Road surface condition determination method and road surface condition determination device

Abstract

When determining the state of the road surface contacted by the tire based on the time-varying waveform of the vibration of the moving tire detected by a vibration detection unit, an empirical mode decomposition algorithm is used to obtain multiple natural vibration modes based on the data of the time-varying waveform of the vibration of the tire. Thereafter, an arbitrary natural vibration mode is selected from these multiple natural vibration modes, and a statistic is calculated based on the distribution of characteristic data calculated by performing a Hilbert transform on the selected natural vibration mode. The statistic is set as a characteristic quantity, and the road surface condition is determined based on the characteristic quantity and the characteristic quantity calculated in advance for each road surface condition.

Description

[0001] This application is a divisional application of the invention patent application with application number 201880031455.6 (PCT / JP2018 / 017082), application date April 26, 2018, and invention name “Road surface condition determination method and road surface condition determination device”. Technical Field

[0002] The invention relates to a method and a device for judging the state of a road surface on which a vehicle is traveling. Background Art

[0003] In the past, as a method for determining the road surface condition using only the time series waveform data of the tire vibration during driving, the following method has been proposed (for example, refer to Patent Document 1): the road surface condition is determined using a function such as a GA kernel, and the GA kernel or the like is calculated based on a feature vector of each time window calculated based on a time series waveform extracted by multiplying the time series waveform of the tire vibration by a window function, that is, a vibration level in a specific frequency band, and a road surface feature vector of each time window calculated based on the time series waveform of the tire vibration, which is calculated in advance for each road surface condition.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Publication No. 2014-35279 Summary of the invention

[0007] Problem that the invention aims to solve

[0008] However, the conventional method has the following problems: since the time series waveform is time-stretched, not only does the calculation of the GA kernel take time, but also the amount of data is large, so the processing is very cumbersome.

[0009] Time expansion is a process required to compare the acquired tire vibration waveforms (acceleration waveforms). For example, when the vehicle is traveling at a speed of 30 km / h, and data of one circumference of the tire (circumference 2m) is acquired at a sampling rate of 10kHz, the number of measurement points is 2400, while when the vehicle is traveling at a speed of 90 km / h, the number of measurement points becomes 800. Therefore, it is difficult to simply compare the waveforms, and the waveforms need to be expanded and contracted on the time axis.

[0010] This time stretching operation is the main reason why the amount of computation does not decrease.

[0011] The present invention has been made in view of the conventional problems, and an object of the present invention is to provide a method and a device thereof that can significantly reduce the amount of calculation and thereby can quickly and accurately determine the state of a road surface.

[0012] Solutions for solving problems

[0013] The present invention is a road surface condition determination method, which determines the state of the road surface contacted by the tire based on the time-varying waveform of the vibration of the running tire detected by a vibration detection unit, and the road surface condition determination method is characterized in that it includes the following steps: detecting the time-varying waveform of the vibration of the tire; using an empirical mode decomposition algorithm to obtain multiple natural vibration modes based on the data of the time-varying waveform; selecting and extracting any natural vibration mode from the multiple natural vibration modes; performing a Hilbert transform on the extracted natural vibration mode to calculate characteristic data such as instantaneous frequency and instantaneous amplitude; calculating a characteristic quantity based on the distribution of the characteristic data; and determining the road surface condition based on the calculated characteristic quantity and the characteristic quantity pre-calculated for each road surface condition, wherein the characteristic quantity is a statistical quantity such as the mean, standard deviation, skewness, kurtosis, etc. of the distribution of the characteristic data.

[0014] In addition, the present invention is a road surface state determination device, which detects the vibration of a running tire to determine the state of the road surface contacted by the tire, and the road surface state determination device is characterized in that it comprises: a vibration detection unit, which is installed on the tire and detects the time-varying waveform of the vibration of the running tire; an inherent vibration mode extraction unit, which uses an empirical mode decomposition algorithm to obtain multiple inherent vibration modes based on the time-varying waveform, and extracts any inherent vibration mode from the acquired multiple inherent vibration modes; a feature data calculation unit, which performs Hilbert transform on the extracted inherent vibration mode to calculate feature data; and a feature quantity calculation unit, which A characteristic quantity is calculated based on the distribution of the characteristic data; a storage unit, which stores the characteristic quantity calculated in advance for each road surface condition using the time-varying waveform of the vibration; a kernel function calculation unit, which calculates a Gaussian kernel function based on the calculated characteristic quantity and the characteristic quantity calculated in advance for each road surface condition; and a road surface condition judgment unit, which judges the road surface condition based on the value of an identification function using the calculated Gaussian kernel function, wherein the characteristic quantity is a statistical quantity such as an average, standard deviation, skewness, kurtosis, etc. of the distribution of the characteristic data, and the road surface condition judgment unit compares the values ​​of the identification function calculated for each road surface condition to judge the road surface condition.

