and the control of this implementation.
[0233] This also allows for a drastic minimization of the development costs of a monitoring solution because the addition of the increment itself will facilitate this process through the use of at least consistency analysis possibly supplemented by temporal persistence logic (performed by warning module 22).
[0234] Estimation device 14 therefore effectively detects new features and drifts, thereby increasing confidence in the estimation model. Estimation device 14 thus provides a solution to the problem of user qualification and confidence in their estimation system 10 in the context of updates to the estimation model used.
[0235] Estimation method 14 does not require the specific development of a dedicated monitoring tool, which can be costly since it only uses the two estimation models to compare them and verify, in particular, that the new MEN+I estimation model complies with both the new and old requirements (those of the ME estimation model). N ) still valid even though they no longer comply with the old, abandoned requirements. The estimation device 14 can also verify in certain cases that the predefined usage domain of the MEN+I estimation model is indeed preserved when relevant. In the example just described, the first estimation module 16, the second estimation module 18, the analysis module 20 and the warning module 22 are each implemented as software, or as a software and / or hardware component (e.g., FPGA), executable by the processor.
[0236] The memory of the estimation device 14 is then capable of storing a first estimation program, a second estimation program, an estimation program, and a warning program. The processor is then capable of executing each of the programs among the first estimation program, the second estimation program, the estimation program, and the warning program.
[0237] All these units can be developed using software technologies such as assembly, C++, Java, C, Rust, Go, or Python. The choice of programming language will be determined by the criticality level of the use of the estimation system and / or the desired performance.
[0238] The platform on which these units will run will depend on the use case. It could be a simple traditional server or a workstation with graphics processing units (GPUs) locally or in edge computing (sometimes translated as edge computing in French), or a distributed platform.
[0239] In an alternative not shown, the first estimation module 16, the second estimation module 18, the analysis module 20 and the warning module 22 are each implemented as a programmable logic component, such as an FPGA (Field Programmable Gate Array), or an integrated circuit, such as an ASIC (Application Specific Integrated Circuit).
[0240] When the estimation device 14 is implemented as one or more software programs, that is, as a computer program, also called a computer program product, it is also capable of being stored on a computer-readable medium (not shown). A computer-readable medium is, for example, a medium capable of storing electronic instructions and being connected to a bus of a computer system. Examples of such a readable medium include an optical disc, a magneto-optical disc, ROM, RAM, any type of non-volatile memory (e.g., FLASH or NVRAM), or a magnetic card. A computer program comprising software instructions is then stored on the readable medium.
[0241] These remarks relating to the physical implementation of the first estimation module 16, the second estimation module 18, the analysis module 20 and the warning module 22 are also valid for the units that these modules contain.
[0242] Other embodiments of the estimation device 14 are conceivable. In particular, simpler implementations of the estimation device 14 can be considered. For example, a simple comparison of predictions P could be used. N and P N +i.
[0243] It is also possible to directly use the CN and CN+I confidence representative values in the verification unit 42, which avoids the use of the two consolidation units 36 and 40.
[0244] An estimation system 10 according to another embodiment is also illustrated in figures 4 and 5.
[0245] Estimation system 10 has the same elements as estimation system 10 in Figures 1 and 2.
[0246] In particular, the estimation device 14 according to Figure 3 comprises the first estimation module 16, the second estimation module 18, the analysis module 20 and the warning module 22.
[0247] Compared to the previous example, the second computing unit 32 applies a second estimation model that is an intermediate version between the Nth version of the MEN estimation model applied by the first computing unit 26 and the (N+1)th version of the ME estimation model N +I.
[0248] To avoid confusion, the second estimation model will be denoted with the notation ME'N+I. Predictions and confidence values will also be denoted with a ' in the form P' N +i and C'N+I.
[0249] The estimation system 10 also includes a third estimation module 60, an inhibition module 62 and a control module 64.
[0250] The third estimation module 60 comprises a third receiving unit 66, a third calculation unit 68 and a third determination unit 70.
[0251] The third receiving unit 66 has the same role as the first receiving unit 24 or the second receiving unit 30.
[0252] In one embodiment, the first receiving unit 24, the second receiving unit 30 and the third receiving unit 66 are combined.
