Method for estimating the wear condition of tires

Abstract

10. A method for estimating the state of wear of a tire, comprising the steps of: recording a vibro-acoustic signal (1001) generated by a tire traveling on a road during a time frame; converting the time signal (1001) into a frequency signal (1002) over a frequency range; dividing the frequency range into frequency bands and associating each band with an item of representative data of the frequency signal, the representative data items forming variables of a matrix (1003); predicting the state of wear of the tire from the matrix by machine learning 1004 of data based on a training database formed by a set of matrices each associated with measurements from the same stage under known driving conditions in a manner representative of the state of wear of the tire; and determining the state of wear of the tire (1005) after a number N of identical predictions within a series M of successive predictions.

Description

Technical Field

[0001] The present invention relates to a method for estimating the wear state of a tire mounted on a vehicle traveling on a road based on noise generated when the tire is in dynamic contact with the ground.

Background Art

[0002] It is important to know the wear state of a tire in order to interact with a driver or a driver assistance system and notify them in real time of changes in the grip conditions of the tire and, in particular, changes in the vehicle's ground contact due to changes in the wear state of the tire.

[0003] That is, patent application WO 2017 / 103474 A1 proposes a method that combines both the wear state of a tire and the state of the ground on which the tire is traveling. However, this method, which is possible when mounted on a vehicle, uses a significant number of parameters that are to be managed by discriminant analysis. This large number of parameters results in a computation time and computational resources that make it difficult to implement this method with an acceptable processing cost, regardless of the type of vehicle, particularly a basic vehicle. In addition, the reliability of this method is also affected by indirectly considering the influence of the traveling speed when processing data.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] One of the objectives of the following invention is to propose a method for highlighting the gradual change in tire wear using an autonomous method, that is, without human intervention and without stopping the vehicle. In addition, the method should be easy to implement with reduced cost and computation time, in particular to facilitate real-time processing when post-processing of the data is performed on the vehicle. [Means for solving the problem]

[0006] The present invention relates to a method for estimating the wear condition of a tire mounting assembly of a vehicle traveling on a road surface, and comprises the following steps: - A step of recording measurements of vibroacoustic signals generated by tires traveling on the road surface during a given time frame, - The step of converting a time signal into a frequency signal over a given frequency range. - A step of dividing the frequency range into at least one frequency band having a predetermined width, and associating at least one data representing a frequency signal within this at least one frequency band to at least one frequency band, wherein at least one representative data derived from the measurement value forms at least one variable in the matrix associated with this measurement value. - A step in which machine learning from data based on a learning database consisting of a set of matrices associated with measurements recorded and performed according to the same steps as above under known driving conditions, each representing the state of tire wear, is used to predict the state of tire wear corresponding to the matrix associated with the completed measurement. - A step in determining the tire wear state after N identical predictions within a sequence M of consecutive predictions.

[0007] The term “vibrational acoustic signal” is understood herein to mean any vibration and / or acoustic signal that can be detected by, for example, a microphone or an accelerometer-type sensor.

[0008] This idea involves using a sensor that, for example, does not modify the response of a tire traveling on a road surface when the tire is mounted on a vehicle through a rim. The conditions under which a tire travels on a road generate a specific sign that it is vibrational and / or acoustic within a broad frequency spectrum. This principle involves a step of measuring a time signal over a short period of time and a step of converting this signal into a frequency signal that is more accessible because the useful information is distributed over a broad frequency band. Assuming the signal is pseudo-periodic, it is easy to envision a step of converting a short-time sample with a certain sampling amount into a broadband frequency spectrum with a given frequency pitch. The sampling frequency determines the width of the frequency spectrum. With this spectrum acquired, it is necessary to divide it into a number of frequency bands, each representing a dimension of the matrix associated with the viviacoustic sign. Next, for each band, one or more items of data representing the frequency spectrum on the frequency band are identified. These representative data represent the second dimension of the matrix. These data representing the frequency spectrum are included in a group consisting of the mean, median, maximum, minimum, or a combination of these quantities of the frequency spectrum. These representative data can also be quantities related to the spectral shape on the frequency band, expressed by the frequency relative to the average level on the frequency band or by the features on the frequency band. The division of the matrix into frequency bands can be performed, for example, as one-third octave bands.

[0009] Finally, the tire wear state is predicted by machine learning based on a matrix constructed from a model of matrices acquired using matrices formed in preceding stages during the learning campaign. During this learning campaign, various tire running tests were performed, vibroacoustic measurements were recorded during the run, and then analyzed for each running test according to the same protocol. Each running test was performed on a specific vehicle with a known tire assembly, including tire wear levels under various pressure and static load conditions applied near nominal conditions. Each running test involved a stage in which the vehicle was run on a road where its polymer properties were predetermined and road weather conditions were identified. For this purpose, the tire wear state was divided into at least two groups, called the new group or the worn group, depending on the tire tread height. The tire state is called "new" when the tread height for a tire mounted on a rim and inflated is located between the maximum height and the maximum intermediate height equidistant from the maximum and minimum heights, and vice versa. It is also possible to separate wear conditions into multiple identical size categories. For example, if the intention is to quantify the wear condition of a tire between "new," "partially worn," and "worn" states, the separation limits between these various states are two-thirds of the difference between the maximum and minimum heights, and one-third of the height difference. That is, each category or group is distributed across a similar range of tread heights.

[0010] In this case, numerous measurements and predictions of the tire wear state statistically enable rapid assessment of the tire wear category through a simple model. This simple mathematical model requires only a small learning database and generates an ultrafast response time. That is, it is assumed that, regardless of road weather conditions, road macroscopic unevenness characteristics, or driving speed, the occurrence N of the same prediction with a series of M measurements will enable the detection of a transition of the tire from one wear category to another. In reality, the tire wear state changes slowly in the time domain compared to other sensitive variables of the system. The same applies to the operating conditions of the tire through parameters such as pressure or applied static load. Of course, the series of measurements M will be determined by the diversity of the measurement conditions, and the occurrence N will be controlled by the prediction accuracy under these measurement conditions. That is, unlike models in the prior art literature, it was decided to increase the number of measurements compared to the basic model in order to predict the inherently slowly changing tire state by information redundancy. The simplicity of the model enables fast response times and post-processing on the vehicle. The prediction also improves in terms of reliability because it does not require examining an excessively large set of sensitive parameters for the tire's vibration acoustic signs.

[0011] Preferably, the method comprises the following steps: - The step of determining the tire speed category, wherein the width of the category is a small portion of the maximum speed, preferably the maximum speed is 300 km / h, and the speed category is determined by machine learning.

[0012] In this particular case, the driving speed category is added as a format in the learning database. This allows for a more rapid prediction of tire wear status, as driving speed significantly affects the vibroacoustic response of a tire traveling at an average level, especially when measured over a wide frequency window. This effect is expressed, in particular, by an increase in the tire's acoustic vibration response proportional to driving speed. The sensitivity of representative data can be attenuated or eliminated in vibroacoustic responses controlled at an average level. That is, by considering driving speed as a format, the sensitivity of data representing frequency signals can be increased, improving the prediction of tire wear status. Driving speed can be assessed by data acquired from the vehicle, such as information from GPS (Global Positioning System), items of information transmitted by the vehicle's CANbus, or items of information directly or indirectly contained through tire-mounted electronic systems such as TMS (Tire Mounting Sensors) or TPMS (Tire Pressure Monitoring System), or by other means. To simplify the system and due to the sensitivity of the tire's vibroacoustic response, it is sufficient to classify speeds by speed category, preferably a speed category range of about 10 km / h, which is a good compromise between the desired evaluation speed in terms of the size of the learning database and the desired accuracy regarding the tire wear state. This reduces the learning database for predictions without excessively affecting accuracy. In addition, changes in the tire wear state are gradual. For this reason, the redundancy of information regarding the tire wear state leads to the statistical determination of the tire wear state from a specific number of M vibroacoustic measurements. By adding a driving speed or driving speed category, this process becomes more efficient in terms of the number of vibroacoustic measurements and converges towards better information regarding the tire wear state.

