METHOD FOR ESTIMATING THE STATE OF WEAR OF A TIRE

Abstract

A method for estimating the state of wear of a tire comprises: recording a vibroacoustic signal (1001) produced by the tire running on a road during a time frame; converting the time signal (1001) into a frequency signal (1002) over a frequency range; segmenting the frequency range into frequency bands and associating a datum representing the frequency signal with each band, with the representative datum forming a variable of a matrix (1003); predicting a state of wear from the matrix, by means of machine learning (1004) from the data based on a learning database, according to modalities each representing a state of wear of the tire; and determining the state of wear of the tire (1005) after a number N of identical predictions in a series M of consecutive predictions.

Claims

1-11. (canceled)

12. A method for estimating a state of wear of a tire of a mounted assembly of a vehicle traveling on a surface of a road, comprising the following steps: measuring a vibroacoustic signal (1001) produced by the tire running on the road surface during a given time frame; converting the time signal (1001) into a frequency signal (1002) over a given frequency range; segmenting the frequency range into at least one frequency band with a predetermined width and associating at least one datum representing the frequency signal in the at least one frequency band with the at least one frequency band, with the at least one representative datum derived from the measurement forming at least one variable of a matrix (1003) associated with the measurement; predicting the state of wear of the tire corresponding to the matrix associated with the completed measurement, by means of machine learning (1004) from the data based on a learning database made up of a set of matrices associated with measurements recorded and carried out according to the same steps as above, under known running conditions, according to modalities each representing a state of wear of the tire; determining a ground surface state category (3002), where the ground surface state category (3002) is a condition for predicting the state of wear of the tire by the machine learning step (1004) if it is a specific ground state category (3003) containing a dry ground surface state category or the dry and a wet ground surface state categories; determining a ground nature category (4002), where the ground nature category (4002) is a condition for predicting the state of wear of the tire by the machine learning (1004) if it is a specific ground nature category (4003) comprising an open ground nature category; and determining the state of wear of the tire (1005) after a number N of identical predictions in a series M of consecutive predictions.

13. The method for estimating the state of wear of a tire according to claim 12, wherein the method further comprises the following step: determining a tire running speed category (2002), a breadth of which is a fraction of a maximum running speed, wherein the running speed category (2002) is a modality of the machine learning (1004).

14. The method for estimating the state of wear of a tire according to claim 13, wherein determining the running speed category (2002) of the tire comprises the following steps: recording a second measurement of a vibroacoustic signal (1001) produced by the tire running on the road surface during a second given time frame; converting the second time signal (1001) into a second frequency signal (1002) over a second given frequency range; segmenting the second frequency range into at least one frequency band with a predetermined width and associating at least one datum representing the second frequency signal in the at least one frequency band with the at least one frequency band, with the at least one representative datum derived from the second measurement forming the at least one variable of a matrix (1003) associated with the second measurement; and determining the running speed category (2002) of the tire corresponding to the matrix associated with the second completed measurement, by means of second machine learning (2004) from the data based on a learning database made up of a set of matrices associated with measurements recorded and carried out according to the same steps as above, under known running conditions, according to modalities each representing a running speed category of the tire.

15. The method for estimating the state of wear of a tire according to claim 12, wherein the ground surface state category (3002) is included in the group consisting of dry, wet, damp, snowy and icy ground surface state categories.

16. The method for estimating the state of wear of a tire according to claim 12, wherein determining the ground surface state category (3002) comprises the following additional steps: recording a third measurement of a vibroacoustic signal (1001) produced by the tire running on the road surface during a given third time frame; converting the third time signal into a third frequency signal (1002) over a third given frequency range; segmenting the third frequency range into at least one frequency band with a predetermined width and associating at least one datum representing the third frequency signal in the at least one frequency band with the at least one frequency band, with the at least one representative datum derived from the third measurement forming the at least one variable of a matrix (1003) associated with the third measurement; and determining a ground surface state category (3002) corresponding to the matrix associated with the third completed measurement, by means of third machine learning (3004) from the data based on a learning database made up of a set of matrices associated with measurements recorded and carried out according to the same steps as above, under known running conditions, according to modalities each representing at least one ground surface state category.

17. The method for estimating the state of wear of a tire according to claim 12, wherein the ground nature category (4002) is included in the group consisting of open, medium and closed ground nature categories.

18. The method for estimating the state of wear of a tire according to claim 12, wherein the specific ground nature category (4003) includes ground with an ATD that is greater than 0.7.

