JP7030531B2 - Road surface condition determination method and road surface condition determination device - Google Patents

Description

The present invention relates to a method for determining a state of a road surface on which a vehicle travels and a device thereof.

Conventionally, as a method of discriminating the road surface condition using only the data of the time-series waveform of the tire vibration during running, each time window calculated from the time-series waveform extracted by applying the window function to the time-series waveform of the tire vibration. Using a function such as a GA kernel calculated from the vibration level of a specific frequency band, which is the feature vector of, and the road surface feature amount for each time window calculated from the time-series waveform of tire vibration obtained in advance for each road surface condition. Therefore, a method for determining whether the road surface condition is DRY / WET / SNOW / ICE has been proposed (see, for example, Patent Document 1).

Japanese Unexamined Patent Publication No. 2014-35279

By the way, in the conventional road surface condition discrimination method, "one-to-other" discrimination such as discrimination between a DRY road surface and a road surface other than the DRY road surface is performed for discrimination of the road surface, so that other types of road surfaces are discriminated at once. If you try, the number of operations will simply increase by the number of road surfaces. This becomes an obstacle when trying to reduce the size of the calculation.
On the other hand, for other types of road surface discrimination, there is a "one-to-one" discrimination method such as discrimination between a DRY road surface and a WET road surface. This "one-to-one" discrimination method has higher discrimination accuracy than the above-mentioned "one-to-other" discrimination method, but the amount of calculation is larger than that of the "one-to-other" discrimination method. There was a problem.

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 an apparatus thereof capable of accurately and surely performing road surface discrimination with a small amount of calculation.

As a result of diligent studies, the present inventors compared the DRY road surface and the WET road surface, and found that the front wheel tires and the water film collide with each other on the WET road surface, but the rear wheel tires Since the water film is separated by the tires of the front wheels, the collision with the water film is reduced. Since the water film on the surface is scraped off and the snow is compacted by the tires of the front wheels on the SNOW road surface, it is preferable to distinguish the DRY road surface from the ICE road surface or the SNOW road surface by the rear wheels. It has arrived.
That is, the present invention is a method for discriminating the state of the road surface in contact with the tire by the road surface condition discriminating device from the time change waveform of the vibration of the running tire detected by the vibration detecting means mounted on the tire. It is a front wheel acceleration waveform which is a time change waveform of the vibration of the front wheel and a time change waveform of the vibration of the rear wheel by the vibration detecting means mounted on the front wheel of the tire and the vibration detecting means mounted on the rear wheel of the tire. A step of detecting the rear wheel acceleration waveform, a step of calculating the front wheel feature amount and the rear wheel feature amount from the front wheel acceleration waveform and the rear wheel acceleration waveform, respectively, and the calculated front wheel feature. The feature amount is provided with a step of discriminating the state of the road surface in which the front wheel and the rear wheel are in contact from the amount, the feature amount of the rear wheel, and the road surface feature amount obtained in advance for each road surface condition . It is a statistic of the distribution of one or both of the instantaneous frequency and the instantaneous amplitude, which is extracted by performing the Hilbert conversion to the natural vibration mode acquired by using the empirical mode decomposition algorithm, and the state of the road surface is described. In the discriminating step, after the kernel function including the feature amount of the front wheel is calculated using the feature amount of the front wheel and the road surface feature amount, the identification using the kernel function including the calculated feature amount of the front wheel is used. The DRY road surface and the WET road surface are discriminated from the value of the function, a kernel function including the rear wheel feature amount is calculated using the rear wheel feature amount and the road surface feature amount, and then the calculated above is performed . Discrimination between DRY road surface and ICE road surface and DRY road surface and SNOW road surface, or DRY road surface and ICE / SNOW road surface is performed from the value of the discrimination function using the kernel function including the features of the rear wheels. It is characterized by.
The ICE / SNOW road surface refers to any of the ICE road surface, the SNOW road surface, and the ice-snow road surface.
As a result, even if a "one-to-one" discrimination method having a higher discrimination accuracy than the "one-to-other" discrimination method is used, the road surface discrimination can be performed accurately and reliably with a small amount of calculation.
In addition, the feature amount extracted from the time-varying waveform of tire vibration is a statistic that does not depend on time, so that the amount of calculation can be significantly reduced, so the road surface condition can be quickly and accurately adjusted. It can be determined.
The above-mentioned feature amount for each road surface condition is machine learning (support vector machine) using the feature amount for each time window calculated from the time-series waveform of tire vibration obtained in advance for each road surface condition as training data. Demanded by.