[0013] In a preferred embodiment, the step of determining the vehicle's speed includes the following steps: - A step of recording a second measurement of a vibroacoustic signal generated by a tire traveling on a road surface during a second given time frame, - A step of converting a second time signal into a second frequency signal over a second given frequency range, - A step of dividing a second frequency range into at least one frequency band having a predetermined width, and associating at least one data representing a second frequency signal in this at least one frequency band, wherein at least one representative data derived from a second measurement forms at least one variable in a matrix associated with the second measurement. - A second step of determining the tire speed category corresponding to the matrix associated with a second completed measurement, using data from a learning database consisting of a set of matrices associated with measurements recorded and performed according to the same steps as above under known driving conditions, each in a format that represents the tire speed category.

[0014] Driving speed can be evaluated through vibroacoustic measurements of tires traveling on the road surface. For this purpose, the use of a second machine learning model, built from a learning database that considers only driving speed or driving speed categories in a specific manner, is recommended. Clearly, the vibroacoustic response of a tire depends not only on the tire wear state but also on other parameters. However, driving speed affects the entire frequency spectrum of the tire, not just a certain specific frequency band, or at least certain bands are not all identical according to the observed parameters. That is, the data representing the vibroacoustic sine matrix associated with driving speed is different from the data regarding the tire wear state. In other words, by dividing the frequency signal into specific frequency bands and evaluating specific representative data for each frequency band, it is possible to identify the driving speed or driving speed category corresponding to the vibroacoustic measurement of the tire using a specific machine learning stage. This frequency division and its representative data are essentially different from those associated with the identification matrix of the tire wear state. Of course, this second vibroacoustic measurement may differ from the vibroacoustic measurement of the wear state. However, there is nothing preventing the same measurement from being used in two machine learning stages. In this particular embodiment, it is necessary to identify the driving speed or driving speed category before determining the tire wear state, thereby determining the sequence of all post-processing steps. Finally, by dividing the driving speed into various speed categories, the number of patterns can be limited, making the method more efficient in terms of both the size of the learning database and the response time of post-processing. Unlike prior art documents, which consider a single data point corresponding to the average power across a given frequency band, the method can consider a series of representative data points from the driving speed evaluation, thereby making the method more reliable in terms of accuracy and reliability. That is, the number of measurements required to statistically predict the tire wear state is significantly reduced.

[0015] Advantageously, this method includes the following additional steps: - A step of determining a ground condition category, wherein the ground condition category is in the form of machine learning or a condition for predicting the tire wear state when it is a specific ground condition category.

[0016] In this case, the ground condition category is either a format in a machine learning database associated with the tire wear state, or a condition for performing the step of predicting the tire wear state. In the first case, the ground condition category is added as a format. This has the advantage of statistically facilitating the prediction of the tire wear state by reducing the number of measurements M that must be performed to converge toward useful information. However, the learning database becomes larger because these new formats increase the number of combinations between formats. That is, the size of the learning database increases, and the mathematical model associated with the tire wear state becomes more complex. In the second case, the inventors have found that when the ground condition category is a specific ground condition category, the step of determining the tire wear state does not involve a step of knowing the ground condition category in order to quickly converge toward a solution. That is, the ground condition category becomes a simple indicator for initiating the prediction of the tire wear state. This limits the size of the learning database and the size of the mathematical model of the tire wear state. However, the number of measurements eligible for prediction is limited to only those measurements that satisfy the conditions for a specific ground condition category. This is not disadvantageous when the tire wear state is a parameter that changes slowly over time. Of course, the ground condition category must be determined before attempting to predict tire wear. While not essential, combining the use of the ground condition category with the driving speed category is certainly possible to further limit the number of measurements needed to converge towards information associated with tire wear. Finally, this road condition can be obtained through weather information associated with a map linked to the vehicle's GPS location, or through any other means on the vehicle, such as a rain sensor or a windshield wiper actuator.

[0017] Specifically, the ground state category is included in the group comprising the categories of dry, wet, moist, snowfall, and freezing.

[0018] The state of the ground corresponds to the meteorological conditions of the ground. The meteorological conditions are included in the group comprising dry state, wet state and moist state, or winter states such as snowfall or freezing state. The snowfall state can be included in the group comprising fresh snow state, compacted snow state, granular snow state, and melting snow state in a preferred embodiment.

[0019] The wet state is characterized by the water level at the same plane as the natural roughness of the road surface. This wet state corresponds, for example, to the ground state obtained by a small amount of rainfall or to a dry road after heavy rain. On the other hand, the moist state is characterized by the water level exceeding the level of the natural roughness of the road surface. In practice, the moist state generally corresponds to a water level in the range of 0.5 to 1 millimeter.

[0020] The meteorological conditions of the ground can be determined by the response of sensors on the vehicle such as weather forecasts associated with the GPS position of the vehicle, or the activation of the front windshield wiper or the rainfall detection sensor on the front windshield. Finally, by associating with the ambient temperature of the environment in which the vehicle moves, it is possible to identify the presence or formation of snow associated with the winter state.

[0021] Advantageously, a specific ground state category comprises the "dry" ground state category.

[0022] Among specific weather conditions that prevent taking the ground state category as a machine learning mode related to the wear state of the tire, the "dry" ground state refers to a surface where the contact between the tire and the ground is not affected by the presence of water regardless of its shape and amount. Specific ground state categories can mix the "humid / wet" ground state category and the "dry" ground state category without hindering the prediction of the tire wear state by machine learning, but the inventor has found that the prediction probability is improved by limiting it to a single "dry" ground state category. However, the "humid" ground state category should preferably be avoided.

[0023] Due to this specific state, vibration-acoustic measurement values taken in dry weather can be selected regardless of the ambient temperature, the number of measurement values used to predict the wear state of the tire can be limited, and the method becomes more efficient. In fact, due to the presence of water, a part of the frequency spectrum obtained is modified compared to that obtained using vibration-acoustic measurements in dry weather, which of course generates variations in the frequency spectrum and inevitably affects the quality of the prediction.

[0024] According to a second embodiment, the step of determining the ground state category comprises the following additional steps: - Recording a third measurement value of the vibration-acoustic signal generated by the tire traveling on the road surface during a given third time frame; - Converting the third time signal into a third frequency signal over a third given frequency range; - Dividing the third frequency range into at least one frequency band having a predetermined width and associating at least one data representing the third frequency signal in at least one frequency band with at least one frequency band, and forming at least one variable of a matrix in which at least one representative data derived from the third measurement value is associated with the third measurement value; - A step of determining the ground condition category corresponding to a matrix associated with a third measurement, completed using machine learning from data based on a learning database consisting of a set of matrices associated with measurements recorded and performed according to the same steps as above under known driving conditions, in a format in which each represents at least one ground condition category.