19. The method for estimating the state of wear of a tire according to claim 12, wherein determining the ground nature category (4002) comprises the following additional steps: recording a fourth measurement of a vibroacoustic signal (1001) produced by the tire running on the road surface during a given fourth time frame; converting the fourth time signal (1001) into a fourth frequency signal (1002) over a given fourth frequency range; segmenting the fourth frequency range into at least one frequency band with a predetermined width and associating at least one datum representing the fourth frequency signal in the at least one frequency band with the at least one frequency band, with the at least one representative datum derived from the fourth measurement forming the at least one variable of a matrix (1003) associated with the fourth measurement; and determining the ground nature category (4002) corresponding to the matrix associated with the fourth completed measurement, by means of fourth machine learning (4004) from the data based on a learning database made up of a set of matrices associated with measurements recorded and carried out according to the same steps as above, under known driving conditions, according to modalities each representing at least one ground nature category.

20. The method for estimating the state of wear of a tire according to claim 19, wherein the step of determining the ground nature category comprises the ground surface state category (3002) as a modality of the fourth machine learning.

21. The method for estimating the state of wear of a tire according to claim 12, wherein the machine learning method (1004) is selected from neural networks, discriminant analysis, support vector machines, boosting, K-nearest neighbors methods and logistic regression.

22. The method for estimating the state of wear of a tire according to claim 12, wherein the state of wear of the tire is selected from new, half-worn, and worn.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The invention will be better understood upon reading the following description, which is provided solely by way of a non-limiting example and with reference to the accompanying figures, in which the same reference numbers in all cases designate identical parts and in which:

[0071] FIG. 1 shows a general overview of the method for estimating the state of wear of a tire according to the invention;

[0072] FIG. 2 shows a vehicle equipped with a measuring device allowing the state of wear of the tire to be determined;

[0073] FIG. 3 shows a discriminant space along two main axes of the method for estimating the state of wear of the tire taking into account only the modalities of the state of wear of the tire according to the invention;

[0074] FIG. 4 shows noise spectra measured on a vehicle travelling on a road, for which the state and ground nature surface are known;

[0075] FIG. 5 shows a discriminant space along a main axis of the method for determining the ground surface state category taking into account only the modality related to the ground surface state according to the invention;

[0076] FIG. 6 shows a discriminant space along two main axes of the method for estimating the state of wear of the tire taking into account the modalities of the state of wear of the tire and the running speed categories for a specific ground state category, namely the “dry” state, according to the invention;

[0077] FIG. 7 shows a discriminant space along two main axes of the method for determining the ground nature category according to the invention;

[0078] FIG. 8 shows a discriminant space along two main axes of the method for estimating the state of wear of the tire by selecting the measurements according to the specific ground state category and the specific ground nature category criteria.

DETAILED DESCRIPTION OF EMBODIMENTS

[0079] FIG. 1 shows a complete overview of the method for estimating the state of wear of a tire using a vibroacoustic measurement 1001 carried out on a vehicle when running, according to several embodiments.

[0080] This temporal measurement 1001 is converted into a frequency spectrum 1002 using standard computation tools such as Fast Fourier Transform.

[0081] The frequency spectrum 1002 is then divided into different frequency bands in accordance with the selected use. On each frequency band, one or more physical quantities of the reduced frequency spectrum is/are associated with the frequency spectrum on the frequency band. The set of quantities forms a vector, the length of which is proportional to the number of computed physical quantities. This allows a matrix 1003 to be filled, one of the dimensions of which is the number of frequency bands obtained from the full width of the frequency spectrum. The second dimension of the matrix 1003 corresponds to the maximum number of physical quantities assessed per selected frequency band. Generally, the matrix is a vector, the length of which is the number of selected frequency bands and the second dimension is a scalar dimension. This also can be a two-dimensional matrix or the second dimension is a vector.

[0082] In a first embodiment, the matrix 1003 is introduced into machine learning 1004 comprising a learning database. The learning database was formed during a prior step of learning by a series of vibroacoustic measurements and of frequency processing on the temporal measurement where the modalities of the learning database were managed. In a conventional embodiment, the modalities are the state of wear categories of the tire comprising at least the worn state and the new state, preferably also the half-worn state. The machine learning provides a prediction of the state of wear of the tire.

[0083] Repeating the measurements, post-processing and predictions allows a series M of prediction results to be formed. Regularly repeating the same result in the series allows the change of the state of wear of the tire to be confirmed. In addition, generally, the initial state of wear of the tire is the new state, which transforms over time and as such changes the prediction series to the half-worn state and then to the worn state. Thus, the redundancy of the prediction results on the same state of wear, knowing that the change in the state of wear can only occur in one direction, generally allows the actual state of wear of the tire to be quickly determined in the form of a state of wear category. The higher the number of wear categories, resulting in greater precision in determining the state of wear of the tire, the less efficient the method since the quality of the result of the prediction will be affected by all the influential parameters other than the state of wear of the tire. For example, the running speed, the ground nature, the meteorological conditions of the ground, but also, the vehicle, the operating conditions of the tire in terms of pressure, of applied load, of outside temperature, etc., can be cited.