Further, the present invention is a road surface condition discriminating device that detects the vibration of a running tire and discriminates the state of the road surface in contact with the tire, and is mounted on the front wheel of the tire to change the vibration of the front wheel over time. A front wheel vibration detecting means for detecting a front wheel acceleration waveform which is a waveform, a rear wheel vibration detecting means for detecting a rear wheel acceleration waveform which is a time-varying waveform of the vibration of the rear wheel mounted on the rear wheel of the tire, and the above. From the front wheel acceleration waveform and the rear wheel acceleration waveform, the feature amount calculation means for calculating the feature amount of the front wheel and the feature amount of the rear wheel, respectively, and the time change waveform of the tire vibration obtained in advance for each road surface condition are obtained. A storage means for storing the road surface feature amount calculated using the method, and a road surface condition determining means for determining the state of the road surface in which the front wheel and the rear wheel are in contact from the calculated feature amount and the road surface feature amount. The feature quantity is a statistic of the distribution of one or both of the instantaneous frequency and the instantaneous amplitude extracted by performing the Hilbert conversion to the natural vibration mode acquired by using the empirical mode decomposition algorithm. The road surface condition determining means calculates a kernel function including the front wheel feature amount using the front wheel feature amount and the road surface feature amount, and then calculates the kernel function including the calculated front wheel feature amount. After discriminating between the DRY road surface and the WET road surface from the value of the discrimination function using the above, and calculating the kernel function including the feature amount of the rear wheel using the feature amount of the rear wheel and the feature amount of the road surface , the above Discrimination between DRY road surface and ICE road surface and discrimination between DRY road surface and SNOW road surface, or DRY road surface and ICE / SNOW road surface from the value of the discrimination function using the calculated kernel function including the feature amount of the rear wheel. It is characterized by making a discrimination.
By using the road surface condition discriminating device having the above configuration, it is possible to improve the discriminating accuracy when the vibration detecting means is attached to the front wheels and the rear wheels to discriminate the road surface.

It should be noted that the outline of the present invention does not list all the necessary features of the present invention, and a subcombination of these feature groups can also be an invention.

It is a functional block diagram which shows the structure of the road surface condition discriminating apparatus which concerns on this embodiment. It is a figure which shows an example of the mounting position of an accelerometer. It is a figure which shows an example of the time-series waveform of a tire vibration. It is a figure which shows the acquisition method of a natural vibration mode. It is a figure which shows the acquisition method of a feature data. It is a schematic diagram which shows the distribution state of a feature amount. It is a schematic diagram which shows the separation hyperplane in an input space and a feature space. It is a flowchart which shows the discriminating method of the road surface condition which concerns on this embodiment. It is a figure which compared the acceleration waveform of the DRY road surface and the acceleration waveform of the WET road surface in the front wheel and the rear wheel.

FIG. 1 is a functional block diagram showing the configuration of the road surface condition determination device 10.
The road surface condition determination device 10 includes acceleration sensors 11 and 21 as tire vibration detection means, front wheel acceleration waveform extraction means 12, rear wheel acceleration waveform extraction means 22, front wheel feature amount calculation means 13, and rear wheel feature amount calculation. The means 23, the front wheel identification function calculation means 14, the rear wheel identification function calculation means 24, the storage means 15, and the road surface state determination means 16 are provided.
The acceleration sensor 11 is an acceleration sensor mounted on the front wheels (hereinafter referred to as a front wheel side acceleration sensor), and the acceleration sensor 21 is an acceleration sensor mounted on the rear wheels (hereinafter referred to as a rear wheel side acceleration sensor). Each means of the waveform extraction means 12 to the road surface state determination means 16, the rear wheel acceleration waveform extraction means 22, and the rear wheel identification function calculation means 24 is composed of, for example, computer software and a memory such as a RAM.
As shown in FIG. 2, the front wheel side acceleration sensor 11 is integrally arranged at substantially the center of the inner liner portion 31 of the front wheel tire (hereinafter referred to as the front wheel 30F) on the tire air chamber 32 side, and is input from the road surface. The vibration of the front wheel 30F is detected.
On the other hand, the rear wheel side acceleration sensor 21 is integrally arranged at substantially the center of the inner liner portion 31 of the rear wheel tire (hereinafter referred to as the rear wheel 30R) on the tire air chamber 32 side, and is based on input from the road surface R. The vibration of the rear wheel 30R is detected.
The signal of the tire vibration of the front wheel 30F, which is the output of the front wheel side acceleration sensor 11, and the signal of the tire vibration of the rear wheel 30R, which is the output of the rear wheel side acceleration sensor 21, are amplified by an amplifier (not shown) and then amplified. It is converted into a digital signal and sent to the front wheel acceleration waveform extracting means 12 and the rear wheel acceleration waveform extracting means 22.
When the front wheel 30F and the rear wheel 30R are not distinguished, it is simply referred to as a tire 30.

The front wheel acceleration waveform extracting means 12 extracts an acceleration waveform (hereinafter referred to as a front wheel acceleration waveform) which is a time-series waveform of the tire vibration for each rotation of the tire 30 from the tire vibration signal detected by the front wheel side acceleration sensor 11. Then, the rear wheel acceleration waveform extracting means 22 extracts an acceleration waveform (hereinafter referred to as a rear wheel acceleration waveform) which is a time-series waveform of the tire vibration from the signal of the tire vibration detected by the rear wheel side acceleration sensor 21.
FIG. 3 is a diagram showing an example of the time-series waveform of the tire vibration. The time-series waveform of the tire vibration has large peaks near the stepping position and the kicking position, and the land portion of the tire touches the ground. In both the front pre-stepping region R _f and the post-kicking region R _k after the land portion of the tire is separated from the road surface, different vibrations appear depending on the road surface condition. On the other hand, since the region before the pre-stepping region R _f and the region after the kicking region R _k (hereinafter referred to as the out-of-road region) are hardly affected by the road surface, the vibration level is small and the information on the road surface is small. Does not include.
As the definition of the off-road surface region, for example, a background level may be set for the acceleration waveform, and a region having a vibration level smaller than this background level may be set as the out-of-road surface region.
In this example, of the acceleration waveforms. The acceleration waveform in the road surface region, which is the region from the pre-stepping region R _f to the post-kicking region R _k , including the road surface information, is defined as measurement data x ₁ (t), and this measurement data x ₁ (t) is empirical. After decomposing into multiple natural mode functions (IMF) using the algorithm of Mode Decomposition (EMD), Hilbert transform is performed for each IMF to calculate the features.