[0025] Road surface weather conditions can be evaluated through vibroacoustic measurements of tires traveling on the road surface. For this purpose, a third machine learning approach is used, constructed from a learning database that considers ground condition categories as a mode. Furthermore, to reduce the size of the learning database for this third machine learning approach, the travel speed category can also be considered as a mode of this third machine learning approach or as a condition for determining ground conditions, thereby making the stage of identifying road surface weather condition categories more efficient. Clearly, the vibroacoustic response of a tire depends not only on the tire wear state but also on other parameters such as travel speed. However, travel speed affects the entire frequency spectrum of the tire, not just a specific frequency band, or at least, not all bands are identical according to the observed parameters. That is, data representing the vibroacoustic sine matrix associated with weather conditions differs from data regarding the tire wear state. In other words, by dividing the frequency signal into specific frequency bands and evaluating specific representative data for each frequency band, it is possible to identify categories associated with road weather conditions corresponding to tire vibroacoustic measurements using specific machine learning. This frequency division and its representative data are essentially different from those associated with the identification matrix for tire wear state and the identification matrix for travel speed. However, by pre-identifying the driving speed category, it becomes possible to reduce the training database for this third machine learning step by taking this parameter into account, making the method more efficient. Of course, this third vibroacoustic measurement may differ from the vibroacoustic measurement of wear condition and the vibroacoustic measurement of driving speed category. However, there is nothing preventing the use of the same vibroacoustic measurement as one of the measurements associated with the other two machine learning steps, thereby reducing the number of steps that need to be performed.

[0026] In this particular embodiment, it is necessary to identify the ground condition category before determining the tire wear state, which requires a specific sequence of post-processing. However, in the third machine learning model, in order to take this parameter into account, the post-processing of the vicoacoustic measurements associated with the ground condition category must be performed after the post-processing of the vicoacoustic measurements for the driving speed category.

[0027] Preferably, the method comprises the following additional steps: - A step of determining a ground characteristic category, wherein the ground characteristic category is a form of machine learning or a condition for predicting the state of tire wear when it is a specific ground characteristic category.

[0028] In this case, the ground characteristic category is either a form of a machine learning database associated with the tire wear state or a condition for performing the step of predicting the tire wear state. In the first case, the ground characteristic category is added as a form. This has the advantage of statistically facilitating the prediction of the tire wear state by reducing the number of measurements that need to be performed to converge towards useful information. However, the learning database becomes larger because these new forms increase the number of combinations between forms. That is, the size of the learning database increases and the mathematical model associated with the tire wear state becomes more complex. In the second case, the inventors have found that when the ground characteristic category is a specific ground characteristic category, the step of determining the tire wear state does not involve a step of knowing the ground characteristic category in order to quickly converge towards a solution. That is, the ground characteristic category becomes a simple indicator for initiating the prediction of the tire wear state. This second case allows for limiting the size of the learning database and the size of the mathematical model of the tire wear state. However, the number of measurements eligible for prediction is reduced to only those measurements that satisfy the conditions regarding a specific ground characteristic category. This is not disadvantageous when tire wear is a parameter that changes slowly over time. Of course, the ground characteristic category must be determined before the stage of predicting the tire wear state. It is certainly possible, but not necessary, to combine the use of the ground characteristic category with the use of the driving speed category or ground condition category in order to further limit the number of measurements required to converge towards information associated with the tire wear state. Finally, this ground characteristic category can be obtained using a ground map associated with the vehicle's GPS position or by any other means such as optical means on the vehicle and / or light and / or sound.

[0029] Specifically, the ground characteristic categories belong to the group that includes the "open," "intermediate," and "closed" categories.

[0030] Pavement is called "closed" or "macroscopically smooth" when it has a smooth appearance and is not rough, such as a concrete slab covered with a layer of asphalt or cement that becomes wet again after being subjected to excessive heat. Pavement is considered "open" or "macroscopically uneven" when it is significantly rough, such as a country road that has been quickly repaired using worn pavement or surface pavement made by spraying gravel onto asphalt. "Intermediate" pavement describes all pavement that is in an intermediate state between the above two states, and more specifically refers to new pavement.

[0031] The macroscopic texture of pavement affects not only the drainage properties of the road but also the acoustic properties of tires. This texture is characterized by the "mean texture depth," also known as "ATD." This ATD is known to those skilled in the art, is also denoted as HSv, and is measured in sand height.

[0032] In other words, various macroscopic textures can be classified as follows: Paving with a closed macroscopic texture has an ATD in the range of 0 to 0.4 mm. Paving with an intermediate macroscopic texture has an ATD in the range of 0.4 to 1.1 mm, and paving with an open macroscopic texture has an ATD greater than 1.1 mm.

[0033] Preferably, a particular ground characteristic category includes a ground characteristic category called "Open".

[0034] Among the specific ground characteristic categories that cannot be taken as machine learning modes associated with tire wear conditions, attention must be paid to the ground characteristic categories characterized by "open" macroscopic textures. In fact, the inventors have found that for these ground characteristics, the frequency spectrum is unique and more readily highlights significant spectral features of various modes, excluding those associated with the ground characteristic category.

[0035] In addition, this specific ground characteristic category allows us to select vibroacoustic measurements taken on fairly open macroscopic textured ground, which limits the number of eligible measurements needed to predict tire wear, making the method more efficient.

[0036] Very preferably, a particular ground characteristic category includes ground having an ATD greater than 0.7, preferably greater than 0.9, and very preferably greater than 1.0.

[0037] However, the inventors have found that this method is also effective when using ground with an ATD of less than 1.1, corresponding to “intermediate” ground. However, only the high range of ATD, such as when the ATD is greater than 0.7 within the ground referred to as “intermediate” ground, enables the efficient implementation of this method. Of course, the higher the ATD, and the closer it is to the “open” ground limit, the more reliable this method becomes.

[0038] In the third embodiment, the step of determining the ground characteristic category includes the following additional steps: - A step of recording a fourth measurement of a vibroacoustic signal generated by a tire traveling on a road surface during a given fourth time frame, - A step of converting a fourth time signal into a fourth frequency signal over a given fourth frequency range, - A step of dividing the fourth frequency range into at least one frequency band having a predetermined width, associating at least one data representing the fourth frequency signal in at least one frequency band with at least one frequency band, and at least one representative data derived from the fourth measurement forms at least one variable in the matrix associated with the fourth measurement, - A step of determining the ground characteristic category corresponding to the matrix associated with a fourth measurement, completed using machine learning from data based on a learning database consisting of a set of matrices associated with measurements recorded and performed under known driving conditions according to the same steps as above, in a format in which each represents at least one ground characteristic category.

[0039] Road surface characteristics can be determined through vibroacoustic measurements of tires traveling on the road surface. For this purpose, the use of a fourth machine learning approach, built from a learning database that considers only ground characteristics in a specific manner, is recommended. Clearly, the vibroacoustic response of a tire depends not only on the tire wear state but also on other parameters such as travel speed and ground weather conditions. However, travel speed affects the entire frequency spectrum of the tire, not just a certain specific frequency band, or at least certain bands are not all identical according to the observed parameters. That is, data representing the vibroacoustic sine matrix associated with travel speed is different from data on the tire wear state, or data on travel speed, or data on ground weather conditions. In other words, by dividing the frequency signal into specific frequency bands and evaluating specific representative data for each frequency band, it is possible to identify ground characteristics or ground characteristic categories corresponding to tire vibroacoustic measurements using specific machine learning approaches. This frequency division and its representative data are essentially different from those associated with the identification matrix for tire wear state, the identification matrix for travel speed, and the identification matrix for ground weather conditions. However, by pre-identifying the driving speed category, it becomes possible to reduce the training database for this fourth machine learning step by taking this parameter into account, making the method more efficient. Of course, this fourth vibroacoustic measurement may differ from the vibroacoustic measurements of wear, driving speed, and ground weather conditions. However, there is nothing preventing the use of the same vibroacoustic measurement as one of the measurements associated with the other three machine learning steps, thereby reducing the number of steps that need to be performed.