[0084] According to a second embodiment, in order to make the method for estimating the state of wear of the tire more reliable, the prediction by the machine learning 1004 can also take into account the running speed of the tire 2001. Indeed, this parameter significantly influences the average level of the frequency spectrum 1002 obtained from the temporal vibroacoustic measurement 1001. Taking into account this parameter in the form of running speed categories as a modality of the machine learning allows incorrect predictions to be reduced.

[0085] This running speed of the tire can be obtained by additional sensors of the vehicle or obtained through information passing through the electronic wiring of the vehicle. This determination can occur directly in the form of categories or can be stored in the form of running speed categories preferred by the machine learning. However, in a variant, the running speed category 2002 is determined using a second vibroacoustic measurement 1001 obtained on the vehicle. This second vibroacoustic measurement 1001 is advantageously the vibroacoustic measurement 1001 that will be used for predicting the state of wear of the tire in step 1004.

[0086] As before, the temporal vibroacoustic measurement 1001 is converted into a frequency spectrum 1002. This frequency spectrum is then divided into frequency bands. One or more physical quantities of the frequency spectrum is/are associated with each frequency band. This allows a matrix 1003 to be completed that is associated with the prediction of the running speed category. However, the division into a frequency band does not need to be similar to that carried out for predicting 1004 the state of wear of the tire. The identification of the speed category 2002 is largely sufficient for predicting 1004 the state of wear of the tire.

[0087] Optionally, it is also possible to determine the meteorological surface state 3001 of the ground where the vehicle is traveling in order to make the method for estimating the state of wear of the tire more reliable, the prediction by the machine learning 1004 can also take into account these meteorological conditions in accordance with two distinct routes.

[0088] The first route involves determining categories 3002 of the state of the ground and taking into account these categories of the state of the ground as a modality of the prediction 1004 of the state of wear of the tire. Knowing the meteorological ground surface state, at least in terms of categories, allows the prediction to be made more reliable, to the detriment of a larger learning database and of a more complex mathematical model.

[0089] The second route involves determining a specific ground state category 3003, for which the prediction of the state of wear of the tire will be carried out. In fact, from among all the vibroacoustic measurements 1001 the ones selected are those that correspond to a state of the ground promoting the prediction 1004 of the state of wear of the tire. Thus, the learning database associated with the prediction is reduced and the mathematical model is more basic, which allows fast computation times with limited resources and reliable prediction.

[0090] The meteorological surface state 3001 of the ground is separated into various categories. In summer or normal conditions, at least the dry state, the wet/damp state are distinguished, and the second group is even separated according to the water level on the road. This can also include the winter conditions, such as the icy state or the snowy state.

[0091] The applicant has found that the specific ground state category must include the “dry” state. Indeed, this ground state category has a high occurrence in the vibroacoustic measurements 2001 and can be more reproducible for division into a frequency band in step 1003. Thus, focusing on the vibroacoustic measurements where the ground state category is specific is not detrimental in terms of the occurrence of the vibroacoustic measurements 1001 in a large majority of territories, while being efficient in terms of prediction due to the similarity of the frequency spectra 1002 for these meteorological conditions.

[0092] Of course, taking into account the ground state categories in this way can occur with or without taking into account the running speed 2001 in the machine learning step 1004. However, determining the ground state category 3002 will necessarily occur after determining the speed category 2002 if the intention is to take everything into account.

[0093] Finally, the meteorological conditions of the ground 3001 can be obtained by additional sensors of the vehicle, such as a rain detector on the windshield, or the actuator of the trigger of the windshield wipers or obtained through the information passing through the electronic wiring of the vehicle. However, in a variant, the ground state category 3002 is determined using a third vibroacoustic measurement 1001 obtained on the vehicle. This third vibroacoustic measurement 1001 is advantageously the vibroacoustic measurement 1001 that will be used for predicting 1004 the state of wear of the tire and/or the vibroacoustic measurement 1004 that was used to determine 2004 the running speed category 2002.