[0015] Furthermore, the described inventive summary does not list all the features required for the present invention, and sub-combinations of these feature groups may also form inventions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Detailed Description of the Invention This is a functional block diagram showing the configuration of the road surface condition determination device according to the present embodiment.

[0017] Figure 2 It is a diagram showing an example of the mounting position of the acceleration sensor.

[0018] Figure 3 FIG. 1 is a diagram showing an example of a time series waveform of tire vibration.

[0019] Figure 4 It is a diagram showing a method of acquiring a natural vibration mode.

[0020] Figure 5 It is a diagram showing a method of acquiring feature data.

[0021] Figure 6 It is a diagram showing an example of a distribution state of feature quantities.

[0022] Figure 7 is a schematic diagram showing the separating hyperplanes in the input space and feature space.

[0023] Figure 8 This is a flowchart showing a method for determining a road surface state according to the present embodiment.

[0024] Fig. 9 This is a graph comparing the road surface recognition accuracy of the conventional method and the present method.

[0025] Fig.10 It is a graph comparing the learning time of the previous method and the present method.

[0026] Fig.11 This is a diagram showing a boundary surface for distinguishing between two types of road surfaces: dry (DRY) and wet (WET).

[0027] Fig.12 This is a graph comparing the dry / wet road discrimination accuracy of the conventional method and the present method. DETAILED DESCRIPTION

[0028] Figure 1 : is a functional block diagram showing the structure of the road surface state determination device 10 .

[0029] The road surface condition determination device 10 includes an acceleration sensor 11 as a tire vibration detection unit, a vibration waveform detection unit 12, an inherent vibration mode extraction unit 13, a feature data calculation unit 14, a feature quantity calculation unit 15, a storage unit 16, a kernel function calculation unit 17 and a road surface condition determination unit 18.

[0030] Each of the vibration waveform detection unit 12 to the road surface state determination unit 18 is constituted by, for example, computer software and a memory such as a RAM.

[0031] like Figure 2 As shown, the acceleration sensor 11 is integrally arranged at the substantially central portion of the tire air chamber 22 side of the inner liner portion 21 of the tire 20, and detects the vibration of the tire 20 caused by the input from the road surface. The tire vibration signal as the output of the acceleration sensor 11 is amplified by an amplifier, for example, and then converted into a digital signal and sent to the vibration waveform detection unit 12.

[0032] The vibration waveform detection unit 12 extracts an acceleration waveform, which is a time-series waveform of the tire vibration, from the tire vibration signal detected by the acceleration sensor 11 every time the tire rotates one revolution.

[0033] Figure 3 The diagram shows an example of a time series waveform of tire vibration. The time series waveform of tire vibration has large peaks near the stepping-in position and the stepping-out position, and has a peak in the pre-stepping-in region R before the contact portion of the tire 20 touches the ground. f and the post-kick-out region R after the contact portion of the tire 20 leaves the road surface. k In both cases, different vibrations occur depending on the road surface conditions. f The area before and after kick-off R k The subsequent area (hereinafter referred to as the off-road surface area) is hardly affected by the road surface, so the vibration level is also small and does not contain information about the road surface.

[0034] In addition, as a definition of the off-road area, for example, a background level is set with respect to the acceleration waveform, and an area having a vibration level smaller than the background level may be defined as the off-road area.

[0035] In this example, the area containing the road surface information in the acceleration waveform, that is, the road surface area (pre-stepping area R f , touchdown area R s And the area after kicking out R k ) is set as the acceleration waveform of the measured data x 1 (t), the measured data x is transformed into 1 (t) Decompose into multiple intrinsic vibration modes (Intrinsic Mode Function; IMF), and then perform Hilbert transform on each IMF to calculate the feature value.

[0036] The natural vibration mode extraction unit 13 uses the EMD algorithm to extract the natural vibration mode according to the measurement data x 1 (t) Obtain multiple IMFs (C 1 , C 2 ,······,C n ), and extract any IMF C from the obtained multiple IMFs k .

[0037] Here, the method of obtaining the IMF is explained.