[0253] The third computing unit 68 is designed to apply a third estimation model to the data received by the third receiving unit 66 in order to obtain a third prediction P N +i.
[0254] The third estimation model corresponds either to a second update of the first estimation model MEN OR an update of the estimation model ME'N+I here by implementing incremental learning.
[0255] The second update is different from the first update. In other words, in the example described, the third estimation model corresponds to version (or increment) N+1 of the estimation model.
[0256] Therefore, the third estimation model is denoted ME N +I and will therefore be referred to hereafter as the MEN+I estimation model to simplify notations.
[0257] The third unit of determination 70 is suitable for determining a third value representing the confidence CN+I in the prediction of the third unit of calculation 68.
[0258] In one embodiment, the second unit of determination 34 and the third unit of determination 70 are coincident.
[0259] The inhibition module 62 is designed to inhibit the reception by the analysis module 20 of the prediction of at least one estimation module among the first estimation module 16, the second estimation module 18 and the third estimation module 60.
[0260] Inhibiting here means either "preventing" or "interrupting" depending on the function chosen for the inhibition module 62. An example for each of these two definitions will be given later in the description.
[0261] According to the example described, the inhibition module 62 is specific to inhibiting only the reception of the prediction P' N +i of the second estimation module 18 or the prediction P N +i of the third estimation module 60.
[0262] This implies that the estimation device 14 always has the first prediction P N , this prediction serving in this context as a reference prediction.
[0263] Furthermore, the inhibition module 62 has two states.
[0264] In such a case with only two states, the inhibition module 62 can be considered as a switching module insofar as the inhibition module 62 allows switching between the prediction of the second estimation module 18 and the prediction of the third estimation module 60.
[0265] The first state is a state of inhibition module 62 in which only the reception of the prediction from the second estimation module 18 is inhibited. In other words, the first state is a state of inhibition of the second estimation module 18.
[0266] The second state is a state of inhibition module 62 in which only the reception of the prediction from the third estimation module 60 is inhibited. The second state is thus a state of inhibition of the third estimation module 60.
[0267] An example of the physical realization of the inhibition module 62 is given for illustrative purposes in Figure 5.
[0268] In this example, the inhibition module 62 has an inhibition unit 71 for each estimation module 18 and 60, of which the inhibition module 62 is specific to inhibit the reception of the prediction by the analysis module 20. For example, the inhibition unit 71 is a switch having an open position in which the reception of the prediction by the analysis module 20 is inhibited and a closed position in which the reception of the prediction by the analysis module 20 is not inhibited.
[0269] The control module 64 is used to control the state of the inhibition module 62, that is, to send it a command to put itself in one of the previous states.
[0270] According to a particular example, the control module 64 is specific to triggering the implementation of a loading operation of the third estimation model MEN+I on the third computing unit 68.
[0271] For this purpose, the control unit 64 is designed to interact with an external device to be informed of the availability of the MEN+I estimation model and to obtain this model.
[0272] Such acquisition is usually implemented via on-the-fly downloading.
[0273] This implies that the estimation system 10 is also equipped with a reliable, efficient, and secure remote download unit (transmitter / receiver) for the estimation model weights. Access to a memory storing the weights for reading and writing is also used.
[0274] In such a case, the control module 64 applies a control law on the inhibition module 62 with the effect that the inhibition module 62 is in the first state when the loading operation has not been implemented and in the second state when the loading operation has taken place.
[0275] This ensures that the analysis module 20 has the most up-to-date version of the estimation model, namely first the intermediate estimation model and then the MEN+I estimation model when it is available.
[0276] The analysis module 20 then works as before, the predictions and confidence values received being those from the first estimation module 16 as well as either those from the second estimation module 18 or those from the third estimation module 60.
[0277] An example of the operation of the estimation system 10 of figures 4 and 5 is now described with reference to figure 6 which illustrates an example of the implementation of a method for estimating a physical quantity by this estimation system 10.
[0278] The estimation process includes a first reception step E72, a first application step E74, a second application step E76, a third application step E78, a second reception step E80 and a obtaining step E82. During the first reception step E72, the data acquired by a detection device 12 are received by the first estimation module 16, the second estimation module 18 and the third estimation module 60.