[0040] In this particular embodiment, it is necessary to identify the ground characteristic category before determining the tire wear state, which requires a specific order with respect to any post-processing. However, post-processing of vicoacoustic measurements associated with ground characteristic categories can be performed before or after post-processing of vicoacoustic measurements of ground weather conditions. Furthermore, in the fourth machine learning, in order to take this parameter into account, post-processing of vicoacoustic measurements associated with ground characteristic categories may need to be performed after post-processing of vicoacoustic measurements of driving speed categories.

[0041] Optionally, the step of determining the ground characteristic category includes a ground state category as a fourth form of machine learning.

[0042] The inventors have found that by placing the step of determining the ground's meteorological conditions before the step of determining the ground's characteristics, the ground's meteorological conditions can be considered as a fourth form of machine learning associated with the ground's characteristics, and the predictions associated with this fourth form of machine learning can be improved. This makes the step of determining the ground's characteristics more efficient by reducing the number of vibroacoustic measurements that need to be analyzed to identify the ground's characteristics.

[0043] Conveniently, at least one of the machine learning methods falls into the group comprising neural networks, discriminant analysis, support vector machines, boosting methods, K-nearest neighbors, and logistic regression.

[0044] Many machine learning methods exist, each with its own advantages and disadvantages. The list provided is not exhaustive, and these methods are not limited to these specific examples of machine learning.

[0045] Conveniently, the time signal is converted into a frequency signal using power spectral density.

[0046] The inventors have found that using power spectral density to convert time signals to frequency signals has the known advantage of conserving signal energy while simplifying the parameters of the frequency signal to quantities, i.e., amplitude, by eliminating the concept of phase.

[0047] Preferably, the tire wear state includes both a new state and a worn state, and preferably also includes a partially worn state.

[0048] To make this method efficient, it is necessary to define at least two wear states of the tire, thereby allowing the vehicle parameters to be adjusted according to a binary mode. That is, the tire's lifespan is divided into two periods: a new state and a worn state, and the tire's behavior transitions from the first state to the second state, which supports the adaptation of safety devices, such as ABS devices (anti-lock braking systems), according to each of these two states. Of course, it is entirely possible to divide the tire's lifespan into a greater number of tire states. In that case, it is recommended that all of these states correspond to the same change in tread height. However, the more tire treads are divided, the more complicated this method becomes, so the inventors have found that the best performance / efficiency compromise for this method requires dividing the tire tread into three states: worn state, semi-worn state, and new state.

[0049] The present invention will be better understood by reading the following description, which is given merely as a non-limiting embodiment and in all cases refers to the accompanying drawings, the same reference number representing the same part. [Brief explanation of the drawing]

[0050] [Figure 1] This figure shows a general overview of the method for estimating the wear state of a tire according to the present invention. [Figure 2] This figure shows a vehicle equipped with a measuring device that allows for the determination of the tire wear condition. [Figure 3] This figure shows the discrimination space along two principal axes for a method of estimating the state of tire wear by considering only the pattern of tire wear according to the present invention. [Figure 4] This figure shows the noise spectrum measured by a vehicle traveling on a road with known conditions and ground characteristics. [Figure 5] This figure shows the discrimination space along the main axis of the method for determining ground condition categories by considering only the style related to ground conditions according to the present invention. [Figure 6] This figure shows a discriminant space along two main axes for a method of estimating the state of tire wear by considering the type of tire wear and a specific ground condition category, namely the driving speed category for a "dry" condition, according to the present invention. [Figure 7] This figure shows the discrimination space along two main axes of the method for determining ground characteristic categories according to the present invention. [Figure 8] This figure shows a discriminant space along two main axes for a method of estimating the state of tire wear by selecting measurements according to specific ground condition categories and specific ground characteristic category criteria. [Modes for carrying out the invention]

[0051] Figure 1 shows an overall overview of a method for estimating the state of tire wear using vibroacoustic measurements 1001 performed using a vehicle during driving, according to several embodiments.

[0052] This temporal measurement 1001 is converted to a frequency spectrum 1002 using standard computational tools such as the Fast Fourier Transform.

[0053] The frequency spectrum 1002 is then divided into different frequency bands according to the selected application. For each frequency band, one or more physical quantities relating to the reduced frequency spectrum are associated with the frequency spectrum on that frequency band. This set of quantities forms a vector whose length is proportional to the number of physical quantities calculated. This allows us to fill matrix 1003, one of which is the number of frequency bands obtained from the full width of the frequency spectrum. The second dimension of matrix 1003 corresponds to the maximum number of physical quantities evaluated for each selected frequency band. Generally, this matrix is ​​a vector, its length is the number of selected frequency bands, and its second dimension is a scalar dimension. This can also be a two-dimensional matrix, i.e., its second dimension is a vector.

[0054] In the first embodiment, matrix 1003 is introduced into machine learning 1004, which includes a learning database. The learning database is formed during a preliminary stage of learning by a series of vicoacoustic measurements and frequency processing on temporal measurements in which the form of the learning database is dealt with. In a conventional embodiment, the form is a tire wear state category having at least a worn state and a new state, preferably a semi-worn state. The machine learning provides predictions about the tire wear state.

[0055] By repeating measurement, post-processing, and prediction, a series M of predicted results can be formed. By regularly forming a series and repeating similar results, changes in the tire wear state can be confirmed. Similarly, generally, the initial state of tire wear is a new state, which deforms over time, and therefore the predicted series changes to a semi-worn state, and then to a worn state. That is, knowing that changes in the wear state occur only in one direction, the redundancy of predicted results for the same wear state allows for the rapid determination of the actual tire wear state in the form of a wear state category. The more wear categories there are, the higher the accuracy of determining the tire wear state, but the quality of the predicted results is affected by all influential parameters other than the tire wear state, making this method less efficient. For example, in addition to driving speed, ground characteristics, and weather conditions, tire operating conditions related to the vehicle, air pressure, load, and ambient temperature can also be included.

[0056] In the second embodiment, to make the method for estimating the tire wear state more reliable, the prediction by machine learning 1004 can also take into account the tire's travel speed 2001. In fact, this parameter significantly affects the mean level of the frequency spectrum 1002 obtained from the temporal viviacoustic measurement 1001. By considering this parameter in the form of a travel speed category as a form of machine learning, erroneous predictions can be reduced.

[0057] The speed of this tire can be obtained by additional sensors on the vehicle or through information passing through the vehicle's electronic wiring. This determination can be made directly in the form of a category or stored in the form of a speed category preferred by machine learning. However, in a variation, the speed category 2002 is determined using a second vibroacoustic measurement 1001 acquired on the vehicle. It is advantageous that this second vibroacoustic measurement 1001 is the same vibroacoustic measurement 1001 used to predict the tire wear condition in step 1004.

[0058] As before, this temporal viviacoustic measurement 1001 is converted into a frequency spectrum 1002. This frequency spectrum is then divided into frequency bands. One or more physical quantities relating to the frequency spectrum are associated with each frequency band. This allows us to complete the matrix 1003 associated with the predicted driving speed category. However, the division into frequency bands does not need to be the same as that performed to predict the tire wear condition 1004. Identifying the speed category 2002 is generally sufficient to predict the tire wear condition 1004.

[0059] Optionally, to make the method for predicting tire wear more reliable, it is also possible to determine the meteorological surface conditions 3001 of the ground on which the vehicle is driving, and the machine learning prediction 1004 can take these meteorological conditions into account according to two different paths.