[0038] First, if Figure 4 As shown, extract the measurement data x 1 All the maximum and minimum points of (t) are connected, and the upper envelope e is obtained by connecting the maximum points. max (t) and the lower envelope e formed by connecting the minimum points min (t), then calculate the upper envelope e max (t) and the lower envelope e min The local average m of (t) 1 (t)=(e max (t)+e min (t)) / 2.

[0039] Next, find the measured data x 1 (t) and the local average m 1 (t) difference waveform y 1 (t) = x 1 (t)-m 1 (t). Difference waveform y 1 (t) lacks symmetry and cannot be considered an IMF. 1 (t) and the measured data x 1 (t) The same process as before is performed to obtain the difference waveform y 2 Then, the process is repeated to obtain the difference waveform y 3 (t), y 4 (t),······,y m (t). Difference waveform y k The larger the k of (t), the higher the symmetry is, and thus the closer it is to the IMF.

[0040] As a condition for the difference waveform to become IMF, y is proposed k The number of zero crossing points and the number of peaks of (t) remain unchanged for 4 to 8 times in the process of finding IMF, and the number of zero crossing points and the number of peaks are consistent. Alternatively, the local average m k The difference waveform y at the time when the standard deviation of (t) becomes below the threshold k-1 (t) is set to IMF.

[0041] Based on the measured data x 1 (t) The extracted IMF is called the first IMF C 1 .

[0042] Next, according to the first IMF C 1 and the measured data x 1 (t) Extract the second IMF C 2 Specifically, the measured data x 1 (t) minus the first IMF C 1 The data after x 2 (t) = x 1 (t)-IMF C 1 Set as new measurement data, for this new measurement data x 2 (t) Perform the above-mentioned operation on the measured data x 1 (t) is processed in the same way to extract the second IMF C 2 .

[0043] Repeat this process, and in the nth IMF C n The IMF finding process is terminated at the time when the waveform becomes a waveform with an extremum value of less than 1. The number of IMFs to be extracted varies depending on the basic waveform (measurement data), but generally 10 to 15 IMFs can be extracted.

[0044] In addition, the IMF k It is extracted sequentially starting from the high-frequency components.

[0045] In addition, all IMF C k The sum of the measured data x 1 (t) are equal.

[0046] In order to perform road surface judgment, it is necessary to focus on the high-frequency components of tire vibration, so the first IMF C is used as the IMF for calculating the feature value. 1 、Second IMF C 2 Just wait for the IMF with a lower number.

[0047] In addition, in order to reduce the amount of calculation, only the IMF to be used is extracted and the calculation is stopped at this point. 3 In this case, the fourth IMF C can be omitted. 4 Future IMF calculations.

[0048] Next, we will use the kth IMF C as the IMF to be used. k Set to X k (t).

[0049] The characteristic data calculation unit 14 calculates the IMF X obtained. k (t) Perform Hilbert transform to calculate the instantaneous frequency f at the zero crossing point of the waveform k (t) and instantaneous amplitude a k (t). Instantaneous frequency f k (t) is the phase function θ k The time derivative of (t).

[0050] X can be obtained by the following formula (1): k The Hilbert transform Y of (t) k (t).

[0051] [Number 1]

[0052]

[0053] By using this Hilbert transform, the analytical waveform Z used to calculate the characteristic data is expressed as the following equations (2) to (4): k (t). In addition, the instantaneous frequency f is obtained by equation (5): k (t).

[0054] [Number 2]

[0055]

[0056]

[0057]

[0058]

[0059] like Figure 5 As shown, each IMF X k The waveform of (t) has multiple moments t j The zero crossing point at time t j With time t j+1 There is a maximum instantaneous amplitude between them.

[0060] Therefore, the time t indicated by the bold line in the figure j With time t j+1 The waveform between is considered as frequency f kj is the instantaneous frequency f k (t j ) and amplitude a kj is the instantaneous amplitude a k (t j ') waveform c k,j Part of (λ k,j / 2), the frequency f kj and amplitude akj Set each IMF X k Here, t j '=(t j +t j+1 ) / 2.

[0061] The feature quantity calculation unit 15 calculates the feature quantity based on the IMF X k (t) characteristic data, amplitude a k,j Relative to frequency f k,j The distribution of the statistic is used to calculate the mean μ k , standard deviation σ k and skewness b 1k .

[0062] These statistics are independent of time, so they are used as feature quantities. k Find the characteristic quantity.