[0279] During the first application step E74, the first computing unit 26 applies the MEN estimation model to the received data to obtain the first prediction PN of the physical quantity to be estimated.
[0280] During the second application step E76, the second computing unit 32 applies the estimation model ME'N+I to the received data to obtain the second prediction P'N+I of the physical quantity to be estimated.
[0281] During the third application step E78, the third computing unit 60 applies the estimation model MEN+I to the received data to obtain the third prediction PN+I of the physical quantity to be estimated.
[0282] Application steps E74, E76 and E78 can be implemented simultaneously.
[0283] During the second reception stage E80, the analysis module 20 receives exactly two predictions systematically including the reception of the prediction from the first estimation module 16, and the reception of a prediction from the second estimation module 18 or the third estimation module 60, the other being inhibited by an inhibition module 62.
[0284] During the E82 acquisition step, the analysis module 20 obtains an estimate of the physical quantity to be estimated for the estimation device 14 from the predictions received.
[0285] Such an E82 obtaining step is, for example, implemented by a consistency analysis as described previously.
[0286] Such a general operation can be used for several different use cases. A first example was given for the case of downloading the ME estimation model N +I in use. The intermediate estimation model ME' N +I is then used due to the lack of availability of the MEN+I estimation model, particularly because its training takes too long.
[0287] This ensures continuity in the use of the estimation model, as updated estimation models are made available continuously and the transition from one version of the estimation model to another is easy since it does not involve interruption of use (shutdown and restart avoided).
[0288] In a second example, the ME'N+I intermediate estimation model presents a major security problem. Therefore, an update of the ME'N+I intermediate estimation model could be considered to address this major issue. In such a case, the intermediate update would only affect the weight values, while maintaining the rest of the neural network architecture (e.g., the same number of neurons, layers, and operators). This involves performing a minor adjustment (more often referred to by the English term "fine-tuning").
[0289] This allows only the weights of the estimation model to be loaded. This technique can also be used for loading version N+1 of the MEN+I estimation model.
[0290] Indeed, such a technique is compatible with the constraints of embedded systems where resources are limited.
[0291] Through these non-limiting examples, it appears that the estimation system 10 allows for the development and qualification of incremental learning of an estimation model with relaxed constraints.
[0292] To this end, the estimation system 10 offers a three-way processing architecture allowing the implementation and monitoring of an N version of the estimation model, an intermediate version of the estimation model and an N+1 version of the estimation model (for example obtained by a patch loaded on the fly).
[0293] The estimation system 10 thus takes advantage of the usage time to continue the qualification of version N+1 (carried out under constraints) and thus have a version N+1 that is as robust, mature and stable as possible.
[0294] Other implementation methods are also conceivable.
[0295] In particular, it is possible to share the components of the second estimation module 18 and the third estimation module 60.
[0296] For example, the unit of determination which is specific to determining the representative values of confidence in the prediction can be shared, with the inhibition module 62 then being positioned upstream of this common unit of determination.
[0297] It may also be envisaged that an inhibition module 62 is positioned so as to prevent the sending of data to the computing unit that the inhibition module 62 wishes to inhibit.
[0298] Each of the 10 observation systems just described can be used in a collaborative context.
[0299] Such a case is shown schematically in figure 7 in which two sets 100 and 102 are represented.
[0300] An assembly here is a carrier, typically a drone, a fighter, a transporter, a tanker or a surveillance aircraft. Each assembly 100 or 102 includes an estimation system 10 and a communication system 104 allowing communication with the other assembly 100 or 102.
[0301] The observation systems 10 are either (i) homogeneous, i.e., of the same type, either according to the first example in Figure 2 or according to the second example in Figure 3, or (ii) heterogeneous combining the two examples.
[0302] Each communication system 104 is designed to exchange information between each set 100 or 102.
[0303] In the example described, the information consists of values representing the consistency between the predictions of the estimation models used by each estimation system 10.
[0304] More specifically, in the case of Figure 7, the information is the result of comparisons of the 42 verification units of each 10 estimation system.
[0305] The first set 100 also includes a coordination system 106. The coordination system 106 is, for example, a computer.