[0094] As before, the temporal vibroacoustic measurement 1001 is converted into a frequency spectrum 1002. This frequency spectrum is then divided into frequency bands. One or more physical quantities of the frequency spectrum is/are associated with each frequency band. This allows a matrix 1003 to be completed that is associated with determining the meteorological conditions category 3002 of the ground. However, the division into a frequency band does not need to be similar to that carried out for predicting 1004 the state of wear of the tire or for determining the running speed category 2002. In fact, the identification of the ground state category 3002 is largely sufficient for predicting 1004 the state of wear of the tire.

[0095] Optionally, it is also possible to determine the nature 4001 of the texture of the ground where the vehicle travels in order to make the method for estimating the state of wear of the tire more reliable, the prediction by the machine learning 1004 can also take into account ground nature categories in accordance with two distinct routes.

[0096] The first route involves determining ground nature categories 4002 and taking into account these categories of the ground nature as a modality of the prediction 1004 of the state of wear of the tire. Knowing the ground nature 4001, at least in terms of categories, allows the prediction 1004 to be made more reliable, to the detriment of a larger learning database and of a more complex mathematical model.

[0097] The second route involves determining a specific ground nature category 4003, for which the prediction 1004 of the state of wear of the tire will be carried out. In fact, from among all the vibroacoustic measurements 1001 the ones selected are those that correspond to a ground nature promoting the prediction 1004 of the state of wear of the tire. Thus, the learning database associated with the prediction is reduced and the mathematical model is more basic, which allows fast computation times with limited resources and a reliable prediction.

[0098] The ground nature 4001 is separated into various categories according to the roughness on a millimetric scale. This ground nature is characterized by the ATD.

[0099] The applicant has found that the specific ground nature category must include ground surfaces referred to as “open” ground surfaces. Indeed, this category allows more reproducibility with respect to the frequency band division of step 1003. Thus, focusing on the vibroacoustic measurements where the ground nature category is specific is efficient in terms of prediction due to a certain similarity of the frequency spectra 1002. However, it is quite possible to extend the specific ground nature category to all ground surfaces having an ATD of more than 0.7, which also covers the upper part, made of ATD material, of the ground surfaces referred to as “medium” ground surfaces.

[0100] Of course, the ground nature categories can be taken into account with or without taking into account the running speed 2001 or the surface state 3001 in the machine learning 1004. However, taking into account the ground nature category will necessarily occur after determining the speed category 2002 if the intention is to take into account these two parameters.

[0101] Finally, the ground nature 4001 can be obtained by additional sensors of the vehicle, such as laser or sound measurement devices or obtained through the information passing through the electronic wiring of the vehicle. However, in a variant, the ground nature category 4002 is determined using a fourth vibroacoustic measurement 1001 obtained on the vehicle. This fourth vibroacoustic measurement 1001 is advantageously the vibroacoustic measurement 1001 that will be used for predicting the state of wear of the tire in step 1004 and/or the vibroacoustic measurement 1001 used for determining the running speed category 2002 and/or the vibroacoustic measurement 1001 used for determining the ground state category 3002.

[0102] As before, the temporal vibroacoustic measurement 1001 is converted into a frequency spectrum 1002. This frequency spectrum is then divided into frequency bands. One or more physical quantities of the frequency spectrum is/are associated with each frequency band. This allows a matrix 1003 to be completed that is associated with determining 4002 the ground nature category 3002. However, the division into a frequency band does not need to be similar to that carried out for predicting 1004 the state of wear of the tire or for determining the running speed category 2002 or for determining the ground state category 3002. In fact, the identification of the ground state category 3002 is largely sufficient for predicting 1004 the state of wear of the tire.

[0103] Finally, the applicant has found that taking into account the meteorological conditions and then the ground nature allows the prediction 1004 of the state of wear of the tire to be made more reliable. Indeed, the separating power of the ground state category is stronger than the separating power of the ground nature category.

[0104] In conclusion, optionally taking into account the ground state category 3002 and the ground nature category 4002 allows the prediction to be made more reliable. The combination is the most efficient configuration in the aforementioned order. The efficiency is measured by the error rate of the prediction.

[0105] However, by taking the specific categories route, the method becomes more efficient due to the reduced size of the learning database of the prediction 1004 of the state of wear of the tire and the speed and simplicity of the computations that can be carried out on board the vehicle and in real time.

[0106] In the main embodiment, the sound signal generated by the tire (T) is measured using a microphone (1) placed in the vehicle. In FIG. 2, the microphone is placed in the front part of a wheel housing located at the rear of the vehicle (C). However, other locations such as, for example, the rear bumper, can be contemplated. The selection of the location for the microphone depends, for example, on the type of data to be estimated, the type of vehicle and the external constraints related to its installation, to its maintenance, and to its durability.

