JP6734713B2 - Road condition determination method - Google Patents

Description

The present invention relates to how to determine the road surface condition during travel.

Conventionally, as a method for determining the road surface condition, the vibration waveform of the running tire tread detected by the acceleration sensor installed on the inner liner of the tire is used before the stepping on the front side of the peak on the stepping side appearing at the stepping end. A region R1, a stepping region R2 that forms a peak on the stepping side, a pre-starting region R3 between the peak on the stepping side and a peak on the starting side that appears at the starting end, and a peak on the starting side are formed. It is divided into a kick-out area R4 and a kick-out area R5 after the kick-out area R4, and the vibration waveform of each area is frequency-analyzed. From the obtained frequency spectrum, a vibration level in a specific frequency band is obtained. A method for determining a road surface state during traveling based on a value obtained by obtaining a plurality of band values P _ij and substituting these band values P _ij into a previously determined identification function F(P _ij ). Has been proposed (for example, see Patent Document 1).
The band value P _ij indicates a band value in which the area number is i and the frequency band is j.

JP, 2011-242303, A

However, in Patent Document 1, the time width of the regions R1 to R5, the magnitude of the band value P _ij , and the like are not considered to change depending on the tire condition such as the tire internal pressure and the tire internal temperature, and thus the road surface condition. The judgment accuracy of was not sufficient.

The present invention has been made in view of the conventional problems, even if the tire condition is changed, and an object thereof is to provide a way capable of determining road surface conditions accurately.

The present invention is a road surface state determination method for determining a road surface state from a time-varying waveform of vibration of a running tire detected by a vibration detection means, a step of acquiring a vibration waveform of the tire, and a state of the tire. acquiring information, after changing Te discrimination parameters for determining a road surface condition obtained from the vibration waveform to the status information, and Luz step to determine the road surface condition from the modified determining parameter , And the state information is one or more of the internal pressure of the tire, the internal temperature of the tire, and the external temperature of the tire, and the determination parameter is the time width of the window applied to the vibration waveform. It is characterized by
As this, according to the acquired status information of the tire, since the change the time width of the window to be applied to the vibration waveform, it is possible to improve the determination accuracy of the road surface condition.
Further , since the state information is one or more of the internal pressure of the tire, the internal temperature of the tire, and the external temperature of the tire, the state of the tire can be accurately grasped .

The above summary of the invention does not enumerate all the necessary features of the present invention, and sub-combinations of these feature groups can also be the invention.

It is a figure which shows the structure of the road surface state determination device which concerns on Embodiment 1 of this invention. It is a figure which shows the example of arrangement|positioning of an acceleration sensor. It is a figure which shows an example of the time series waveform of vibration, and the extraction area|region of a band value. It is a figure which shows the relationship between a zone value and a tire internal pressure. 3 is a flowchart showing a road surface state determination method according to the first embodiment. It is a figure which shows the relationship between an appropriate frequency band and a tire internal pressure. It is a figure which shows the structure of the road surface state determination device which concerns on this Embodiment 2. It is a figure which shows the extraction method of the time series waveform of the tire vibration for every time window. It is a figure which shows an example of a road surface HMM. It is a figure which shows the road surface HMM used for likelihood calculation. It is a schematic diagram of a state transition series. 7 is a flowchart showing a road surface state determination method according to the second embodiment. It is a figure which shows the structure of the road surface state determination device which concerns on this Embodiment 3. It is a schematic diagram which shows an input space. It is a figure which shows the DRY road surface feature vector and the road surface feature vector other than a DRY road surface in an input space. It is a figure which shows the calculation method of GA kernel of a DRY road surface feature vector and a road surface feature vector other than a DRY road surface. It is a figure which shows the calculation method of GA kernel of the calculated feature vector and road surface feature vector. 7 is a flowchart showing a road surface state determination method according to the third embodiment.