The front wheel feature amount calculation means 13 includes a natural vibration mode extraction unit 13a, a feature data calculation unit 13b, and a feature amount calculation unit 13c, and obtains a front wheel feature amount from the front wheel acceleration waveform extracted by the front wheel acceleration waveform extraction means 12. calculate.
The natural vibration mode extraction unit 13a acquires a plurality of IMFs (C ₁ , C ₂ , ..., C _n ) from the measurement data x ₁ (t) using an EMD algorithm, and also acquires a plurality of acquired IMFs. Extract any IMFC _k from.
Here, how to obtain the IMF will be described.
First, as shown in FIG. 4, all the maximum points and the minimum points of the measurement data x ₁ (t) are extracted, and the upper envelope line e _max (t) connecting the maximum points and the lower side connecting the minimum points. After finding the envelope e _min (t), the local _average of the upper envelope e _max (t) and the lower envelope e _min (t) = (e _max (t) + e min (e max (t) + e _min ( t))) / 2 is calculated.
Next, the difference waveform y ₁ (t) = x ₁ (t) −m ₁ (t) between the measurement data x ₁ (t) and the local average m ₁ (t) is obtained. The difference waveform y ₁ (t) has poor symmetry and cannot be said to be an IMF. Therefore, the same processing as that performed on the measurement data x ₁ (t) is performed on the difference waveform y ₁ (t) to obtain the difference waveform y ₂ (t). Further, this process is repeated to obtain the difference waveforms y ₃ (t), y ₄ (t), ..., Y _m (t). The difference waveform y _k (t) has higher symmetry as k becomes larger, and becomes closer to the IMF.
The conditions for the difference waveform to be the IMF are that the number of zero cross points and the number of peaks of y _k (t) do not change 4 to 8 times in a row in the process of obtaining the IMF, and the number of zero cross points and the number of peaks. If they match, the conditions are proposed. The difference waveform y _k-1 (t) at the time when the standard deviation of the local average m _k (t) becomes equal to or less than the threshold value may be used as the IMF.
The IMF extracted from this measurement data x ₁ (t) is called the first IMFC ₁ .

Next, the second IMFC ₂ is extracted from the first IMFC ₁ and the measurement data x ₁ (t). Specifically, the data x ₂ (t) = x ₁ (t) -IMFC ₁ obtained by subtracting the first IMF ₁ from the measurement data x ₁ (t) is used as new measurement data, and this new measurement data x ₂ The second IMFC ₂ is extracted by performing the same processing on (t) as the above-mentioned processing on the measurement data x ₁ (t).
This process is repeated, and when the nth IMFC _n becomes a waveform having one extremum, the process for obtaining the IMF ends. The number of IMFs to be extracted varies depending on the original waveform (measurement data), but usually 10 to 15 IMFs are extracted.
IMFC _k is extracted in order from the high frequency component.
Also, the sum of all IMFC _ks is equal to the measurement data x ₁ (t).
By the way, in order to discriminate the road surface, it is necessary to pay attention to the high frequency component of the tire vibration, so as the IMF for calculating the feature amount, a low number IMF such as the first IMFC ₁ or the second IMFC ₂ is used. You can use it.
In order to reduce the amount of calculation, only the IMF to be used should be extracted and the calculation should be stopped there. For example, when only the third IMFC ₃ is used, the calculation for extracting the fourth IMFC ₄ and later may be omitted.
Hereinafter, the kth IMFC _k , which is the IMF to be used, is referred to as X _k (t).