[0107] FIG. 2 shows a vehicle C traveling on a ground surface G, shown schematically, comprising front and rear wheel housings, in which the wheels fitted with tires T are housed.

[0108] When the vehicle C moves, the tire T generates a noise, the amplitude and frequency of which depend on many factors. This sound pressure is in fact the superposition of noises from various origins, such as noises generated by tread pattern elements coming into contact with the ground G, by the movements of air between the tread pattern elements, by the water particles lifted by the tire, or even by the air flows associated with the speed of the vehicle. Listening for these noises is also superimposed with the noises associated with the environment of the vehicle, such as the engine noise. All these noises are also dependent on the speed of the vehicle.

[0109] A measurement means, such as a microphone 1, is installed in the vehicle. It should be noted herein that various positions can be contemplated for the measurement means, with only one being shown in FIG. 1, but without limiting the scope of the invention to this configuration. Thus, the sensor can be positioned on a wall of the rear bumper, for example, but not necessarily oriented so as to detect a sound signal originating from the rear of the vehicle.

[0110] It is also possible to contemplate a position on a wall of the front bumper of a vehicle. The measurement means also can be positioned in a wheel housing in order to listen for the running noises as close as possible to the site where they are generated. Ideally, installing a vibroacoustic sensor in each of the wheel housings can be considered to be the best means of detecting all the noises and running vibrations generated by the tires. However, in order to determine the ground state (meteorological conditions) and the ground nature (macro-texture of the coating), a single microphone is sufficient. In this latter case, isolating it from the aerodynamic and engine noises is preferable.

[0111] Of course, the operating precautions are taken in order to protect the measurement means from external aggressions such as projections of water, mud or gravel.

[0112] The vehicle also comprises a computer 2, connected to the measurement means, and configured to execute the operations for shaping and analyzing, as will be described in detail hereafter, the raw information derived from the measurement means, and to estimate the state of the tire as a function of a measurement of the vibroacoustic emission detected by the measurement means.

[0113] FIG. 3 is a view showing the state of wear of the tire distributed in three identical size categories for vehicle running speeds ranging between 20 and 130 km/h. The first category is the “new” wear category corresponding to an upper third of the useful height of the tire tread. The useful height is delimited by the maximum height of the tread corresponding to the radially external end of the tire relative to its natural axis of rotation and the minimum height is defined by the radially external end of the wear indicator at the bottom of the groove. The second category, referred to as the “half-worn” category, corresponds to the middle third of the useful height of the tire tread. Finally, the last category, referred to as the “worn” category, corresponds to the lower third of the useful height of the tread.

[0114] Herein, a set of vibroacoustic measurements carried out on a vehicle is displayed irrespective of the conditions of the vehicle in terms of applied load and inflation pressure starting from the nominal vehicle configuration. The vehicle has followed a road circuit comprising an urban route, a route on a country road and a highway route over several sessions in an intermediate season, which allows all the conditions of the ground surface state to be mixed, in particular the “dry”, “wet” and “damp” ground states and on various types of ground surface nature, in particular “closed”, “medium” and “open” ground surface natures. The learning database takes into account the state of wear of the tire as a modality. The measurements are taken in this case with tires that have the three types of state of wear of the tire.

[0115] Machine learning relative to its conditions mathematically defines main directions. FIG. 3 shows the discriminant space in the two first main directions that represent the two axes of representation of the graph. The machine learning identifies three series of circles in this two-dimensional representation. The first series, represented by dashed circles, represents the various probabilities of the “new” state of wear in this two-dimensional discriminant space. The circles are concentric. The highest probability, i.e., greater than 0.9, is defined by the surface delimited by the smallest circle. Then, the following circle defines a degressive probability of 0.1, that is a value of 0.8. Furthermore, the following circle represents a probability that has further decreased by 0.1, that is a value of 0.7. The second series of circles represented by circles as a grey solid line represents the various probabilities of the “half-worn” state of wear in this two-dimensional discriminant space. Finally, the third series of circles represented by circles as a black solid line in the same way represents the various probabilities of the “worn” state of wear. It should be noted that the three series of circles are generally separate, in particular in the vicinity of their small circle, referred to as the main circle, which allows the measurements to be sorted according to the three states of wear. However, the circles are big and the secondary circles mutually overlap. As a result, there is uncertainty with respect to the prediction that has been made. It is possible for an incorrect prediction to be made. Thus, by statistically multiplying the measurements, incorrect predictions are minimized in all the predictions, which allows the state of wear of the tire to be determined. In this configuration, the series of predictions N that identified the same state of wear needs to be large in a series M of significant predictions. In this case, a point symbol is assigned to each measurement. The round “o” symbol represents a measurement, the prediction of which indicates a tire with intermediate wear. The plus “+” symbol represents a measurement for which the prediction indicates a severely worn tire. Finally, the cross “x” symbol represents a measurement for which the prediction indicates a tire in the practically new state.