Embodiment 1.
FIG. 1 is a functional block diagram of a road surface state determination device 10 according to the present embodiment. In FIG. 1, 11 is an acceleration sensor as vibration detection means, 12 is an internal pressure sensor as tire state detection means, and 13 is a tire state. Judgment means, 14 is a vibration waveform detection means, 15 is a region signal extraction means, 16 is a band value calculation means, 17 is a band value correction means, and 18 is a road surface state determination means.
The acceleration sensor 11 and the internal pressure sensor 12 configure a sensor unit 10A, and the vibration waveform detection unit 13 to the road surface state determination unit 18 configure a storage/calculation unit 10B.
Each unit that configures the storage/calculation unit 10B includes, for example, computer software and a storage device such as a RAM.

As shown in FIG. 2, the acceleration sensor 11 is arranged in the tire width direction center of the inner liner portion 2 of the tire 1 so that the detection direction is the tire circumferential direction. As a result, the acceleration sensor 11 detects the acting tire circumferential acceleration input from the road surface to the tread 3. Hereinafter, the position of the acceleration sensor 11 (strictly speaking, the position of the surface of the tread 3 that is radially outside the acceleration sensor 11) is referred to as a measurement point.
For example, the output of the acceleration sensor 11 is sent by the transmitter 11F to the vibration waveform detecting means 14 of the storage/calculation unit 10B (not shown) provided on the vehicle body side. Further, the determination result of the road surface state determination device 10 is sent to the vehicle control device 20 arranged on the vehicle body side.
Further, the internal pressure sensor 12 is provided integrally with the acceleration sensor 11 and measures the internal pressure P of the tire 1 (hereinafter referred to as the tire internal pressure) P. The measured tire internal pressure P is sent by the transmitter 11F to the tire condition determination means 13 of the storage/calculation unit 10B.
The storage/calculation unit 10B may be provided on the tire 1 side and the determination result of the road surface state determination device 10 may be transmitted to the vehicle control device 20 provided on the vehicle body side.

The tire condition determining means 13 determines from the tire internal pressure P measured by the internal pressure sensor 12 whether or not the condition of the tire 1 is a condition enabling determination of the road surface condition. Specifically, it is determined whether or not the measured tire internal pressure P is within a preset determinable internal pressure range [P _min , P _Max ], and P<P _min or P>P _Max . In this case, since it is difficult to determine the road surface state from the detected vibration waveform, an interruption command signal for interrupting the detection of the vibration waveform is sent to the vibration waveform detection means 14, and the vehicle control device 20 detects the interruption. It sends an undecidable signal indicating that the accuracy of the road surface condition determined from the vibration waveform is low.
On the other hand, when the measured tire internal pressure P is within the determinable internal pressure range [P _min , P _Max ], the measured data of the tire internal pressure P is output to the band value correction means 17.
The determinable internal pressure range [P _min , P _Max ] varies depending on the tire type. For example, if the standard internal pressure is P ₀ , P _min =(1-0.3)P ₀ , P _Max =(1+0. 3) It is preferable that P ₀ or P _min =P ₀ -100 kPa and P _Max =P ₀ +100 kPa.

The vibration waveform detection means 14 detects a time-series waveform in which tire circumferential vibrations that are output from the acceleration sensor 11 and that are input to the running tire 1 are arranged in time series. As shown in FIG. 3, a peak (positive peak) P _f that first appears in the time-series waveform of vibration is a peak that occurs when the measurement point collides with the road surface, and the position of this peak P _f is the depression point. P _f . The next peak (negative peak) P _{k that} appears when the measurement point leaves the road surface, and the position of this peak P _k is the start point.
Domain signal extracting means 15, a time-series waveform detected by the vibration waveform detection means 13, than the peak P _f of leading side is before the area with the depression front region R1, a region for forming the peak P _f of leading side A certain stepping region R2, a pre-kicking region R3 that is a region between the stepping side peak P _f and the kicking side peak P _k, and a kicking region that is a region that forms the kicking side peak P _k It is divided into R4 and a region after kicking R5, which is a region after the region R4 after kicking, and a time series waveform of vibration in each of the regions R1 to R5 is extracted.
When the interruption instruction signal is input from the tire condition determination means 13, the time series waveform extraction operation is interrupted.