このヒルベルト変換により、特徴データを算出するための解析波形Ｚ_k(t)は、以下の式（２）～（４）のように表せる。
［数２］

図５に示すように、各IMFＸ_k(t)の波形は、複数の時刻t_jにおいてのゼロクロス点を有し、時刻t_jと時刻t
_j+1との間に、瞬時振幅の極大値を有する。
そこで、同図の太線で示す、時刻t_jと時刻t _j+1との間の波形を、周波数ｆ_kjが瞬時周波数ｆ_k(t_j)で、振幅ａ_kjが瞬時振幅ａ_k(t_j ^’)の波形ｃ_k,jの一部（λ_k,j/２）であるとみなし、この周波数ｆ_kjと振幅ａ_kjとを各IMFＸ_k(t)の特徴データとする。ここで、t_j ^’＝（t_j＋t_j+1）/２である。
特徴量算出部１３ｃは、IMFＸ_k(t)の特徴データである、周波数ｆ_k,jに対する振幅ａ_k,jの分布から、統計量である、平均μ_k、標準偏差σ_k、及び、歪度ｂ_{1 k}を算出する。
これらの統計量は、時間に依存しない統計量であるので、これらの統計量を特徴量として採用する。なお、特徴量はＣ_k毎に求まる。 The feature data calculation unit 13b performs a Hilbert transform on the obtained IFX _k (t), and calculates the instantaneous frequency f _k (t) at the zero cross point of the waveform and the maximum value of the instantaneous amplitude a _k (t). The instantaneous frequency f _k (t) is the time derivative of the phase function θ _k (t).
The Hilbert transform Y _k (t) of X _k (t) is obtained by the following equation (1).
[Number 1]

By this Hilbert transform, the analysis waveform Z _k (t) for calculating the feature data can be expressed by the following equations (2) to (4).
[Number 2]

As shown in FIG. 5, the waveform of each IFX _k (t) has zero crossing points at a plurality of time t _j , and the time t _j and the time t j.
It has a maximum value of instantaneous amplitude between _{j + 1} and j + 1.
Therefore, the waveform between time t _j and time t _{j + 1} , shown by the thick line in the figure, has a frequency f _k j as an instantaneous frequency f _k (t _j ) and an amplitude a _k j as an instantaneous amplitude a _k (t _j ). It is regarded as a part (λ _{k, j} / 2) of the waveform c _{k, j} of ^' ), and the frequency f k _j and the amplitude a k j are used as the feature data of each _{IFX k} (t). Here, t _j ^' = (t _j + t _{j + 1} ) / 2.
The feature amount calculation unit 13c is based on the distribution of the amplitudes a _{k and} j with respect to the frequencies f _{k and j} , which are the feature data of the IFX _k (t), and are statistics such as the mean μ _k , the standard deviation σ _k , and the skewness. Calculate the degree b _{1 k} .
Since these statistics are time-independent statistics, these statistics are adopted as features. The feature amount is obtained for each C _k .

FIG. 6 is a schematic diagram showing the input space of the feature amount when the feature amount is X = (μ, σ, b ₁ ). The a1 axis is the _average μ, and the _a2 axis is the standard deviation σ, _a3 . The axis has a skewness b ₁ .
In the figure, if the set of the features X _i when the group C is traveling on the DRY road surface and the set of the features _X'i when the group C'is traveling on the WET road surface, the group C and If it can be distinguished from Group C', it is possible to determine whether the road surface on which the tire is traveling is a DRY road surface or a WET road surface.
Similarly, the distribution of the feature amount on the SNOW road surface and the distribution of the feature amount on the ICE road surface can be obtained from the acceleration waveform when traveling on the SNOW road surface or the ICE road surface.
Although omitted in FIG. 1, the configuration and operation of the rear wheel feature amount calculation means 23 are the same as those of the front wheel feature amount calculation means 13, and the IMFC _k is obtained from the rear wheel acceleration waveform extracted by the rear wheel acceleration waveform extraction means 22. After extracting, the average μ _k , standard deviation σ _k , and skewness b _{1 k} , which are the features of the rear wheels, are obtained for each C _k .
Hereinafter, the feature amount to be used will be referred to as the feature amount of the first IMFC ₁ .

The storage means 15 stores four discriminative models, a D / W discriminative model, a D / S discriminative model, a D / I discriminative model, and an S / I discriminative model, which have been obtained in advance.
The D / W discrimination model is a reference feature quantity Y _DSV (y _jk ) and Y _WSV (y _jk ), which are features for separating a DRY road surface and a WET road surface by a discrimination function f _DW (x) representing a separation hyperplane. ) And the Lagrange multipliers λ _D and λ _W that weight the reference features Y _DSV (y _jk ) and Y _WSV (y _jk ), respectively.
The D / S discrimination model stores the reference feature quantities Y _DSV (y _jk ), Y _SSV (y _jk ), and the Lagrange multipliers λ _D , λ _S , and the D / I discrimination model stores the reference feature quantities Y _DSV (y jk). Stores y _jk ), Y _ISV (y _jk ), and Lagrange multipliers λ _D and λ _I.
Further, the S / I discrimination model stores the reference feature quantities Y _SSV (y _jk ), Y _ISV (y _jk ), and the Lagrange multipliers λ _S and λ _I.
Each identification model is calculated from the time-series waveform of tire vibration obtained by running a test vehicle equipped with a tire with an acceleration sensor attached to the tire at various speeds on each road surface of DRY, WET, SNOW, and ICE. After obtaining the obtained feature quantity Y _A = (μ _A , σ _A , b ₁ A), it is constructed by a support vector machine (SVM) using Y _A as training data. Here, the subscript A indicates DRY, WET, SNOW, and ICE. Further, the feature amount in the vicinity of the identification boundary selected by the SVM is referred to as a road surface feature amount Y _ASV .
The D / W discrimination model is constructed using the front wheel acceleration waveform that is the output of the front wheel side acceleration sensor 11, and the D / S discrimination model, the D / I discrimination model, and the S / I discrimination model are the rear wheels. It is constructed using the rear wheel acceleration waveform which is the output of the side acceleration sensor 21.