[0306] The coordination system 106 is designed to obtain, through interactions with the communication systems 104, the results of the comparisons of the verification units 42 of each estimation system 10.
[0307] The 106 coordination system then applies an aggregation function to the representative values of the consistency between the predictions to obtain an aggregate value.
[0308] In general, aggregation functions are mathematical operations that allow you to combine values to obtain a final value.
[0309] In a particular example, the aggregation function is a majority vote.
[0310] This vote can be weighted, for example by taking into account the fact that a set may be better placed to observe the environment and therefore, a priori, its prediction is more reliable.
[0311] Alternatively, the aggregation function is a fuzzy logic function.
[0312] According to another embodiment, the aggregation function is a statistical function, notably by calculating performance statistics based on the task performed by the estimation model.
[0313] The 106 coordination system is suitable for deducing an estimate of the physical quantity to be estimated from the aggregated value.
[0314] As an example for the case described, the 106 coordination system implements the deduction according to the following truth table. Aggregation
[0315] Result for Result for system of
[0316] Prediction deduced set 100 set 102 coordination
[0317] 106
[0318] Disagreement Disagreement MEN model exit Disagreement Disagreement MEN model exit Agreement Disagreement Disagreement MEN model exit ME model exit N +I Agreement Agreement Agreement
[0319] or ME'N+I
[0320]
[0321] The coordination system 106 is then able to share with each estimation system 10 a consolidated recommendation based on an overall consistency status which indicates the estimation of the physical quantity to be used by each estimation device 14.
[0322] For this purpose, the coordination system 106 includes a communication unit (not shown) specifically designed to communicate with each of the communication systems 104.
[0323] The associated operation corresponds to the flowchart shown schematically in figure 8. The estimation process includes a obtaining step E110, an application step E112 and a deduction step E114.
[0324] During the acquisition step 110, each analysis module 20 analyzes the consistency of values from predictions obtained by applying estimation models to data acquired by a detection device 12, an estimation model being an update of another estimation model, to obtain a value representative of the consistency between the predictions.
[0325] During the application step E112, the coordination system 106 applies an aggregation function to the representative values of the consistency between the predictions to obtain an aggregated value.
[0326] During the deduction step E114, the coordination system 106 deduces an estimate of the physical quantity to be estimated from the aggregated value.
[0327] The estimation process just described allows for the addition of a collective type consolidation logic.
[0328] This helps to increase the reliability of the prediction.
[0329] The estimation system 10 therefore allows us to benefit from the simultaneous presence of several sets 100 and 102 to consolidate the prediction, which makes it possible to increase the reliability of the simultaneous use of the two estimation models MEN and MEN+I OR MEN and ME'N+I. Other modes of embodiment are conceivable.
[0330] In particular, with reference to Figure 9, the coordination system 106 can be a central system positioned at a distance from each of the sets 100 and 102.
[0331] The coordination system 106 remains in contact with each of the communication systems 104.
[0332] Alternatively or in addition, as shown schematically in Figure 10, the values representing the consistency between the predictions of the estimation models are the values obtained by the alert modules 22.
[0333] Typically, the coordination system 106 will use the percentage of disagreement of each alert module 22 to deduce the prediction common to each estimation system 10.
[0334] More broadly, each warning module 22 is designed to obtain a value representative of the temporal evolution of the consistency analysis results, and the values representative of consistency between predictions are these values representative of temporal evolution.
[0335] As a useful example, one could consider using a cascading system of individual consistency statuses (denoted by the acronym SIC), individual collaborative consistency statuses (denoted by the abbreviation SICC), and persistent collaborative consistency statuses (SPCC). This allows for the combination of two consolidations: consolidation between collaborative units (SICC) and temporal consolidation (SPCC).
[0336] This can be illustrated for a case with two units by the following table:
[0337] Time SIC unit 1 SIC unit 2 SICC SPCC 0 agreement disagreement disagreement
[0338] 1 disagreement agreement disagreement agreement agreement agreement disagreement K-1 agreement agreement agreement agreement K agreement disagreement disagreement disagreement
[0339]
[0340] In this example, unit 1 is the master unit, while unit 2 is the follower unit. Extending this to more than one follower unit is straightforward, as it simply requires adding as many columns as there are follower units.