[0060] The first approach involves determining ground condition categories 3002 and taking these ground condition categories into account as a form of predicted tire wear 1004. Knowing meteorological ground conditions, at least in terms of categories, allows for more reliable predictions, although this comes at the cost of a larger learning database and more complex mathematical models.

[0061] The second path involves a step in determining a specific ground condition category 3003 in which the prediction of tire wear is performed. In fact, from all vibroacoustic measurements 1001, those corresponding to ground conditions that facilitate the prediction of tire wear 1004 are selected. That is, the learning database associated with the prediction is reduced, and the mathematical model becomes more basic, enabling fast computation time and reliable predictions using limited resources.

[0062] The meteorological surface conditions of the ground are divided into various categories. Under summer or normal conditions, at least dry and wet / moist conditions are distinguished, and the second group may even be divided according to the water level on the road. Similarly, winter conditions such as freezing or snowfall conditions may also be included.

[0063] The applicant found that a specific ground condition category must include a “dry” condition. In fact, this ground condition category is frequently observed in vibroacoustic measurement 2001 and may be more reproducible with respect to the division into frequency band ranges in step 1003. That is, focusing on vibroacoustic measurement values ​​where the ground condition category is specific is not disadvantageous in terms of the frequency of vibroacoustic measurement 1001 in most areas, while being more efficient in terms of prediction due to the similarity of frequency spectra 1002 for these meteorological conditions.

[0064] Of course, taking ground condition categories into consideration in this way may be done in machine learning stage 1004, whether or not the speed 2001 is considered. However, determining the ground condition category 3002 will inevitably occur after determining the speed category 2002, when the intention is to take everything into account.

[0065] Finally, the ground weather conditions 3001 can be obtained by additional sensors on the vehicle, such as a rainfall detector on the windshield or an actuator for the trigger of the windshield wipers, or through information passing through the vehicle's electronic wiring. However, in a variation, the ground condition category 3002 is determined using a third vibroacoustic measurement 1001 obtained on the vehicle. It is advantageous that this third vibroacoustic measurement 1001 is a vibroacoustic measurement 1001 used to predict the tire wear condition 1004, and / or a vibroacoustic measurement 1004 used to determine the driving speed category 2002.

[0066] As before, the temporal viviacoustic measurements 1001 are converted into a frequency spectrum 1002. This frequency spectrum is then divided into frequency bands. One or more physical quantities relating to the frequency spectrum are associated with each frequency band. This allows for the completion of the matrix 1003 associated with determining the ground weather condition category. However, the division into frequency bands does not need to be the same as that performed to predict tire wear 1004 or to determine the driving speed category. In fact, the identification of the ground condition category 3002 is almost sufficient to predict tire wear 1004.

[0067] Optionally, to make the method for predicting tire wear more reliable, it is also possible to determine the texture characteristics 4001 of the ground the vehicle is driving on, and the machine learning prediction 1004 can take the ground characteristic category into account according to two different paths.

[0068] The first path involves determining the ground condition categories 4002 and taking these categories of ground characteristics into account as a form of prediction 1004 of tire wear. Knowing the ground characteristics 4001 at least in terms of categories allows for more reliable predictions, although this has the drawback of making the learning database larger and the mathematical model more complex.

[0069] The second path involves determining a specific ground characteristic category 4003 on which the tire wear prediction 1004 is performed. In practice, from all vibroacoustic measurements 1001, those corresponding to ground characteristics that facilitate the tire wear prediction 1004 are selected. That is, the learning database associated with the prediction is reduced, and the mathematical model becomes more basic, enabling faster computation time and reliable predictions using limited resources.

[0070] Ground characteristic 4001 is divided into various categories according to its roughness on a millimeter scale. This ground characteristic is characterized by ATD.

[0071] The applicant has found that a specific ground characteristic category must include ground referred to as "open" ground. In fact, this category allows for higher reproducibility with respect to the frequency band division of stage 1003. That is, focusing on vibroacoustic measurements where the ground characteristic category is specific is efficient in terms of prediction due to the predetermined similarity of the frequency spectrum 1002. However, it is quite possible to extend the specific ground characteristic category to all ground where the ATD is greater than 0.7, thereby also including the upper end of the ATD for ground referred to as "intermediate" ground.

[0072] Of course, machine learning 1004 can take ground characteristic categories into account whether or not it considers the driving speed 2001 or the surface condition 3001. However, taking ground characteristic categories into account is a necessary step after determining the speed category 2002 when the intention is to take these two parameters into account.

[0073] Finally, the ground characteristics 4001 can be acquired by additional sensors on the vehicle, such as lasers or acoustic measuring devices, or through information passing through the vehicle's electronic wiring. However, in a variation, the ground characteristics category 4002 is determined using a fourth vibroacoustic measurement 1001 acquired on the vehicle. It is advantageous that this fourth vibroacoustic measurement 1001 is the vibroacoustic measurement 1001 used to predict the tire wear condition in step 1004, and / or the vibroacoustic measurement 1001 used to determine the driving speed category 2002, and / or the vibroacoustic measurement 1001 used to determine the ground condition category 3002.

[0074] As before, the temporal viviacoustic measurement 1001 is converted into a frequency spectrum 1002. This frequency spectrum is then divided into frequency bands. One or more physical quantities relating to the frequency spectrum are associated with each frequency band. This allows for the completion of the matrix 1003 associated with the determination of the ground characteristic category 3002 4002. However, the division into frequency bands does not need to be the same as that performed to predict the tire wear condition 1004, or to determine the driving speed category 2002, or to determine the ground characteristic category 3002. In fact, the identification of the ground characteristic category 3002 is almost sufficient to predict the tire wear condition 1004.

[0075] Finally, the applicant found that the prediction of tire wear conditions 1004 can be made more reliable by considering weather conditions, and then ground characteristics. In fact, the ability to separate ground condition categories is stronger than the ability to separate ground characteristic categories.

[0076] In conclusion, predictions can be made more reliable by optionally considering ground condition category 3002 and ground characteristic category 4002. The combination described above is the most efficient configuration. Efficiency is evaluated by the prediction error rate.

[0077] However, by taking a specific categorical path, this method becomes more efficient due to the reduction in the size of the learning database for predicting tire wear conditions, and the speed and simplicity of the calculations that can be performed in real time on the vehicle.

[0078] In the primary embodiment, the acoustic signal generated by the tire (T) is measured using a microphone (1) installed on the vehicle. In Figure 2, the microphone is positioned in front of the wheel housing located at the rear of the vehicle (C). However, other locations, such as the rear bumper, are also conceivable. The choice of microphone location depends, for example, on the type of data to be estimated, the type of vehicle, and external constraints associated with its mounting, maintenance, and durability.

[0079] Figure 2 shows a vehicle C traveling on the ground G, schematically represented with front and rear wheel housings that house wheels fitted with tires T.

[0080] As vehicle C moves, the tires T generate noise whose amplitude and frequency depend on numerous factors. This sound pressure is actually a superposition of noises originating from various sources, such as noise generated by the contact of tread pattern elements with the ground G, the movement of air between tread pattern elements, water particles lifted by the tire, or even airflow associated with the vehicle's speed. Similarly, when these noises are heard, they overlap with noises associated with the vehicle environment, such as engine noise. All of these noises also depend on the vehicle's speed.

[0081] A measuring device, such as microphone 1, is installed in the vehicle. Although only one is shown in Figure 1, it should be noted that the scope of the present invention is not limited to this configuration, and various positions for the measuring device can be envisioned. That is, the sensor can be positioned, for example, on the wall of the rear bumper, but it does not necessarily have to be oriented to detect acoustic signals originating from the rear of the vehicle.