[0116] FIG. 4 is a spectral representation of the sound power recorded by the microphone during a time frame. The term “time frame” is understood herein to mean the time interval, which is generally short, during which a recording is carried out, on the basis of which the data used as the basis for a measurement are established. This time frame is less than or equal to 0.5 seconds or ideally is less than or equal to 0.25 seconds.

[0117] This spectral representation represents the received sound power (in dB) as a function of the frequency, over a given frequency range, typically in this case, the audible frequency range, ranges between 0 Hz and 20 KHz.

[0118] More specifically, the spectral representation of FIG. 4 is obtained by decomposing the frequency range into frequency bands with predetermined widths, and by assigning a characteristic value to each frequency band that is equal to the average power measured in this frequency band. In this case, dividing the frequency range as one-third octave bands was used. Thus, each point of each of the curves of FIG. 4 represents an average sound power for a frequency band given and measured during a time frame under running conditions, in which, all other things being equal, only the running speed is varied (typically from 30 km/h to 110 km/h).

[0119] It then can be seen that the curves representing the spectral powers are offset relative to one another, and that the total dissipated sound power increases as a function of the speed. Although the general shape of the curves remains similar, which supports the fact that this parameter does not systematically subvert the prediction of the state of wear, certain specific features of the spectrum are more or less marked, which generates noise in the prediction constructed on its spectra. Taking into account the running speed, in particular in terms of a speed category of approximately 30 km/h, allows the prediction of the state of wear of the tire to be improved according to a second embodiment of the invention.

[0120] These observations are reproduced when one or more modalities of the other categories is/are changed and the curves obtained are compared by only varying the speed parameter.

[0121] FIG. 5 is a two-dimensional representation of a series of measurements made on a vehicle for the same tire train, irrespective of its state of wear and irrespective of its operating conditions on a vehicle when traveling on various roads. However, some measurements were taken on wet or damp ground that attest to the activation of the windshield wipers of the vehicle. However, no measurement of the water level on the roads was recorded, these measurements are classified as two unique categories: dry and wet/damp.

[0122] FIG. 5 highlights the efficiency of the third machine learning step of the method for determining the ground surface state, in the form of a category, on the basis of a vibroacoustic temporal measurement. In a discriminant space, in this case in a representation through the single main vector, the measurements are easily classified according to the two aforementioned modalities. The graph represents, for a set of measurements indexed from 1 to 1,290 along the first axis of the graph and stored as the windshield wiper activation indicator, the norm of the main vector along the vertical axis of the graph. The two ground state categories are easily distinguished, forming two groups of measurements, for which the values of the norm of the main vector are significantly differentiated. In addition, a notion of Gaussian distribution of the measurements is found around a central value for each group. Thus, the discriminant power of the vibroacoustic measurement according to the third machine learning is confirmed on a main vector derived from the vibroacoustic measurement on a vehicle. Although some points are aberrant, their limited number easily leads to the meteorological conditions of the ground where the vehicle circulates being determined with low measurement redundancy. The other solution involves not taking into account these aberrant points, since uncertainty remains, and in re-initiating a method for estimating the state of wear of the tire.

[0123] FIG. 6 is a view according to the state of wear of the tire distributed as three categories of identical size, the “new” category, the “half-worn” category and the “worn” category, for vehicle running speeds ranging between 20 and 130 km/h.

[0124] In this case, a set of vibroacoustic measurements carried out on a vehicle is displayed irrespective of the conditions of the vehicle in terms of applied load and inflation pressure starting from the nominal vehicle configuration. The vehicle has followed a road circuit comprising an urban route, a route on a country road and a highway route over several sessions in an intermediate season, which allows all the conditions of the ground surface state to be mixed, in particular the “dry”, “wet” and “damp” ground states and on various types of ground surface nature, in particular “closed”, “medium” and “open” ground surface natures. The learning database takes into account the state of wear of the tire as a modality, but also the running speed category of the vehicle. The measurements are taken in this case with tires that have the three types of state of wear of the tire. However, a condition for carrying out the prediction of the state of wear of the tire is put in place on the ground surface state through the specific ground state category corresponding to the “dry” ground state category. Thus, if the ground surface state does not correspond to the specific ground state category, the prediction of the state of wear of the tire by the machine learning is not carried out. This excludes a number of vibroacoustic measurements but the occurrence of the measurements on a vehicle is sufficient relative to the slow temporal evolution of the wear of the tire.