[0063] Next, the feature quantity to be used is set to the first IMF C 1 The characteristic quantity of .

[0064] Figure 6 The graph shows the distribution of characteristic quantities calculated from acceleration waveforms when a vehicle travels on dry and wet road surfaces. The ξ axis is the mean μ, the η axis is the standard deviation σ, and the ζ axis is the skewness b. 1 In addition, the light-colored circles represent the data for dry roads, and the dark-colored circles represent the data for wet roads.

[0065] Here, when the feature quantity is set to X = (μ, σ, b 1 ) Figure 6 By distinguishing, within the input space of the feature quantity X, a group of vehicles represented by light-colored circles traveling on a dry road and a group of vehicles represented by dark-colored circles traveling on a wet road, it is possible to determine whether the road surface on which the vehicle is traveling is a dry road or a wet road based on the feature quantity X.

[0066] Similarly, the distribution of characteristic quantities of a snowy road surface and the distribution of characteristic quantities of an icy road surface can be obtained based on the acceleration waveform when the vehicle travels on a snowy road surface and an icy road surface.

[0067] The storage unit 16 stores four pre-calculated road surface models, which are used to separate dry road surfaces from road surfaces other than dry road surfaces, wet road surfaces from road surfaces other than wet road surfaces, snowy road surfaces from road surfaces other than snowy road surfaces, and icy road surfaces from road surfaces other than icy road surfaces by using an identification function f(x) representing a separating hyperplane.

[0068] In finding the characteristic quantity YA =(μ A , σ A , b 1A ) and then Y A As learning data, a road surface model is constructed using a support vector machine (SVM). A =(μ A , σ A , b 1A ) is calculated from the time series waveform of tire vibration obtained by driving a test vehicle equipped with a tire with an acceleration sensor installed on the tire at various speeds on dry, wet, snowy, and icy roads. In addition, the subscript A represents dry, wet, snowy, and icy. In addition, the feature quantity near the recognition boundary selected by SVM is called the road surface feature quantity Y ASV .

[0069] Figure 7 is the dry road feature quantity Y in the input space. DSV and the road surface characteristic quantity Y of the road surface other than the dry road surface nDSV The black circles in the figure are the feature quantities of dry roads, and the white circles are the feature quantities of roads other than dry roads. In addition, the number of feature quantities in the actual input space is three, but this figure is represented by two dimensions (the horizontal axis is p 1 , the vertical axis is p 2 In addition, the storage unit 16 does not need to store all Y D , Y W , Y S , Y I , only store the above Y DSV , Y WSV , Y SSV , Y ISV That's it.

[0070] The identification boundary of the group cannot be separated linearly in general. Therefore, the kernel method is used to convert the road feature vector Y DSV and Y nDSV The nonlinear mapping φ is used to map to a high-dimensional feature space for linear separation, thereby converting the road feature vector Y in the original input space into DSV and Y nDSV Perform non-linear classification.

[0071] Specifically, using the data set X = (x 1 , x 2 , ...x n ) and the category z = {1, -1} to find the best recognition function f(x) = w for identifying data T φ(x)-b.

[0072] Here, the data is the road feature vector Y D , Y nD , the category z=1 is the image in χ 1 The dry road data shown in Figure 2 are shown in Figure 2. z = -1 is the value in χ 2 The data of the road surface other than the dry road surface is shown in FIG. 1 . In addition, w is a weight coefficient, b is a constant, and f(x)=0 is a recognition boundary.

[0073] For example, using the Lagrange undetermined multiplier method, we can identify the function f(x) = w T φ(x)-b is optimized. The optimization problem is replaced by the following equations (6) and (7).

[0074] [Number 3]

[0075]

[0076]

[0077] Here, α and β are indicators of the existence of a plurality of learning data. In addition, λ is a Lagrange multiplier, and λ>0.

[0078] At this time, by using the inner product φ(x α )φ(x β ) is replaced by the kernel function K(x α , x β ), which enables the identification function f(x)=w T φ(x)-b is nonlinear. In addition, φ(x α )φ(x β ) is to convert x α and x β The inner product after mapping to a higher-dimensional space through mapping φ.

[0079] The Lagrange multiplier λ can be obtained for the above formula (2) using an optimization algorithm such as the steepest descent method or SMO (Sequential Minimal Optimization). At this time, since a kernel function is used, it is not necessary to directly obtain a high-dimensional inner product. Therefore, the calculation time can be greatly reduced.