[0082] It is also possible to assume a position on the wall surface of the vehicle's front bumper. Similarly, the measuring means can be positioned on the wheel housing to listen to the road noise as close as possible to the source of the noise. Ideally, installing a vibration acoustic sensor on each wheel housing can be considered the best means of detecting all noise and road vibrations generated by the tires. However, a single microphone is sufficient to determine the ground conditions (weather conditions) and ground characteristics (macroscopic texture of the pavement). In the latter case, it is preferable to isolate it from aerodynamic noise and engine noise.

[0083] Of course, operational precautions are taken to protect the measuring instruments from external aggression such as splashes of water, mud, or gravel.

[0084] The vehicle also includes a computer 2 configured to be connected to a measuring means and to perform calculations for shaping and analyzing the raw information derived from the measuring means, as detailed below, and to estimate the tire condition as a function of the measured values ​​of vibroacoustic emissions detected by the measuring means.

[0085] Figure 3 shows that tire wear is distributed into three categories for the same tire size at vehicle speeds ranging from 20 to 130 km / h. The first category is the "new" wear category, which corresponds to the upper third of the useful height of the tire tread. The useful height is determined by the maximum height of the tread corresponding to the radial outer edge with respect to the tire's natural axis of rotation, and the minimum height is determined by the radial outer edge of the wear indicator at the bottom of the groove. The second category, called the "partially worn" category, corresponds to the middle third of the useful height of the tire tread. Finally, the last category, called the "worn" category, corresponds to the lower third of the useful height of the tread.

[0086] This section presents a set of vibroacoustic measurements performed on a vehicle, starting from the nominal vehicle configuration and regardless of vehicle conditions regarding load and air pressure. The vehicle traversed a road circuit including urban, rural, and main road routes several times during the intermediate seasons, thereby mixing all ground conditions, particularly "dry," "wet," and "damp" ground conditions, with various ground characteristics, especially "closed," "intermediate," and "open" ground characteristics. The learning database considers tire wear conditions as a format. In this case, measurements are performed on tires with three types of wear conditions.

[0087] Machine learning mathematically determines principal directions in relation to its conditions. Figure 3 shows the discrimination space with two first principal directions representing the two axes of the graph display. Machine learning identifies three series of circles in this two-dimensional display. The first series, represented by dashed circles, represents the various probabilities of the "new" wear state in this two-dimensional discrimination space. The circles are concentric. The highest probability, i.e., the probability greater than 0.9, is determined by the plane defined by the smallest circle. Next, the next circle sets a probability of a decrease of 0.1, i.e., a value of 0.8. Furthermore, the next circle represents a probability of a further decrease of 0.1, i.e., a value of 0.7. The second series of circles, represented by solid gray lines, represents the various probabilities of the "partially worn" wear state in this two-dimensional discrimination space. Finally, the third series of circles, similarly represented by solid black lines, represents the various probabilities of the "worn" wear state. The three series of circles are largely separate, especially near the smaller circles called the principal circles, allowing the measurements to be classified according to three wear states. However, the circles are large, and the secondary circles overlap each other. As a result, uncertainty exists regarding the predictions made. False predictions can be made. That is, by statistically multiplying the measurements, the false predictions can be minimized within all predictions, thereby determining the tire's wear state. In this configuration, the series N of predictions that identify the same wear state must be large within the series M of significant predictions. In this case, each measurement is assigned a dot symbol. A round "o" symbol represents a measurement that predicts a tire with intermediate wear. A plus "+" symbol represents a measurement that predicts a heavily worn tire. Finally, a cross "x" symbol represents a measurement that predicts a tire that is virtually new.

[0088] Figure 4 shows the spectral display of the acoustic power recorded by the microphone during a time frame. In this specification, the term “time frame” is generally understood to mean a short time interval during which recordings are made, and the data used as the basis for the measurement is established accordingly. This time frame is 0.5 seconds or less, or ideally 0.25 seconds or less.

[0089] This spectral representation shows the received acoustic power (in dB) as a function of frequency over a given frequency range, typically the audible frequency range in this case, i.e., from 0 Hz to 20 kHz.

[0090] More specifically, the spectral representation in Figure 4 is obtained by decomposing the frequency range into frequency bands with predetermined widths, and by assigning characteristic values ​​equal to the average power measured in these frequency bands to each frequency band. In this case, dividing the frequency range into one-third octave bands was used. That is, each point on each curve in Figure 4 represents the average acoustic power measured over a time frame under driving conditions with respect to a given frequency band, where all other conditions are equal and only the driving speed varies (typically from 30 km / h to 110 km / h).

[0091] Next, the curves representing spectral power are offset from one another, and it can be seen that the total dissipated acoustic power increases as a function of velocity. The schematic shapes of the curves remain similar, supporting the fact that this parameter does not systematically overturn the prediction of wear conditions, however certain specific features of the spectrum are more or less prominent, which introduces noise into the predictions built on that spectrum. By considering the driving speed, particularly the driving speed category of about 30 km / h, the prediction of tire wear conditions can be improved according to a second embodiment of the present invention.

[0092] These observations are reproducible when one or more modes related to other categories are modified, and the resulting curves are compared by changing only the velocity parameter.

[0093] Figure 5 is a two-dimensional representation of a series of measurements taken on a vehicle for the same tire row, regardless of its wear condition and the driving conditions on various roads. However, as indicated by the activation of the vehicle's windshield wipers, some measurements were taken on wet or damp ground. However, measurements regarding water levels on the road were not recorded, and these measurements are classified into two distinct categories: dry and wet / damp.

[0094] Figure 5 highlights the efficiency of the third machine learning stage of a method for determining ground conditions in category form based on vibroacoustic time measurements. In the discriminative space, in this case represented by a single principal vector, the measurements are easily classified according to the two modes described above. The graph is ticked from 1 to 1290 along the first axis of the graph and represents the norm of the principal vector along the longitudinal axis of the graph for a set of measurements stored as an operating indicator for the windshield wiper. Two ground condition categories are easily distinguishable, forming two groups of measurements where the norm values ​​of the principal vectors are clearly identified. Furthermore, the concept of a Gaussian distribution of measurements can be found around the center value of each group. That is, the discriminative function of the vibroacoustic measurements by the third machine learning stage is confirmed for the principal vector derived from the vibroacoustic measurements of the vehicle. Although there are anomalies, their number is limited, resulting in low measurement redundancy and easy determination of the weather conditions of the ground the vehicle is circling. Another solution involves not considering these anomalies, as uncertainty remains, and restarting the method for estimating the state of tire wear.

[0095] Figure 6 shows the wear status of tires distributed across three identical size categories—"new," "partially worn," and "worn"—for vehicle speeds ranging from 20 to 130 km / h.

[0096] In this case, starting from the nominal vehicle configuration, regardless of the vehicle's conditions regarding load and tire pressure, a set of vibroacoustic measurements performed on the vehicle is displayed. This vehicle travels a road circuit including urban routes, rural routes, and main road routes several times during the intermediate season, thereby mixing all ground conditions, especially "dry," "wet," and "humid" ground conditions, with various types of ground characteristics, especially "closed," "intermediate," and "open" ground characteristics. The learning database considers not only the tire wear state but also the vehicle's travel speed category as a format. In this case, measurements are performed on tires with three types of wear states. However, the conditions for performing predictions of tire wear state are appropriately attributed to the ground condition through a specific ground condition category corresponding to the "dry" ground condition category. That is, when the ground condition does not correspond to a specific ground condition category, machine learning does not perform predictions of tire wear state. For this purpose, a large number of vibroacoustic measurements are excluded, but the occurrence of measurements related to the vehicle in relation to the gradual temporal progression of tire wear is sufficient.