The band value calculating means 16 passes the time-series waveforms of the respective regions R1 to R5 through a bandpass filter to calculate the band value A _ij which is the magnitude of the vibration component in the predetermined frequency region. The subscript i indicates the time-series waveform regions R1 to R5, and the subscript j indicates the extracted frequency region.
For example, A ₁₁ is a band value selected from the 2 kHz to 8 kHz band of the pre-depressing region R1, A ₂₃ is a band value selected from the 4 kHz to 10 kHz band of the depressing region R2, and A ₅₂ is 2 kHz of the after-start region R5. Is a band value selected from the band of 4 kHz.
This band value A _ij corresponds to the discrimination parameter of the present invention.
The band value correction means 17 corrects the band value A _ij calculated by the band value calculation means 16 using the data of the tire internal pressure P sent from the tire condition determination means 13. Specifically, a graph showing the relationship between the tire internal pressure P and the band value ratio K _ij =a _ij (P)/a _ij (P ₀ ) as shown in FIG. 4 is prepared, and when the internal pressure is P The band value A _ij (P) of is corrected.
Letting the corrected band value be A _ij (P), A _ij (P)=K _ij ×A _ij .
It should be noted that all the band values A _ij do not increase as the tire internal pressure P increases, as shown in FIG. Since the band value A _ij may decrease as the tire internal pressure P increases, as described above, the relationship between the band value A _ij and the tire internal pressure P is the relationship between the tire internal pressure P and the band value ratio K _ij. It is necessary to prepare the data shown for each extraction region Ri and extraction frequency region j.
The method for correcting the band value A _ij is not limited to the method described above, and as shown in FIG. 4B, a straight line indicating the relationship between A _ij and A _ij (P) for each tire internal pressure P or obtained in advance curve, the a _ij on the line or on a curve corresponding to the measured tire pressure P of a _ij (P) may be used as the correction value.

The road surface state determining unit 18 replaces the _Aij calculated by the band value calculating unit 16 with a plurality of preset discriminating functions Fk( _Aij ), and the band value _Aij (corrected by the band value correcting unit 17). The road surface state is estimated using the function value fk obtained by substituting P).
In this example, as the discriminant function Fk, the discriminant function F1=which is proposed by the present applicant in Japanese Patent Application No. 2010-115730 (Patent Document 1) for determining whether or not there are inclusions such as water and snow on the road surface w ₁₁ ·A ₁₁ +w ₁₂ ·A ₁₂ −K 1 and the discriminant function F 2 =w ₂₁ ·A ₂₁ +w ₂₂ ·A ₅₁ −K 2 for determining whether the road surface is a snowy road and inclusions on the road surface Discriminant function F3=w ₃₁ ·A ₅₂ +w ₃₂ ·A ₃₁ +w ₃₃ ·A ₄₁ +w ₃₄ ·A for judging whether it is water or snow, that is, whether the road surface is a deep WET surface or a deep sherbet-like snow road ₅₃ is used to determine whether the road surface is a "snow road" and whether the "snow road" is a "snowy road" or a "shallow sorbet-like snowy road".

Next, a road surface condition determining method according to the first embodiment will be described with reference to the flowchart of FIG.
First, the acceleration sensor 11 detects the tire circumferential vibration of the running tire 1, and the internal pressure sensor 12 measures the tire internal pressure (step S10).
Next, it is determined from the measured tire internal pressure whether or not the tire internal pressure P is within a preset determinable internal pressure range [P _min , P _Max ] (step S11).
If the tire internal pressure P is within the determinable internal pressure range [P _min , P _Max ], the process proceeds to step S12, and a time-series waveform in which tire circumferential vibration output from the acceleration sensor 11 is arranged in time series is detected. To do. On the other hand, when the measured tire internal pressure P is P<P _min or P>P _Max , the road surface determination operation is stopped. In addition, after a predetermined time has elapsed, it may be determined again whether or not the measured tire internal pressure P is within the determinable internal pressure range [P _min , P _Max ].