[0080] In this example, as the kernel function K(x α , x β ), the Gaussian kernel (RBF kernel) shown in the following formula was used.

[0081] [Number 4]

[0082]

[0083] In order to distinguish between dry roads and roads other than dry roads, the identification function f(x) as a separating hyperplane is given a margin, so that dry roads and roads other than dry roads can be distinguished with high accuracy. The separating hyperplane is used to convert the dry road feature vector Y D The road surface feature vector Y of the road surface other than the dry road surface nD The margin is the distance from the separating hyperplane to the nearest sample (support vector), and the separating hyperplane that serves as the identification boundary is f(x)=0.

[0084] Moreover, if Figure 7 As shown, the dry road feature vector Y D All roads are in the region of f(x) ≥ +1, except for dry roads. W It is in the region of f(x)≤-1.

[0085] The dry road model for distinguishing dry roads from roads other than dry roads has a support vector Y at a distance of f(x) = +1. DSV and the support vector Y at a distance of f(x) = -1 nDSV The input space of Y DSV and Y nDSV There are usually multiple ones.

[0086] The kernel function calculation unit 17 calculates the feature value X calculated by the feature value calculation unit 15 and the support vectors Y of the dry model, wet model, snow model and ice model recorded in the storage unit 16. DSV , Y WSV , Y SSV and Y ISV , respectively calculate the kernel function K D (X, Y), K W (X, Y), K S (X, Y) and K I (X, Y).

[0087] In the road surface state determination unit 18, the kernel function K is used based on the following equations (9) to (12): D (X, Y), K W (X, Y), K S (X, Y), K I Four identification functions f of (X, Y) D (x), f W (x), f S (x), f I The value of (x) is used to determine the road surface condition.

[0088] [Number 5]

[0089]

[0090]

[0091]

[0092]

[0093] f D is the identification function used to distinguish dry roads from other roads, f W is the identification function used to distinguish wet roads from other roads, f S is the identification function used to distinguish snowy roads from other roads, f I is a recognition function used to distinguish icy roads from other roads.

[0094] In addition, N DSV is the number of support vectors for the dry model, N WSV is the number of support vectors for the wet model, N SSV is the number of support vectors for the snow model, N ISV is the number of support vectors for the icing model.

[0095] Identify the Lagrange multiplier λ of a function D The values ​​of φ and φ are obtained by learning when obtaining a discrimination function for discriminating between a dry road surface and other road surfaces.

[0096] In this example, the identification function f is calculated separately D 、f W 、f S 、f I , according to the calculated identification function f A The road condition is determined by the identification function of the maximum value.

[0097] Next, refer to Figure 8 The flowchart of FIG. 1 illustrates a method for determining the state of the road surface on which the tire 20 is traveling using the road surface state determination device 10 .

[0098] First, the acceleration sensor 11 detects tire vibration generated by input from the road surface R on which the tire 20 travels (step S10 ), and the time series waveform of the tire vibration is extracted from the detected tire vibration signal (step S11 ).

[0099] Then, the EMD algorithm was used to obtain multiple IMF C based on the extracted tire vibration time series waveform data. 1 ~C n (Step S12) Then, the first to third IMFs with lower numbers are extracted from these IMFs. 1 ~C3 , select the IMF C to be used in the judgment of road surface condition k , and set it to X k (t)(Step S13).

[0100] Next, in the case of X k (t) Perform Hilbert transform to calculate the instantaneous frequency f at the zero crossing point as characteristic data k (t) and instantaneous amplitude a k After the maximum value of (t) is reached (step S14), the instantaneous amplitude a k (t) relative to the instantaneous frequency f k The calculated statistic is set as the feature value X k (Step S15). In this example, the statistic is set to the average μ k , standard deviation σ k and skewness b 1k .

[0101] Next, based on the calculated feature quantity X k and the support vector Y of the road surface model recorded in the storage unit 16 A Find the kernel function K A (X, Y) (step S16) Here, the subscript A represents dryness, wetness, snow accumulation, and ice formation.

[0102] Moreover, the kernel function K is used to calculate A Four identification functions f of (X, Y) D (x), f W (x), f S (x), f I (x) (step S17) After that, the calculated identification function f A The values ​​of (x) are compared, and the road surface condition of the identification function showing the largest value is determined to be the road surface condition of the road on which the tire 20 is traveling (step S18).