[0097] Machine learning mathematically determines principal directions in relation to its conditions. Figure 6 shows the discriminant space along two first principal directions representing the two axes of the graph display. Machine learning identifies three series of circles in this two-dimensional display. The first series, represented by dashed circles, represents various probabilities of the "new" wear condition. In this two-dimensional discriminant space, the circles are concentric with each other. The highest probability, i.e., the probability greater than 0.9, is determined by the plane defined by the smallest circle. Next, the next circle defines a probability of a decrease of 0.1, i.e., a value of 0.8. Furthermore, the next circle represents a probability of a further decrease of 0.1, i.e., a value of 0.7. The second series of circles, represented by solid gray lines, represents various probabilities of the "partially worn" wear condition in this two-dimensional discriminant space. Finally, the third series of circles, similarly represented by solid black lines, represents various probabilities of the "worn" wear condition. The series of three circles is more generally separated than in Figure 3, particularly near the smaller circles called the principal circles, allowing for more efficient classification of measurements according to the three wear states. However, the circles are large, and the secondary circles overlap each other. As a result, uncertainty exists regarding the predictions made. False predictions may be made. That is, by statistically multiplying the measurements, the false predictions can be minimized within all predictions, thereby determining the tire wear state. In this case, the method becomes more reliable by minimizing false predictions by excluding a certain number of vicoacoustic measurements attributable to the ground category. In this configuration, the series N of predictions that identify the same wear state must be large within the series M of significant predictions. In this case, each measurement is assigned a dot symbol. A round "o" symbol represents a measurement that predicts a tire with intermediate wear. A plus "+" symbol represents a measurement that predicts a heavily worn tire. Finally, a cross "x" symbol represents a measurement that predicts a tire that is virtually new.

[0098] Figure 7 is a two-dimensional representation of a series of measurements taken on a vehicle for the same tire row, regardless of its wear condition, driving conditions on various roads, and weather conditions. The route is a test course for a vehicle where the pavement texture is periodically monitored by ATD-type measurements. The fourth machine learning approach defines the vector-characterized principal directions by considering only the patterns associated with the three ground characteristic categories: "open," "intermediate," and "closed." In this case, two principal vectors are used to represent the discriminant space dimensionally.

[0099] The fourth machine learning method identifies three series of circles in relation to each style in this two-dimensional representation. The first series, represented by dashed circles, represents the various probabilities of “intermediate” ground characteristics. In this two-dimensional discriminative space, the circles are concentric with each other. The highest probability, i.e., the probability greater than 0.9, is determined by the face defined by the smallest circle. Next, the following circle sets a probability that decreases by 0.1, i.e., a value of 0.8. Furthermore, the next circle represents a probability that decreases by another 0.1, i.e., a value of 0.7. The second series of circles, represented by solid gray lines, represents the various probabilities of “closed or smooth” ground characteristics in this two-dimensional discriminative space. Finally, the third series of circles, similarly represented by solid black lines, represents the various probabilities of “open or macroscopic uneven” ground characteristics. It should be noted that the three series of circles are almost separated, and are completely separated between “open” and “closed,” and at least in their smallest circle, called the principal circle, so that the measurements can be classified according to the three ground characteristics. However, uncertainty remains regarding the predictions made. Incorrect predictions may occur. That is, by statistically multiplying the measurements, incorrect predictions can be minimized within the total number of predictions, thereby determining the tire wear state according to the three categories mentioned above. In this case, the method becomes more reliable by minimizing incorrect predictions by excluding a certain number of vicoacoustic measurements attributable to the ground characteristic category. In this configuration, the series N of predictions identifying the same wear state must be large within the series M of significant predictions. In this case, each measurement is assigned a dot symbol. A round "o" symbol represents a measurement indicating a ground with an intermediate type of characteristic. A plus "+" symbol represents a measurement indicating a smooth or closed ground. Finally, a cross "x" symbol represents a measurement indicating a macroscopic uneven or open type of ground.

[0100] Figure 7 highlights the efficiency of the fourth machine learning method for determining ground characteristics based on vibroacoustic time measurements. In the discriminative space, in this case a representation considering two principal vectors, the measurements are easily classified according to the three modes described above. Three ground characteristic categories are identified by machine learning and are easily distinguished, with the norm values ​​of the principal vectors clearly distinguishing them, particularly with respect to the first vector, forming three sets of measurements. That is, this two-dimensional representation derived from the vehicle's vibroacoustic measurements confirms the discriminative function of the vibroacoustic measurements by the fourth machine learning method. Although there are some anomalies, their number is limited, resulting in low measurement redundancy and easy determination of the ground texture over which the vehicle travels.

[0101] Figure 8 shows the wear status of tires distributed across three identical size categories—"new," "partially worn," and "worn"—for vehicle speeds ranging from 20 to 130 km / h.

[0102] In this case, starting from the nominal vehicle configuration, regardless of the vehicle's conditions regarding load and air pressure, a set of vibroacoustic measurements performed on the vehicle is displayed. This vehicle travels a road circuit including urban routes, rural routes, and main road routes several times during the intermediate season, thereby mixing all ground conditions, particularly "dry," "wet," and "damp" ground conditions, with various types of ground characteristics, particularly "closed," "intermediate," and "open" ground characteristics. The learning database considers not only the tire wear state but also the vehicle's travel speed category as a format. In this case, measurements are performed on tires with three types of wear states. However, in order to perform tire wear state prediction, two conditions are required before the prediction can be made: one relating to the ground condition and the other to the ground characteristics. That is, when the ground condition does not correspond to a specific ground condition category, machine learning prediction of tire wear state is not performed. Similarly, when the ground characteristics do not correspond to a specific ground characteristic category where ATD exceeds 0.7, machine learning prediction of tire wear state is not performed. For this purpose, a large number of vibroacoustic measurements are excluded, but there are still enough vehicle-related measurements to occur in relation to the gradual temporal progression of tire wear.

[0103] Machine learning mathematically determines principal directions in relation to the conditions. Figure 8 shows the discrimination space with two first principal directions representing the two axes of the graph display. Machine learning identifies three series of circles in this two-dimensional display. The first series, represented by dashed circles, represents various probabilities of the "new" wear condition. In this two-dimensional discrimination space, the circles are concentric with each other. The highest probability, i.e., the probability greater than 0.9, is determined by the plane defined by the smallest circle. Next, the next circle defines a probability of a decrease of 0.1, i.e., a value of 0.8. Furthermore, the next circle represents a probability of a further decrease of 0.1, i.e., a value of 0.7. The second series of circles, represented by solid gray lines, represents various probabilities of the "partially worn" wear condition in this two-dimensional discrimination space. Finally, the third series of circles, similarly represented by solid black lines, represents various probabilities of the "worn" wear condition. The series of three circles is more generally separated than in Figures 3 and 6, particularly near the smaller circles called the principal circles, allowing for more efficient classification of measurements according to the three wear states. As a result, the uncertainty regarding the predictions made is small. There is a possibility of incorrect predictions being made. That is, by statistically multiplying the measurements, the incorrect predictions can be minimized within all predictions, thereby determining the tire wear state. In this case, the method becomes more reliable by minimizing the incorrect predictions by excluding a certain number of vibroacoustic measurements attributable to the ground condition category. In this configuration, the series N of predictions identifying the same wear state must be large within the series M of consecutive predictions. In this case, each measurement is assigned a dot symbol. A round "o" symbol represents a measurement that indicates the prediction of a tire with intermediate wear. A plus "+" symbol represents a measurement that indicates the prediction of a heavily worn tire. Finally, a cross "x" symbol represents a measurement that indicates the prediction of a tire in virtually new condition. In this example, no incorrect classifications are observed. Finally, by selecting "specific" categories in terms of the conditions under which predictions are made, it is possible to limit the computation time and resources required for machine learning while considering all of these categories as forms of machine learning.This corresponds to minimizing the learning database to only the format that can be maintained by machine learning. Furthermore, this helps to utilize a low-cost method for estimating tire wear status on a vehicle in real time.