FIG. 7 is a conceptual diagram showing a DRY road surface feature amount Y _D and a WET road surface feature amount Y _W on the input space. In the figure, black circles are DRY road surfaces and white circles are WET road surfaces.
As described above, both the DRY road surface features Y _D and the WET road surface features Y _W are matrices, but in order to explain how to obtain the identification boundary of the group, in FIG. 7, the DRY road surface features Y _D and WET The road surface features Y _W are shown as two-dimensional vectors.
Group identification boundaries are generally not linearly separable.
Therefore, by using the kernel method to map the road surface features Y _D and Y _W to a high-dimensional feature space by the nonlinear mapping φ and performing linear separation, the road surface features Y _D and Y _W can be obtained in the original input space. On the other hand, non-linear classification is performed.
When distinguishing between the DRY road surface and the WET road surface, a margin should be provided for the discrimination function f _DW (x) which is a separation hyperplane that separates the DRY road surface feature amount Y _D and the WET road surface feature amount Y _W. Therefore, the DRY road surface and the WET road surface can be accurately distinguished. The margin means the distance from the separation hyperplane to the nearest sample (support vector), and the separation hyperplane which is the discrimination boundary is f (x) = 0.
Then, as shown in FIG. 7, the DRY road surface features Y _D are all in the region of f _DW (x) ≧ + 1, and the WET road surface features Y _W are in the region of f _DW (x) ≦ -1.
The D / W discriminative model that distinguishes between the DRY road surface and the WET road surface is a support vector Y _DSV at a distance of f _DW (x) = + 1 and a support vector Y _WSV at a distance of f _DW (x) = -1. It is an input space equipped with. Generally, there are a plurality of Y _DSVs and Y _WSVs .
The same applies to the D / S discriminative model, the D / I discriminative model, and the S / I discriminative model.

ここで、α，βは複数ある学習データの指標である。また、λはラグランジュ乗数で、λ＞０である。なお、λ＝０である路面特徴量Ｙ_Ajは、識別関数ｆ（ｘ）に寄与しない（サポートベクトルではない）ベクトルデータである。
ラグランジュ乗数は、φ（ｘ_α）φ（ｘ_β）は、ｘ_αとｘ_βを写像φで高次元空間へ写像した後の内積である。
また、φ（ｘ_α）φ（ｘ_β）は、ｘ_αとｘ_βを写像φで高次元空間へ写像した後の内積で、内積φ^T（ｘ_α）φ（ｘ_β）を直接求めずに、カーネル関数Ｋ（ｘ_α，ｘ_β）に置き換えることで、識別関数ｆ（ｘ）＝ｗ^Tφ（ｘ）－ｂを非線形できる。
ラグランジュ乗数λは、前記の式（７）について、最急下降法やＳＭＯ（Sequential Minimal Optimization）などの最適化アルゴリズムを用いて求めることができる。
Ｄ/Ｓ識別モデル、Ｄ/Ｉ識別モデル、及び、Ｓ/Ｉ識別モデルについても、同様に、基準特徴量Ｙ_ASV（ｙ_jk）、及び、ラグランジュ乗数λ_Aを求めることができる。
本例では、カーネル関数Ｋ（ｘ_α，ｘ_β）として、以下の式に示す、ガウシアンカーネル（ＲＢＦカーネル）を用いた。
［数４］

Next, the optimal discriminant function f _DW (x) = that discriminates the data using the set of data X = (x ₁ , x ₂ , ... x _n ) and the belonging class z = {1, -1}. w ^T φ (x) −b is obtained. Here, w is a vector representing the weighting coefficient, and b is a constant.
The data are DRY road surface features Y _Dj and WER road surface features Y _Wj , and the belonging class is DRY road surface data in which z = 1 is indicated by χ ₁ in the figure, and z = -1 is WET indicated by χ ₂ . It is the data of the road surface. f (x) = 0 is the discrimination boundary, and 1 / || w || is the distance between the road surface feature amount Y _Aj (A = D, W) and f (x) = 0.
The discriminant function f _DW (x) = w ^T φ (x) −b is optimized using, for example, the Lagrange undetermined multiplier method. The optimization problem is replaced by the following equations (6) and (7).
[Number 3]

Here, α and β are indices of multiple learning data. Further, λ is a Lagrange multiplier, and λ> 0. The road surface feature amount Y _Aj in which λ = 0 is vector data that does not contribute to the discrimination function f (x) (not a support vector).
The Lagrange multiplier is φ (x _α ) φ (x _β ), which is the inner product after mapping x _α and x _β to a higher dimensional space with the mapping φ.
Further, φ (x _α ) φ (x _β ) is the inner product after mapping x _α and x _β to a high-dimensional space with the mapping φ, and the inner product φ ^T (x _α ) φ (x _β ) is not directly obtained. By substituting the kernel function K (x _α , x _β ), the discriminant function f (x) = w ^T φ (x) −b can be made non-linear.
The Lagrange multiplier λ can be obtained for the above equation (7) by using an optimization algorithm such as a steepest descent method or SMO (Sequential Minimal Optimization).
Similarly, for the D / S discriminative model, the D / I discriminative model, and the S / I discriminative model, the reference feature quantity Y _ASV (y _jk ) and the Lagrange multiplier λ _A can be obtained.
In this example, the Gaussian kernel (RBF kernel) shown in the following equation was used as the kernel function K (x _α , x _β ).
[Number 4]