Next, after dividing the detected time-series waveform into a pre-stepping-in area R1, a stepping-in area R2, a pre-kicking-out area R3, a kicking-out area R4, and a post-kicking-out area R5 (step S13), The band value A _ij is calculated from the time series waveform of the vibration in each of the regions R1 to R5 (step S14).
The band value A _ij is a band value in which the extraction region of the time-series waveform is Ri and the frequency region is [f _ja , f _jb ]. For example, A ₂₃ is a band value in the frequency region of the stepping region R2 [4 kHz, 10 kHz].
In the present invention, the calculated band value A _ij is corrected using the data of the tire internal pressure P (step S15). The correction is performed for each band value A _ij .
Next, it is determined whether or not there are inclusions such as water and snow on the road surface (step S16).
Specifically, the relationship between the band value A ₁₁ and the band value A _{12 under} various road surface conditions is experimentally obtained in advance, and the discriminant function F1=w ₁₁ ·A ₁₁ +w ₁₂ ·A ₁₂ −K 1 is set. The band value A ₁₁ (P) and the band value A ₁₂ (P) corrected with the tire internal pressure P are respectively substituted into the band value A ₁₁ and the band value A ₁₂ of the discriminant function F1 to obtain the function value f1.
When the function value f1 satisfies f1≧0, it is determined that there are inclusions such as water and snow on the road surface, and therefore the process proceeds to step S17, where the inclusions on the road surface are covered with fresh snow. Determine if it is a soft pile. On the other hand, when f1<0, it is determined that the road surface is “not snowy”.

In step S17, it is determined whether or not the road is a snowy road covered with fresh snow.
Specifically, to set the discrimination function _{F2 = w 21 · A 21 +} w 22 · A 51 -K2 in advance experimentally obtained relation between bandwidth value A ₂₁ and band value A ₅₁ in the various road conditions, the The band value A ₂₁ (P) and the band value A ₅₁ (P) corrected by the tire internal pressure P are respectively substituted into the band value A ₂₁ and the band value A ₅₁ of the discriminant function F2 to obtain the function value f2. When f2<0, it is determined that the road surface is a snowy road.
On the other hand, when the function value f2 is f2≧0, the process proceeds to step S18, and the inclusion on the road surface is water or snow, that is, the road surface is a deep WET road surface or a deep sherbet-like snow road. Determine whether or not.
That is, in step S18, the pre-discrimination function _{F3 = w 31 · A 52 +} w 32 · A 31 + w 33 · A 41 + w 34 · relative bandwidth value _{A 52, A 31, A 41} , A 53 at various road conditions A ₅₃ is set, and the band values A ₅₂ , A ₅₃₁ , A ₄₁ and the band value A ₅₃ of the discriminant function F3 are corrected by the tire internal pressure P, respectively, band values A ₅₂ (P), A ₃₁ (P), A _The function value f3 obtained by substituting ₄₁ (P) and the band value A ₅₃ (P) is obtained. If f3≧0, it is determined that the road is a deep WET road (not a snow road), and if f3<0, it is determined that the road surface is a sherbet-like snow road.

As described above, in the first embodiment, the acceleration sensor 11 detects the tire circumferential vibration of the running tire 1, the internal pressure sensor 12 measures the tire internal pressure P, and the measured tire internal pressure P Is within the determinable internal pressure range [P _min , P _Max ], the band value A _ij obtained from the time-series waveform of the tire circumferential vibration, which is the vibration information, is corrected by the tire internal pressure P, and this correction is performed. Since the road surface condition is determined using the band value A _ij (P), the accuracy of determining the road surface condition can be improved.
Further, when the tire internal pressure P is outside the determinable internal pressure range [P _min , P _Max ], the determination of the road surface condition is interrupted, so that the erroneous determination of the road surface condition can be prevented.