[0125] Machine learning relative to its conditions mathematically defines main directions. FIG. 6 shows the discriminant space along the two first main directions that represent the two axes of representation of the graph. The machine learning identifies three series of circles in this two-dimensional representation. The first series, represented by dashed circles, represents the various probabilities of the “new” state of wear. The circles are mutually concentric in this two-dimensional discriminant space. The highest probability, i.e., greater than 0.9, is defined by the surface delimited by the smallest circle. Then, the following circle defines a degressive probability of 0.1, that is a value of 0.8. Furthermore, the following circle represents a probability that has further decreased by 0.1, that is a value of 0.7. The second series of circles represented by circles as a grey solid line represents the various probabilities of the “half-worn” state of wear in this two-dimensional discriminant space. Finally, the third series of circles represented by circles as a black solid line in the same way represents the various probabilities of the “worn” state of wear. It should be noted that the three series of circles are generally separate, more than those of FIG. 3, in particular in the vicinity of their small circle, referred to as the main circle, which allows the measurements to be more efficiently sorted according to the three states of wear. However, the circles are big and the secondary circles mutually overlap. As a result, there is uncertainty with respect to the prediction that has been made. It is possible for an incorrect prediction to be made. Thus, by statistically multiplying the measurements, incorrect predictions are minimized in all the predictions, which allows the state of wear of the tire to be determined. In this case, excluding a certain number of vibroacoustic measurements due to the ground surface category makes the method more reliable by minimizing incorrect predictions. In this configuration, the series of predictions N that identified the same state of wear needs to be large in a series M of significant predictions. In this case, a point symbol is assigned to each measurement. The round “o” symbol represents a measurement, the prediction of which indicates a tire with intermediate wear. The plus “+” symbol represents a measurement for which the prediction indicates a severely worn tire. Finally, the cross “x” symbol represents a measurement for which the prediction indicates a tire in the practically new state.

[0126] FIG. 7 is a two-dimensional representation of a series of measurements made on a vehicle for the same tire train, irrespective of its state of wear and irrespective of its operating conditions on a vehicle when traveling on various roads, with this being irrespective of the meteorological conditions. The routes taken are vehicle test tracks, for which the coating texture is regularly monitored by an ATD type measurement. The fourth machine learning, by taking into account only the modalities associated with the three ground nature categories, “open”, “medium” and “closed”, defines main directions characterized by vectors. In this case, the discriminant space is represented using the two main vectors to dimensionally represent the discriminant space.

[0127] The fourth machine learning identifies three series of circles in this two-dimensional representation in association with each of the modalities. The first series, represented by dashed circles, represents the various probabilities of the “medium” ground nature. The circles are mutually concentric in this two-dimensional discriminant space. The highest probability, i.e., greater than 0.9, is defined by the surface delimited by the smallest circle. Then, the following circle defines a degressive probability of 0.1, that is a value of 0.8. Furthermore, the following circle represents a probability that has further decreased by 0.1, that is a value of 0.7. The second series of circles represented by circles as a grey solid line represents the various probabilities of the “closed or smooth” ground nature in this two-dimensional discriminant space. Finally, the third series of circles represented by circles as a black solid line in the same way represents the various probabilities of the “open or macro-rough” ground nature. It should be noted that the three series of circles are generally separate, perfectly between the “open” and “closed” categories, and at least at their smallest circle, called main circle, which allows the measurements to be sorted according to the three ground natures. However, uncertainty remains with respect to the prediction that has been made. It is possible for an incorrect prediction to be made. Thus, by statistically multiplying the measurements, incorrect predictions are minimized in all the predictions, which allows the ground nature to be determined according to the three aforementioned categories. In this case, excluding a certain number of vibroacoustic measurements due to the ground state category makes the method more reliable by minimizing incorrect predictions. In this configuration, the series of predictions N that identified the same state of wear needs to be large in a series M of significant predictions. In this case, a point symbol is assigned to each measurement. The round “o” symbol represents a measurement, the prediction of which indicates a ground surface for which the nature would be of the medium type. The plus “+” symbol represents a measurement for which the prediction indicates a smooth or closed type ground surface. Finally, the cross “x” symbol represents a measurement for which the prediction indicates a macro-rough or open type ground surface.