なお、Ｎ_DSVはＤＲＹモデルのサポートベクトルの数、Ｎ_WSVはＷＥＴモデルのサポートベクトルの数である。また、識別関数のラグランジュ乗数λ_D，λ_Wなどの値は、ＤＲＹ路面とＷＥＴ路面とを識別する識別関数を求める際の学習により求められる。
後輪識別関数演算手段２４は、カーネル関数算出部２４ａと識別関数演算部２４ｂとを備え、ＤＲＹ路面特徴量Ｙ_DとＳＮＯＷ路面特徴量Ｙ_Sとを分離するる識別関数ｆ_DS（ｘ）の値、ＤＲＹ路面特徴量Ｙ_DとＩＣＥ路面特徴量Ｙ_Iとを分離する識別関数ｆ_DI（ｘ）の値、及び、ＳＮＯＷ路面特徴量Ｙ_SとＩＣＥ路面特徴量Ｙ_Iとを分離する識別関数ｆ_SI（ｘ）の値を計算する。
カーネル関数算出部２４ａは、後輪特徴量算出手段２３にて算出された特徴量Ｘと記憶手段１５に記録されているＤ/Ｓモデル、Ｄ/Ｉモデル、Ｓ/Ｉモデルの各サポートベクトルＹ_DSV，Ｙ_SSV，Ｙ_ISVから、上記式（８）を用いて、ガウシアンカーネルＫ_D（Ｘ，Ｙ_DSV）とＫ_S（Ｘ，Ｙ_SSV）とＫ_I（Ｘ，Ｙ_ISV）とを算出する。
識別関数演算部２４ｂでは、カーネル関数Ｋ_D（Ｘ，Ｙ_DSV），Ｋ_S（Ｘ，Ｙ_SSV），Ｋ_I（Ｘ，Ｙ_WSV）を用いて、識別関数ｆ_DS（ｘ），ｆ_DI（ｘ），ｆ_SI（ｘ）の値を求める。
識別関数ｆ_DS（ｘ），ｆ_DI（ｘ），ｆ_SI（ｘ）の値は、下記の式（１０）～（１２）を用いて計算する。
［数６］

路面状態判別手段１６は、前輪識別関数演算手段１４で計算された識別関数ｆ_DW（ｘ）の値と、後輪識別関数演算手段２４で計算された識別関数ｆ_DS（ｘ），ｆ_DI（ｘ），ｆ_SI（ｘ）の値とから、路面状態がＤＲＹ/ＷＥＴ/ＳＮＯＷ/ＩＣＥのいずれかであるかを判別する。 The front wheel discrimination function calculation means 14 includes a kernel function calculation unit 14a and a discrimination function calculation unit 14b, and is a separation superplane that separates the DRY road surface feature amount Y _D and the WET road surface feature amount Y _W (discrimination function f _DW ( Calculate the value of x).
The kernel function calculation unit 14a uses the above equation (8) from the feature amount X calculated by the front wheel feature amount calculation means 13 and the support vectors Y _DSV and Y _WSV of the D / W model recorded in the storage means 15. ) Is used to calculate the Gaussian kernels _{KD (X, Y DSV )} and K _W (X, Y _WSV ).
In the discriminant function calculation unit 14b, the value of the discriminant function f _DW (x) for discriminating between the DRY road surface and the WET road surface by using the kernel functions _KD (X, Y _DSV ) and K _W (X, Y _WSV ). Ask for.
The value of the discriminant function f _DW (x) is calculated using the following equation (9).
[Number 5]

N _DSV is the number of support vectors of the DRY model, and N _WSV is the number of support vectors of the WET model. Further, the values of the Lagrange multipliers λ _D , λ _W , etc. of the discriminant function are obtained by learning when finding the discriminant function for discriminating between the DRY road surface and the WET road surface.
The rear wheel identification function calculation means 24 includes a kernel function calculation unit 24a and an identification function calculation unit 24b, and has an identification function f _DS (x) that separates the DRY road surface feature amount Y _D and the SNOW road surface feature amount Y _S. The value, the value of the discrimination function f _DI (x) that separates the DRY road surface feature amount Y _D and the ICE road surface feature amount Y _I , and the discrimination function that separates the SNOW road surface feature amount Y _S and the ICE road surface feature amount Y _I. f Calculate the value of _SI (x).
The kernel function calculation unit 24a has a feature amount X calculated by the rear wheel feature amount calculation means 23 and each support vector Y of the D / S model, the D / I model, and the S / I model recorded in the storage means 15. From _DSV , Y _SSV , and Y _ISV , the Gaussian kernels KD (X, Y _DSV ), _KS ( _X , Y _SSV ), and KI ( _X , Y _ISV ) are calculated using the above equation (8). ..
The discriminant function calculation unit 24b uses the kernel functions _{KD (X, Y DSV )} , KS ( _X , Y _SSV ), and KI ( _X , Y _WSV ) to discriminate functions f _DS (x), f _DI (. Find the values of x) and f _SI (x).
The values of the discriminant functions f _DS (x), f _DI (x), and f _SI (x) are calculated using the following equations (10) to (12).
[Number 6]

The road surface condition discriminating means 16 includes the value of the discriminating function f _DW (x) calculated by the front wheel discriminating function calculating means 14 and the discriminating functions f _DS (x) and f _DI calculated by the rear wheel discriminating function calculating means 24. From the values of x) and f _SI (x), it is determined whether the road surface condition is DRY / WET / SNOW / ICE.