In the first embodiment, the tire condition is the tire inner pressure, but it may be the tire inner temperature or both the tire inner pressure and the tire inner temperature. Alternatively, the tire outer surface temperature may be added to either or both of the tire inner pressure and the tire inner temperature.
The temperature inside the tire may be measured by disposing a temperature sensor 19 on the tire air chamber 5 side of the wheel rim 4 or on the inner liner portion 2 of the tire 1 as shown in FIG. Further, the tire outer surface temperature may be measured by, for example, a temperature sensor installed in a tire house of a vehicle body (not shown) at a position facing the tire 1.
Further, in the first embodiment, the discrimination parameter to be corrected according to the tire condition is set as the pre-depression area R1, the depressing area R2, the pre-kicking area R3, the kicking area R4, and the after-kicking extracted from the time-series waveform. The band value A _ij , which is the magnitude of the vibration component in the predetermined frequency region calculated from the time-series waveform of the region R5, is used. It may be the function value fk. Alternatively, it may be w _kl (weight level) which is a coefficient of the discriminant function Fk, or a constant K _k .
The case where the discriminant parameter to be corrected according to the tire condition is the function value fk of the discriminant function Fk corresponds to the first discriminating step for discriminating the road surface condition from the discriminant parameter and condition information of the present invention.

Further, the discrimination parameter may be the range of the specific frequency band.
For example, as shown in FIG. 6, since the upper limit frequency f _ja and the lower limit frequency f _jb of the specific frequency region, or the proper value of the region width (f _jb -f _ja ) of the specific frequency region changes depending on the tire internal pressure P. By changing these values according to the tire internal pressure P, the road surface condition determination accuracy can be improved.
Further, the time width of the window applied to the vibration waveform may be used as the discrimination parameter. That is, the position of the peak P _f on the stepping side of the vibration waveform, the position of the peak P _k on the kicking side, or the position of the interval between the peak P _f on the stepping side and the peak P _k on the kicking side also determines the tire internal pressure and the tire. Since it changes depending on the internal temperature, it is applied to the vibration waveform that is the region width of the extraction region of the vibration waveform (pre-depression region R1, depressing region R2, pre-kicking region R3, kicking-out region R4, and post-kicking region R5). If the time width of the window is changed according to the tire information, the road surface condition determination accuracy can be further improved.
Further, by simultaneously correcting and changing a plurality of determination parameters, it is possible to further improve the determination accuracy of the road surface condition.

Embodiment 2.
FIG. 7 is a functional block diagram of a road surface state determination device 30 according to the second embodiment. In FIG. 7, 11 is an acceleration sensor as vibration detection means, 12 is an internal pressure sensor as tire state detection means, and 13 is a tire. State determination means, 14 is a vibration waveform detection means, 31 is a windowing means, 32 is a feature vector calculation means, 33 is a feature vector correction means, 34 is a storage means, 35 is a likelihood calculation means, and 36 is a road surface state determination means. is there.
The acceleration sensor 11 and the internal pressure sensor 12 compose a sensor unit 10A, and the tire condition determination unit 13, the vibration waveform detection unit 14, and the window mounting unit 31 to the road surface condition determination unit 36 are storage/calculation units 30B. Make up.
Each unit that configures the storage/calculation unit 30B includes, for example, computer software and a storage device such as a RAM.
The acceleration sensor 11, the internal pressure sensor 12, the tire state determination means 13, and the vibration waveform detection means 14 having the same reference numerals as those in the first embodiment are the same as those in the first embodiment, and the acceleration sensor 11 is a running tire. 1, the tire circumferential direction vibration is detected, and the internal pressure sensor 12 measures the tire internal pressure.
The tire condition determining means 13 determines whether or not the measured tire internal pressure P is within a preset determinable internal pressure range [P _min , P _Max ] based on the measured tire internal pressure data, and a vibration waveform. The detection means 14 detects a time series waveform in which tire circumferential vibrations are arranged in time series as shown in FIG.