Claims
DEMANDS 1. Estimation device (14) for a physical quantity, the estimation device (12) comprising: - a first estimation module (16), the first estimation module (16) comprising: - a first receiving unit (24) suitable for receiving data acquired by a detection device (12), and - a first computing unit (26) designed to apply a first estimation model to the data received by the first receiving unit (24) in order to obtain a first prediction (PN) of the physical quantity to be estimated, - a second estimation module (18), the second estimation module (18) comprising: - a second receiving unit (30) suitable for receiving data acquired by a detection device (12), and - a second computing unit (32) suitable for applying a second estimation model to the data received by the second receiving unit (30) to obtain a second prediction (P'N+I) of the physical quantity to be estimated, the second estimation model being a first update of the first estimation model, - a third estimation module (60), the third estimation module (60) comprising: - a third receiving unit (66) suitable for receiving data acquired by a detection device (12), and - a third computing unit (68) designed to apply a third estimation model to the data received by the third receiving unit (66) to obtain a third prediction (PN+I) of the physical quantity to be estimated, the third estimation model being a second update of the first estimation model, - an analysis module (20), the analysis module (20) being designed to receive each prediction and to obtain an estimate of the physical quantity to be estimated for the estimation device (14) from the predictions received, and - an inhibition module (62), the inhibition module (62) being adapted to inhibit the reception by the analysis module (20) of the prediction of at least one estimation module among the first estimation module (16), the second estimation module (18) and the third estimation module (60).
2. Estimation device according to claim 1, in which the inhibition module (62) comprises an inhibition unit (71) for each estimation module (18, 60) whose inhibition module (62) is adapted to inhibit the reception of the prediction by the analysis module (20).
3. Estimation device according to claim 2, wherein each inhibition unit (71) comprises at least one switch having an open position in which the reception of the prediction by the analysis module (20) is inhibited and a closed position in which the reception of the prediction by the analysis module (20) is not inhibited.
4. Estimation device according to any one of claims 1 to 3, wherein the inhibition module (62) is suitable for inhibiting only the reception of the prediction from the second estimation module (18) and the third estimation module (60).
5. Estimating device according to claim 4, wherein the inhibition module (62) has a first state in which only the reception of the prediction of the second estimating module (18) is inhibited and a second state in which only the reception of the prediction of the third estimating module (60) is inhibited, the estimating device (14) further comprising a control module (64), the control module (64) being adapted to control the state of the inhibition module (62).
6. Estimation device according to claim 5, wherein the control module (64) is adapted to drive the implementation of a loading operation of the third estimation model onto the third computing unit (68) and to control the inhibition module (62) so that the inhibition module (62) is in the first state when the loading operation has not been implemented and in the second state when the loading operation has taken place.
7. Estimation device according to any one of claims 1 to 6, wherein the analysis module (20) is suitable for analyzing the consistency of values from the predictions received and for deducing from the consistency analysis an estimate of the physical quantity to be estimated for the estimation device (14).
8. Estimation device according to any one of claims 1 to 7, wherein the predictions received by the analysis module (20) are the first prediction and one prediction from the second prediction and the third prediction.
9. Estimation system (10), the estimation system (10) being suitable for estimating a physical quantity, the estimation system (10) comprising: - a detection device (12), the detection device (12) being suitable for acquiring data, and - an estimation device (14) for a physical quantity, the estimation device (14) being according to any one of claims 1 to 8.
10. Method for estimating a physical quantity comprising the steps of: - receiving data acquired by a detection device (12), - application of a first estimation model to the received data to obtain a first prediction (PN) of the physical quantity to be estimated, - application of a second estimation model to the received data to obtain a second prediction (PN+I) of the physical quantity to be estimated, the second estimation model being a first update of the first estimation model, - application of a third estimation model to the received data to obtain a third prediction (P N +i) of the physical quantity to be estimated, the third estimation model being a second update of the first estimation model, - reception of each prediction, the reception of the prediction of at least one estimation module among the first estimation module (16), the second estimation module (18) and the third estimation module (60) being inhibited by an inhibition module (62), and - obtaining an estimate of the physical quantity to be estimated for the estimation device (14) from the predictions received.