Next, a method of discriminating the state of the road surface by using the road surface condition discriminating device 10 will be described with reference to the flowchart of FIG.
First, the acceleration sensors 11 and 21 detect the tire vibration generated by the input from the road surface R (step S10), and from the detected tire vibration signal, the front wheel acceleration waveform which is the time-series waveform of the front wheel 30F tire vibration is obtained. The rear wheel acceleration waveform, which is a time-series waveform of the rear wheel 30R tire vibration, is extracted (step S11).
Then, after acquiring a plurality of _{IMFs 1} to Cn from the extracted tire vibration time-series waveform data using the EMD algorithm (step S12), the first of these IMFs has a lower number. -Third IMFCs ₁ to C ₃ are extracted, IMFC _k to be used for determining the road surface condition is selected, and this is set as X _k (t) (step S13).
Next, the Hilbert transform is performed on X _k (t) to calculate the instantaneous frequency f _k (t) at the zero cross point, which is the characteristic data, and the maximum value of the instantaneous amplitude a _k (t) (step S14). Then, a statistic is calculated from the distribution of the instantaneous amplitude a _k (t) with respect to the instantaneous frequency f _k (t), and the calculated statistic is defined as the feature quantity X _k (step S15). In this example, the statistics are mean μ _k , standard deviation σ _k , and skewness b _{1 k} .
Each step from step S12 to step S15 is performed for each of the front wheel acceleration waveform and the rear wheel acceleration waveform, the front wheel feature amount is calculated from the front wheel acceleration waveform, and the rear wheel feature amount is calculated from the rear wheel acceleration waveform. do. In this example, the feature amount X _k is described without distinguishing between the front wheel feature amount and the rear wheel feature amount.
Next, from the calculated feature quantity X _k and the support vectors Y _DSV , Y _WSV , Y _SSV , Y _ISV of the identification model recorded in the storage means 15, the Gaussian kernel KD (X, Y _DSV ) _,. After finding K _W (X, Y _WSV ), K _S ( _X , Y _SSV ), KI (X, Y _ISV ) (step S16), four identifications using the kernel function _KA (X, Y) The functions f _DW (x), f _DS (x), f _DI (x), and f _SI (x) are calculated respectively (step S17).
Finally, the road surface condition is discriminated using the calculated values of the four discriminant functions f _DW (x), f _DS (x), f _DI (x), and f _SI (x) (step S18).
To determine the road surface condition, first, it is determined whether the road surface is a DRY road surface or a WET road surface from the value of the discrimination function f _DW (x) calculated using the front wheel feature amount and the D / W discrimination model.
Specifically, if f _DW (x) <0, it is determined that the road surface is a WET road surface without discriminating between the WET road surface and the SNOW road surface and the WET road surface and the ICE road surface.
On the other hand, when f _DW (x)> 0, the discriminative function f _DS (x) calculated using the rear wheel features, the D / S discriminative model, the D / I discriminative model, and the S / I discriminative model. ), F _DI (x), f _SI (x), it is determined whether the road surface is a DRY road surface, a SNOW road surface, or an ICE road surface.
Specifically, if f _DS (x)> 0 and f _DI (x)> 0, it is determined that the road surface is DRY, and f _DS (x) <0 and f _SI (x)> 0. If so, it is determined to be a SNOW road surface.
Further, if f _DS (x) <0 and f _SI (x) <0, it is determined to be an ICE road surface.

Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the embodiments. It will be apparent to those skilled in the art that various changes or improvements can be made to the above embodiments. It is clear from the claims that such modified or modified forms may also be included in the technical scope of the invention.

For example, in the above embodiment, the tire vibration detecting means is the acceleration sensors 11 and 21, but other vibration detecting means such as a pressure sensor may be used. Also, regarding the installation locations of the acceleration sensors 11 and 21, even if they are installed one by one at a position separated by a predetermined distance in the width direction from the center in the tire width direction, or installed in another location such as in a block. good. Further, the number of acceleration sensors 11 and 21 is not limited to one, and may be provided at a plurality of locations in the tire circumferential direction.
Further, in the above-described embodiment, the first IMFC ₁ is used as the IMF for calculating the feature amount, but other IMFs may be used. As described above, since it is necessary to pay attention to the high frequency component of the tire vibration in order to discriminate the road surface, it is preferable to use a low number IMF as the IMF for calculating the feature amount.
Further, in the above-described embodiment, the feature amounts are the average μ, the standard deviation σ, and the skewness b ₁ , but other statistics such as the kurtosis b ₂ may be added. Alternatively, a plurality of statistics may be combined from among the mean μ, the standard deviation σ, the skewness b ₁ , the kurtosis b ₂ , and the like.