As shown in FIG. 8, the windowing means 31 windows the time series waveform of the tire circumferential vibration with a preset time width (time window width), and extracts the time series waveform of the tire vibration for each time window. ..
The feature vector calculation means 32 calculates the feature vector X _t for each of the extracted time series waveforms of the respective time windows. In this example, as the feature vector X _t , the time-series waveform of the tire vibration is extracted through the k band-pass filters BP(k) each having a frequency domain of f _ka −f _kb and extracted. Vibration level (power value of filtered wave) x _kt was used. Since the dimension of the feature vector X is k-dimensional, and in this example, the specific frequency band is set to 6 of 0-0.5kHz, 0.5-1kHz, 1-2kHz, 2-3kHz, 3-4kHz, and 4-5kHz, k =6.
Since the feature vector X _t is obtained for each time window, if the total number of time windows is N, the number of feature vectors X _t will also be N.
The feature vector correction means 33 uses the data of the tire internal pressure P sent from the tire condition determination means 13 to calculate the N×k power values x _kt (hereinafter referred to as power value x _kt ) calculated by the feature vector calculation means 32. Correct using. The correction method is the same as in the first embodiment, and a graph showing the relationship between the power value ratio and K _k =x _kt (P)/x _kt (P ₀ ) is prepared, and the power when the internal pressure is P is set. The value x _kt is corrected to obtain the corrected feature vector X _t (P).
If the component of the corrected feature vector X _t (P) is x _kt (P), then x _kt (P)=K _k ×x _kt .

The storage unit 34 stores a plurality of hidden Markov models (hereinafter referred to as road surface HMMs) configured for each road surface state. The road surface HMM includes an inside road surface HMM (road) and an outside road surface HMM (silent). The in-road HMM (road) is composed of a vibration waveform that appears in the road surface region in the time-series waveform of tire vibration, and the out-of-road surface HMM (silent) is composed of a vibration waveform in the non-information region.
As shown in FIG. 9, the road surface HMM has seven states S _{1 to} S ₇ corresponding to the time-series waveform of tire vibration, and each state S _{1 to} S ₇ has an output probability of the feature vector X _t . b _ij (X) and transition probability between states a _ij (X) are used (i, j=1 to 7).
In this embodiment, by performing the learning of dividing the time series waveform of the tire vibration in five states in five states S ₂ to S ₆ except the starting state S ₁ of each road HMM and ending state S _7, the road surface The output probability b _ij (X) of the HMM feature vector X and the transition probability a _ij (X) between states were obtained.
The output probability b _ij (X) represents the probability that the feature vector X will be output when the state transits from the state S _i to the state S _j . The output probability b _ij (X) assumes a mixed normal distribution.
The transition probability a _ij (X) represents the probability that a state will transition from state S _i to state S _j .
When the dimension of the feature vector X is k, the output probability b _ij is set for each k component x _k of the feature vector X.

In this example, the time-series waveform data of tire vibration obtained by running a vehicle equipped with the tire 1 equipped with the acceleration sensor 11 in advance on each of DRY, WET, SNOW, and ICE road surfaces is learned. As the data for use, five road surface HMMs including a DRY road surface HMM, a WET road surface HMM, a SNOW road surface HMM, and an ICE road surface HMM, that is, four road surface HMMs (road) and road outside HMMs (silent), were constructed.
Both the road surface HMM (road) and the road surface HMM (silent) are HMMs having seven states S _{1 to} S ₇ including a start state S ₁ and an end state S ₇ .
HMM learning is performed by a known method such as an EM algorithm, a Baum-Welch algorithm, and a forward-back-forward algorithm.