[0103] Fig. 9 This graph compares the road surface discrimination accuracy of the present method with the discrimination accuracy of a previous method that uses a GA kernel to discriminate the road surface condition using the vibration level of a specific frequency calculated from the time series waveform of tire vibration as a feature quantity. It can be seen that the accuracy of the present method is increased by about 3% to 4% compared with the previous method.

[0104] In addition, if Fig.10As shown in the figure, for the same amount of data (approximately 3300 data), when comparing the time taken to learn the support vector machine between this method and the previous method, a considerable improvement was achieved using the feature-extracted data. Therefore, in this method, it was confirmed that the amount of calculation was significantly reduced compared to the previous method.

[0105] The present invention has been described above using the embodiments, but the technical scope of the present invention is not limited to the scope described in the embodiments. It is also obvious to those skilled in the art that various changes or improvements can be made to the embodiments. It is obvious from the claims that such changes or improvements can also be included in the technical scope of the present invention.

[0106] For example, in the above embodiment, the tire vibration detection unit is set as an acceleration sensor 11, but other vibration detection units such as a pressure sensor may also be used. In addition, regarding the installation position of the acceleration sensor 11, one may be provided at each position separated from the tire width direction by a predetermined distance in the width direction, or may be provided at other positions such as within the module. In addition, the number of acceleration sensors 11 is not limited to one, and may be provided at multiple positions in the tire circumferential direction.

[0107] In the above embodiment, the first IMF C is used as the IMF for calculating the feature value. 1 However, other IMFs may be used. In addition, as described above, in order to perform road surface determination, it is necessary to focus on the high-frequency components of tire vibration, so it is preferable to use a lower-numbered IMF as the IMF for calculating the feature amount.

[0108] In addition, in order to reduce the amount of calculation, only the IMF to be used is extracted and the calculation is stopped at this point. 3 In this case, the fourth IMF C can be omitted. 4 Future IMF calculations.

[0109] In addition, in the above embodiment, only the first IMF C is used as the IMF. 1 However, by using multiple IMFs and performing road surface determination for each IMF, the accuracy of road surface determination can be improved.

[0110] In the above embodiment, the feature quantities are set to mean μ, standard deviation σ, and skewness b 1 , but we can also add the kurtosis b 2 Or, you can also use the mean μ, standard deviation σ, skewness b 1 , Kurtosis b 2Combine multiple statistics in .

[0111] In the above embodiment, the characteristic amount is a statistic obtained by finding the distribution of the instantaneous frequency f(t), but a statistic obtained from the distribution of the instantaneous amplitude a(t) may be used.

[0112] In addition, in the embodiment, it is determined whether the road surface on which the tire 20 is traveling is a dry road surface, a wet road surface, a snowy road surface, or an icy road surface. However, in the case where the road surface is distinguished between two types of road surfaces such as dry / wet, if the two boundary surfaces of the distribution of characteristic quantities of one road surface state (dry boundary surface) and the distribution of characteristic quantities of another road surface state (wet boundary surface) drawn by a support vector machine are replaced by a boundary surface (dry-wet boundary surface) for separating one road surface state from another road surface state to distinguish the road surface state, the road surface state distinction accuracy can be further improved.

[0113] Conventionally, the boundary surface (hyperplane) of the distribution of the support vector (feature quantity) of the dry road drawn by the support vector machine and the identification function for separating the dry road from other roads (wet road, snowy road and icy road) are f D = 0, the boundary surface of the distribution of the support vector (feature quantity) of the wet road surface and the identification function for separating the wet road surface from other road surfaces are f W = 0, so in the dry / wet road discrimination, use Fig.11 The boundary surface f for determining a dry road surface shown by a dashed line in (a) D = 0 and the boundary surface f for judging a wet road surface shown by the dotted line in the figure W =0 are used to identify the road surface.

[0114] Therefore, if Fig.11 As shown in (b), by setting the boundary surface as the boundary surface f between the support vectors for dry road surface and the support vectors for wet road surface DW =0, thereby making the boundary surface one, so the accuracy of distinguishing between the two types of dry / wet road surfaces is further improved.

[0115] Furthermore, it goes without saying that also with respect to other two types of road surface discrimination such as dry / snowy, dry / icy, or wet / snowy, if the boundary surface is made one, the accuracy of the two types of road surface discrimination can be further improved.

[0116] [Example]

[0117] The feature value of each time window calculated based on the time series waveform of tire vibration when driving on dry and wet roads, i.e., road surface data, is used as learning data, and the support vector for dry roads and wet roads is obtained through machine learning (SVM).