[0104] The applicant found that considering specific driving speed categories also improves the quality of predictions regarding tire wear. However, the benefit in terms of prediction quality is offset by the generation of vibroacoustic measurements required to define at least a broad range of specific driving speed categories.

[0105] Of course, there are several possible variations regarding the role of categories between the method described in Figure 3 and the method described in Figure 8, by alternating between selecting between different machine learning styles and defining specific categories to perform automated predictions. [Explanation of symbols]

[0106] 1001 Vibration and Acoustic Measurements 1002 Frequency Spectrum 1003 Queue 1004 Machine Learning 1005 Tire wear condition

Claims

1. A method for estimating the wear state of the tires of a vehicle mounted on a road surface, A step of measuring a vibration acoustic signal (1001) generated by the tire traveling on the road surface during a given time frame, The steps include converting the vibration acoustic signal (1001) into a frequency signal (1002) over a given frequency range, The steps of dividing the frequency range into at least one frequency band having a predetermined width, and associating at least one data representing the frequency signal within the at least one frequency band with the at least one frequency band, wherein the at least one data derived from the measurement value forms at least one variable in the matrix (1003) associated with the measurement value, A step of predicting the tire wear state corresponding to the matrix associated with the measured value, using machine learning (1004) from the data, which is based on a learning database consisting of a set of matrices associated with measurements recorded and performed according to the steps of: measuring the vibration acoustic signal (1001) under known driving conditions according to a format in which each represents the wear state of the tire; transforming the vibration acoustic signal (1001); and dividing the frequency range and associating the at least one data; A step of determining a ground condition category (3002), wherein the ground condition category (3002) is a specific ground condition category (3003) that includes the "dry" ground condition category or the "dry" and "wet" ground condition categories, and the machine learning (1004) is used to predict the wear state of the tire, A step of determining a ground characteristic category (4002), wherein if the ground characteristic category (4002) is a specific ground characteristic category (4003) indicating that the ground is relatively rough, the machine learning (1004) determines the ground characteristic category (4002), which is a condition for predicting the wear state of the tire. In the step of predicting the wear state of the tire over multiple consecutive cycles, when the number of times the wear state of the tire is predicted to be the same reaches a predetermined value, the step of determining the wear state (1005) of the tire, A method characterized by comprising:

2. A step of determining the tire's running speed category (2002), wherein the running speed category (2002) is obtained by dividing the maximum running speed into a plurality of ranges, and the running speed category (2002) is in the form of the machine learning (1004), A method for estimating the wear state of a tire according to claim 1, characterized by comprising the following:

3. The step of determining the tire's speed category (2002) is: A step of recording a second measurement of a second vibroacoustic signal (1001) generated by the tire running on the road surface during a second given time frame, The steps include converting the second vibration acoustic signal (1001) into a second frequency signal (1002) over a given second frequency range, The steps of dividing the second frequency range into at least one frequency band having a predetermined width, and associating at least one data representing the second frequency signal within the at least one frequency band with the at least one frequency band, wherein the at least one data derived from the second measurement value forms at least one variable in the matrix (1003) associated with the second measurement value, A second machine learning (2004) from the data, based on a learning database comprising a set of matrices associated with measurements recorded and performed according to the steps of: recording the second measurement under known driving conditions in a manner that each represents a driving speed category of the tire; converting the second vibroacoustic signal (1001); and dividing the second frequency range and associating the at least one data, thereby determining the driving speed category (2002) of the tire corresponding to the matrix associated with the second measurement; Equipped with, The method for estimating the wear state of a tire according to feature 2.

4. The method for estimating the wear state of a tire according to any one of claims 1 to 3, characterized in that the ground condition category (3002) is included in the group comprising the categories of dry, wet, damp, snowy, and frozen.

5. The step of determining the aforementioned ground condition category (3002) involves the following additional steps: A step of recording a third measurement of a third vibroacoustic signal (1001) generated by the tire running on the road surface during a given third time frame, The steps include converting the third vibration acoustic signal (1001) into a third frequency signal (1002) over a given third frequency range, An additional step of dividing the third frequency range into at least one frequency band having a predetermined width, and associating at least one data representing the third frequency signal within the at least one frequency band with the at least one frequency band, wherein the at least one data derived from the third measurement forms the at least one variable of the matrix (1003) associated with the third measurement, A third machine learning (3004) from the data, based on a learning database comprising a set of matrices associated with the measurements recorded and performed according to the steps of: recording the third measurement under known driving conditions in a manner in which each represents at least one ground condition category; transforming the third vibroacoustic signal (1001); and dividing the third frequency range and associating the at least one data, to determine the ground condition category (3002) corresponding to the matrix associated with the third measurement; Equipped with, A method for estimating the wear state of a tire according to any one of claims 1 to 4.

6. The method for estimating the wear state of a tire according to any one of claims 1 to 5, wherein the ground characteristic category (4002) includes an open category, an intermediate category, and a closed category, the open category indicating that the ground is relatively rough, the closed category indicating that the ground is not relatively rough, and the intermediate category indicating that the roughness of the ground is in an intermediate state between the open category and the closed category.

7. The method for estimating the wear state of a tire according to any one of claims 1 to 6, characterized in that the specific ground characteristic category (4003) includes ground having an average texture depth greater than 0.

7.

8. The step of determining the aforementioned ground characteristic category (4002) involves the following additional steps: A step of recording a fourth measurement of a fourth vibration-acoustic signal (1001) generated by the tire running on the road surface during a given fourth time frame, The steps include converting the fourth vibration acoustic signal (1001) into a fourth frequency signal (1002) over a given fourth frequency range, An additional step of dividing the fourth frequency range into at least one frequency band having a predetermined width, and associating at least one data representing the fourth frequency signal within the at least one frequency band with the at least one frequency band, wherein the at least one data derived from the fourth measurement forms the at least one variable of the matrix (1003) associated with the fourth measurement, A fourth machine learning (4004) from the data, based on a learning database comprising a set of matrices associated with measurements recorded and performed according to the steps of: recording the fourth measurement under known operating conditions in a manner in which each represents at least one ground characteristic category; transforming the fourth vibroacoustic signal (1001); and dividing the fourth frequency range and associating the at least one data, to determine the ground characteristic category (4002) corresponding to the matrix associated with the fourth measurement; Equipped with, A method for estimating the wear state of a tire according to any one of claims 1 to 7.

9. The method for estimating the tire wear state according to claim 8, characterized in that the step of determining the ground characteristic category comprises the ground condition category (3002) as the fourth form of machine learning.

10. A method for estimating the wear state of a tire according to any one of claims 1 to 9, characterized in that at least one of the machine learning methods (1004, 2004, 3004, 4004) is included in the group comprising a neural network, discriminant analysis, support vector machine, boost method, K-nearest neighbor method, and logistic regression.

11. A method for estimating the wear state of a tire according to any one of claims 1 to 10, characterized in that the wear state of the tire is included in the group comprising new tires and worn tires.

Images