In addition, as the feature amount used for identifying the road surface state, other feature amounts such as the vibration level of a specific frequency band extracted from the vibration waveform or the calculated value calculated from the vibration levels of a plurality of specific frequency bands are used. You may.
In addition, when the outside air temperature is low (for example, 5 ° C or less) using the temperature information acquired by a temperature sensor or the like, only the rear wheel acceleration waveform is used to determine whether the road surface is a DRY road surface, a SNOW road surface, or an ICE. If it is determined whether the road surface is a DRY road surface or a WET road surface by using only the front wheel acceleration waveform when the outside air temperature is not low, the calculation amount can be further reduced.

10 Road surface condition discriminator, 11 Front wheel side accelerometer,
12 Front wheel acceleration waveform extraction means, 13 Front wheel feature amount calculation means,
13a natural vibration mode extraction unit, 13b feature data calculation unit, 13c feature amount calculation unit,
14 front wheel identification function calculation means, 14a kernel function calculation unit, 14b identification function calculation unit, 15 storage means, 16 road surface condition determination means, 21 rear wheel side acceleration sensor,
22 Rear wheel acceleration waveform extraction means, 23 Rear wheel feature amount calculation means,
24 Rear wheel identification function calculation means,
30F front tires, 30R rear tires, 31 inner liner,
32 Tire air chamber, R road surface.

Claims (2)

It is a method of discriminating the state of the road surface in contact with the tire by the road surface condition discriminating device from the time change waveform of the vibration of the running tire detected by the vibration detecting means mounted on the tire.
With the vibration detecting means mounted on the front wheel of the tire and the vibration detecting means mounted on the rear wheel of the tire, the front wheel acceleration waveform which is the time change waveform of the vibration of the front wheel and the time change waveform of the vibration of the rear wheel are used. A step to detect a certain rear wheel acceleration waveform,
A step of calculating the feature amount of the front wheel and the feature amount of the rear wheel from the front wheel acceleration waveform and the rear wheel acceleration waveform, respectively.
From the calculated feature amount of the front wheel and the feature amount of the rear wheel and the road surface feature amount obtained in advance for each road surface condition, the step of determining the state of the road surface in which the front wheel and the rear wheel are in contact with each other is performed. Prepare,
The feature quantity is a statistic of the distribution of one or both of the instantaneous frequency and the instantaneous amplitude, which is extracted by performing the Hilbert transform on the natural vibration mode acquired by using the algorithm of empirical mode decomposition.
In the step of determining the state of the road surface,
After calculating the kernel function including the front wheel feature amount using the front wheel feature amount and the road surface feature amount, DRY is obtained from the value of the discrimination function using the kernel function including the calculated front wheel feature amount . Distinguish between the road surface and the WET road surface,
After calculating the kernel function including the feature amount of the rear wheel using the feature amount of the rear wheel and the road surface feature amount, the discriminating function using the kernel function including the calculated feature amount of the rear wheel. A road surface condition determination method comprising discriminating between a DRY road surface and an ICE road surface and a DRY road surface and a SNOW road surface from a value , or discriminating between a DRY road surface and an ICE / SNOW road surface . It is a road surface condition determination device that detects the vibration of the running tire and determines the condition of the road surface in contact with the tire.
A front wheel vibration detecting means mounted on the front wheel of the tire and detecting a front wheel acceleration waveform which is a time-varying waveform of the vibration of the front wheel.
A rear wheel vibration detecting means mounted on the rear wheel of the tire and detecting a rear wheel acceleration waveform which is a time-varying waveform of the vibration of the rear wheel.
A feature amount calculating means for calculating the feature amount of the front wheel and the feature amount of the rear wheel from the front wheel acceleration waveform and the rear wheel acceleration waveform, respectively.
A storage means for storing the road surface feature amount calculated by using the time change waveform of the tire vibration obtained in advance for each road surface condition, and a storage means.
It is provided with a road surface condition determining means for determining the state of the road surface in which the front wheel and the rear wheel are in contact from the calculated feature amount and the road surface feature amount.
The feature quantity is a statistic of the distribution of one or both of the instantaneous frequency and the instantaneous amplitude, which is extracted by performing the Hilbert transform on the natural vibration mode acquired by using the algorithm of empirical mode decomposition.
The road surface condition determining means is
After calculating the kernel function including the front wheel feature amount using the front wheel feature amount and the road surface feature amount, DRY is obtained from the value of the discrimination function using the kernel function including the calculated front wheel feature amount . Distinguish between the road surface and the WET road surface,
After calculating the kernel function including the feature amount of the rear wheel using the feature amount of the rear wheel and the road surface feature amount, the discriminating function using the kernel function including the calculated feature amount of the rear wheel. A road surface condition discriminating device characterized by discriminating between a DRY road surface and an ICE road surface, discriminating between a DRY road surface and a SNOW road surface, or discriminating between a DRY road surface and an ICE / SNOW road surface from a value .

Images