[0118] Specifically, as shown in Table 1 below, the road data used is divided into training data and test data. After the support vectors for dry roads and wet roads are obtained, the boundary surface between the support vectors for dry roads and wet roads is obtained. At this time, the hyperparameters C and σ of the support vector machine are set to the values ​​with the highest accuracy under each condition. At this time, the maximum number of support vectors is 415.

[0119] [Table 1]

[0120] data train test dry 2130 1071 Wet 694 346

[0121] Fig.12 This is a graph comparing the dry / wet discrimination accuracy when one boundary surface is used and the dry / wet discrimination accuracy when two boundary surfaces are used in the conventional method.

[0122] As shown in the figure, by using only one boundary surface, the discrimination accuracy is improved by about 3%. Fig.11 By making the boundary surface one as shown in (b), the accuracy of distinguishing between the two types of road surfaces, dry / wet, is improved.

[0123] The present invention is described above using the implementation mode, but the present invention can also be described as follows. That is, the present invention is a road surface state determination method, which determines the state of the road surface contacted by the tire according to the time-varying waveform of the vibration of the running tire detected by the vibration detection unit, and the road surface state determination method is characterized in that it includes the following steps: detecting the time-varying waveform of the vibration of the tire; using the empirical mode decomposition algorithm to obtain multiple natural vibration modes according to the data of the time-varying waveform; selecting and extracting any natural vibration mode from the multiple natural vibration modes; performing Hilbert transform on the extracted natural vibration mode to calculate characteristic data such as instantaneous frequency and instantaneous amplitude; calculating characteristic quantities according to the distribution of the characteristic data; and determining the road surface state according to the calculated characteristic quantities and the characteristic quantities pre-calculated for each road surface state, wherein the characteristic quantities are statistical quantities such as the average, standard deviation, skewness, and kurtosis of the distribution of the characteristic data.

[0124] By setting the feature quantity extracted from the time-varying waveform of tire vibration as a statistic that does not depend on time in this way, the amount of calculation can be greatly reduced, and thus the road surface condition can be determined quickly and accurately.

[0125] The feature amount for each road surface condition described above is obtained by machine learning (support vector machine) using the feature amount for each time window calculated from the time series waveform of tire vibration, which is obtained in advance for each road surface condition, as learning data.

[0126] In addition, in the step of judging the road surface condition, after calculating the Gaussian kernel function based on the calculated feature value and the feature value pre-calculated for each road surface condition, the road surface condition is judged based on the value of the identification function using the calculated Gaussian kernel function, thereby reliably reducing the amount of calculation.

[0127] In addition, in the case where the road surface is distinguished into two types of road surface distinctions, instead of using two boundary surfaces, namely, a boundary surface of the distribution of characteristic quantities of one road surface state and a boundary surface of the distribution of characteristic quantities of another road surface state drawn by a support vector machine, a boundary surface drawn by a support vector machine for separating one road surface state from another road surface state is used to distinguish the road surface state, thereby further improving the road surface state distinction accuracy.

[0128] In addition, the present invention is a road surface state determination device, which detects the vibration of a running tire to determine the state of the road surface contacted by the tire, and the road surface state determination device is characterized in that it comprises: a vibration detection unit, which is installed on the tire and detects the time-varying waveform of the vibration of the running tire; an inherent vibration mode extraction unit, which uses an empirical mode decomposition algorithm to obtain multiple inherent vibration modes based on the time-varying waveform, and extracts any inherent vibration mode from the acquired multiple inherent vibration modes; a feature data calculation unit, which performs Hilbert transform on the extracted inherent vibration mode to calculate feature data; and a feature quantity calculation unit, which A characteristic quantity is calculated based on the distribution of the characteristic data; a storage unit, which stores the characteristic quantity calculated in advance for each road surface condition using the time-varying waveform of the vibration; a kernel function calculation unit, which calculates a Gaussian kernel function based on the calculated characteristic quantity and the characteristic quantity calculated in advance for each road surface condition; and a road surface condition judgment unit, which judges the road surface condition based on the value of an identification function using the calculated Gaussian kernel function, wherein the characteristic quantity is a statistical quantity such as an average, standard deviation, skewness, kurtosis, etc. of the distribution of the characteristic data, and the road surface condition judgment unit compares the values ​​of the identification function calculated for each road surface condition to judge the road surface condition.

[0129] By using the road surface condition determination device having the above-described structure, the amount of calculation can be significantly reduced, and the road surface condition can be determined quickly and accurately.