[0128] FIG. 7 highlights the efficiency of the fourth machine learning of the method for determining the ground nature on the basis of a vibroacoustic temporal measurement. In the discriminant space, in this case in a representation taking into account the two main directions, the measurements are easily classified according to the three aforementioned modalities. The three ground nature categories are identified by the machine learning and are easily distinguished, forming three groups of measurements, for which the values of the norm of the main vectors are significantly differentiated, in particular for the first vector. Thus, the discriminant power of the vibroacoustic measurement according to the fourth machine learning is confirmed by this two-dimensional representation derived from the vibroacoustic measurement on a vehicle. Although some points are aberrant, their limited number easily leads to the texture of the ground where the vehicle circulates being determined with low measurement redundancy.

[0129] FIG. 8 is a view according to the state of wear of the tire distributed as three categories of identical size, the “new” category, the “half-worn” category and the “worn” category, for vehicle running speeds ranging between 20 and 130 km/h.

[0130] In this case, a set of vibroacoustic measurements carried out on a vehicle is displayed irrespective of the conditions of the vehicle in terms of applied load and inflation pressure starting from the nominal vehicle configuration. The vehicle has followed a road circuit comprising an urban route, a route on a country road and a highway route over several sessions in an intermediate season, which allows all the conditions of the ground surface state to be mixed, in particular the “dry”, “wet” and “damp” ground states and on various types of ground surface nature, in particular “closed”, “medium” and “open” ground surface natures. The learning database takes into account the state of wear of the tire as a modality, but also the running speed category of the tire. The measurements are taken in this case with tires that have the three types of state of wear of the tire. However, two conditions for carrying out the prediction of the state of wear of the tire are required before undertaking the prediction, one on the ground surface state and the other on the ground nature. Thus, if the ground surface state does not correspond to the specific ground state category, the prediction on the state of wear of the tire by the machine learning step is not carried out. Similarly, if the ground nature does not correspond to the specific ground nature category, which has an ATD higher than 0.7, the prediction of the state of wear of the tire by the machine learning is not carried out. This excludes a number of vibroacoustic measurements but the occurrence of the measurements on a vehicle is sufficient relative to the slow temporal evolution of the wear of the tire.

[0131] Machine learning relative to its conditions mathematically defines main directions. FIG. 8 shows the discriminant space in the two first main directions that represent the two axes of representation of the graph. The machine learning identifies three series of circles in this two-dimensional representation. The first series, represented by dashed circles, represents the various probabilities of the “new” state of wear. The circles are mutually concentric in this two-dimensional discriminant space. The highest probability, i.e., greater than 0.9, is defined by the inner surface delimited by the smallest circle. Then, the following circle defines a degressive probability of 0.1, that is a value of 0.8. Furthermore, the following circle represents a probability that has further decreased by 0.1, that is a value of 0.7. The second series of circles represented by circles as a grey solid line represents the various probabilities of the “half-worn” state of wear in this two-dimensional discriminant space. Finally, the third series of circles represented by circles as a black solid line in the same way represents the various probabilities of the “worn” state of wear. It should be noted that the three series of circles are generally separate, more than those of FIG. 3 and of FIG. 6, in particular in the vicinity of their small circle, referred to as the main circle, which allows the measurements to be more efficiently sorted according to the three states of wear. As a result, there is low uncertainty with respect to the prediction that has been made. It is possible for an incorrect prediction to be made. Thus, by statistically multiplying the measurements, incorrect predictions are minimized in all the predictions, which allows the state of wear of the tire to be determined. In this case, excluding a certain number of vibroacoustic measurements due to the ground surface category and the ground nature category makes the method more reliable by minimizing incorrect predictions. In this configuration, the series of predictions N that identified the same state of wear needs to be large in a series M of consecutive predictions in order to determine the state of wear. In this case, a point symbol is assigned to each measurement. The round “o” symbol represents a measurement, the prediction of which indicates a tire with intermediate wear. The plus “+” symbol represents a measurement for which the prediction indicates a severely worn tire. Finally, the cross “x” symbol represents a measurement for which the prediction indicates a tire in the practically new state. No incorrect sorting is identified in this example. Finally, selecting “specific” categories in terms of conditions for carrying out the prediction allows the computation times and the computation resources required with machine learning to be limited, taking into account all these categories as a modality of the machine learning. This amounts to minimizing the learning database to only those modalities retained by the machine learning. In addition, this supports the low cost use of the method for estimating the state of wear of the tire on board a vehicle and in real time.

[0132] The applicant has found that taking into account a specific running speed category also improves the quality of the prediction of the state of wear of the tire. However, the gain in terms of prediction quality is counterbalanced by the occurrence of the vibroacoustic measurements at least to define a broad specific running speed category.

[0133] Of course, between the method described in FIG. 3 and the method described in FIG. 8, there are several possible variants relating to the role of a category by alternating both between the selection between a modality of the machine learning or a condition for carrying out the automatic prediction by defining specific categories.