[0130] In addition, instead of the Gaussian kernel function, an indefinite-valued kernel function such as a polynomial kernel function or a Laplace kernel function may be used.

[0131] Description of Reference Numerals

[0132] 10: road surface condition determination device; 11: acceleration sensor; 12: vibration waveform detection unit; 13: natural vibration mode extraction unit; 14: feature data calculation unit; 15: feature quantity calculation unit; 16: storage unit; 17: kernel function calculation unit; 18: road surface condition determination unit; 20: tire; 21: inner lining layer; 22: tire air chamber; R: road surface.

Claims

1. A road surface condition determination method, which determines the condition of a road surface contacted by a tire based on a time-varying waveform of a vibration of a running tire detected by a vibration detection unit, wherein the road surface condition determination method is characterized in that: The following steps are involved: detecting a time-varying waveform of vibration of the tire; Using an empirical mode decomposition algorithm to obtain a plurality of natural vibration modes based on the data of the time-varying waveform; Selecting and extracting any natural vibration mode from the plurality of natural vibration modes; Performing Hilbert transform on the extracted natural vibration mode to calculate characteristic data; calculating a feature quantity according to the distribution of the feature data; and The road surface condition is determined based on the calculated characteristic amount and the characteristic amount obtained in advance for each road surface condition. The characteristic data is either or both of the instantaneous frequency and the instantaneous amplitude. The feature quantity is a statistic of the distribution of the feature data.

2. A road surface condition determination method, which determines the condition of a road surface contacted by a tire based on a time-varying waveform of a vibration of a running tire detected by a vibration detection unit, wherein the road surface condition determination method is characterized in that: The following steps are involved: detecting a time-varying waveform of vibration of the tire; Using an empirical mode decomposition algorithm to obtain a plurality of natural vibration modes based on the data of the time-varying waveform; Selecting and extracting any natural vibration mode from the plurality of natural vibration modes; Performing Hilbert transform on the extracted natural vibration mode to calculate characteristic data; calculating a feature quantity according to the distribution of the feature data; and The road surface condition is determined based on the calculated characteristic amount and the characteristic amount obtained in advance for each road surface condition. The characteristic quantity is a statistic of the distribution of the characteristic data, In the step of determining the road surface condition, After an indefinite-valued kernel function is calculated based on the calculated feature amount and the feature amount previously determined for each road surface condition, the road surface condition is determined based on the value of a discrimination function using the calculated indefinite-valued kernel function.

3. The road surface condition determination method according to claim 2, It is characterized in that The characteristic data is either or both of the instantaneous frequency and the instantaneous amplitude.

4. The road surface condition determination method according to claim 2 or 3, It is characterized in that In the step of determining the road surface condition, The unvalued kernel function is a polynomial kernel function or a Laplace kernel function.

5. The road surface condition determination method according to claim 2 or 3, It is characterized in that When the road surface is judged as two types of road surface, Instead of two boundary surfaces drawn by a support vector machine, namely a boundary surface of distribution of characteristic quantities of one road surface state and a boundary surface of distribution of characteristic quantities of another road surface state, a boundary surface drawn by a support vector machine for separating one road surface state from another road surface state is used to distinguish the road surface state.

6. A road surface condition determination device, which detects the vibration of a tire during driving to determine the state of a road surface contacted by the tire, wherein the road surface condition determination device is characterized by comprising: a vibration detection unit mounted on the tire to detect a time-varying waveform of vibration of the tire during travel; an intrinsic vibration mode extraction unit, which uses an empirical mode decomposition algorithm to obtain a plurality of intrinsic vibration modes according to the time-varying waveform, and extracts any intrinsic vibration mode from the plurality of intrinsic vibration modes obtained; A characteristic data calculation unit, which performs Hilbert transform on the extracted natural vibration mode to calculate characteristic data; a feature quantity calculation unit, which calculates a feature quantity according to the distribution of the feature data; a storage unit that stores a feature quantity calculated using a time-varying waveform of vibration, which is obtained in advance for each road surface state; an indeterminate-value kernel function calculation unit, which calculates an indeterminate-value kernel function based on the calculated feature quantity and the feature quantity calculated in advance for each road surface state; as well as a road surface condition determination unit that determines a road surface condition based on a value of a discrimination function using the calculated indefinite-valued kernel function, The characteristic quantity is a statistic of the distribution of the characteristic data. The road surface condition determination unit determines the road surface condition by comparing the values ​​of the identification function obtained for each road surface condition.

Images