Ｘ；データ系列
ｔ；時刻
Ｓ；状態数
Ｍ_s；混合ガウス分布の成分の数
ｃ_jsm；ｍ番目の混合成分の混合比
μ；ガウス分布の平均ベクトル
σ；ガウス分布の分散共分散行列
遷移確率π（Ｘ_t）は、路面ＨＭＭが７状態であるので、７×７の行列で表わせる。この遷移確率π（Ｘ_t）としては、前記路面ＨＭＭの学習により求められた特徴ベクトルＸ_tの状態間の遷移確率ａ_ij（Ｘ_t）を用いればよい。
そして、算出した出力確率Ｐ（Ｘ_t）と遷移確率π（Ｘ_t）との積である時間窓毎の出現確率Ｋ（Ｘ_t）を求め、この時間窓毎の出現確率Ｋ（Ｘ_t）を全ての時間窓について掛け合わせて尤度Ｚを求める。すなわち、尤度Ｚは、Ｚ＝ΠＰ（Ｘ_t）・遷移確率π（Ｘ_t）により求められる。あるいは、それぞれの時間窓毎に計算された出現確率Ｋ（Ｘ_t）の対数をとって、全ての時間窓について足し合わせることで尤度Ｚを求めてもよい。 As shown in FIG. 10, the likelihood calculating means 35 calculates the likelihood of the corrected feature vector X _t (P) for each of the plurality (here, four) of road surface HMMs.
As the likelihood, as proposed by the present applicants in Japanese Patent Application No. 2011-140943, first, the output probability P(X _t (P)) is calculated for each time window using the following equations (1) and (2). To calculate. In the following, X _t is the corrected feature vector X _t (P).

X; data series t; time S; number of states M _s ; number of components of mixed Gaussian distribution c _jsm ; mixture ratio of m-th mixed component μ; mean vector of Gaussian distribution σ; variance-covariance matrix of Gaussian distribution transition probability π(X _t ) can be represented by a 7×7 matrix because the road surface HMM is in seven states. As the transition probability π(X _t ), the transition probability a _ij (X _t ) between the states of the feature vector X _t obtained by learning the road surface HMM may be used.
The calculated output probabilities P (X _t) and the transition probabilities π seeking (X _t) the emergence of each time window which is the product of the probability K (X _t), occurrence probability K (X _t) of the time per window Is multiplied for all time windows to obtain the likelihood Z. That is, the likelihood Z is obtained by Z=ΠP(X _t )·transition probability π(X _t ). Alternatively, the likelihood Z may be obtained by taking the logarithm of the appearance probability K(X _t ) calculated for each time window and adding them up for all time windows.

By the way, as shown in FIG. 11, there are a plurality of routes (state transition series) through which the state of the road surface HMM transits from state S ₁ to state S ₇ . That is, for each road surface HMM, the likelihood Z differs for each state transition sequence.
In this example, a well-known Viterbi algorithm is applied to obtain a state transition series Z _M having the largest likelihood Z, and this state transition series is used as a state transition series corresponding to a detected time series waveform of tire vibration. Let the likelihood Z _{M be} Z of the road surface HMM.
The likelihood Z _M is calculated for each road surface HMM.
The road surface state determination unit 36 compares the likelihoods of the plurality of hidden Markov models calculated by the likelihood calculation unit 35, and the tire travels in the road surface state corresponding to the hidden Markov model having the largest likelihood. It is determined that the road surface is on the existing road surface.

Next, a road surface condition determination method according to the second embodiment will be described with reference to the flowchart in FIG.
First, the acceleration sensor 11 detects the tire circumferential vibration of the running tire 1, and the internal pressure sensor 12 measures the tire internal pressure (step S20).
Next, it is determined from the measured tire internal pressure data whether or not the measured tire internal pressure P is within a preset determinable internal pressure range [P _min , P _Max ] (step S21).
If the tire internal pressure P is within the determinable internal pressure range [P _min , P _Max ], the process proceeds to step S22, and a time-series waveform in which the tire circumferential vibration output from the acceleration sensor 11 is arranged in time series is detected. After that, the time-series waveform that is the tire vibration data is windowed with a preset time window, and the time-series waveform of the tire vibration is extracted for each time window (step S23).
In this example, the time window width is set to 2 msec.
On the other hand, when the measured tire internal pressure P is P<P _min or P>P _Max , the extraction of the time series waveform of tire vibration is interrupted. In addition, after a predetermined time has elapsed, it may be determined again whether or not the measured tire internal pressure P is within the determinable internal pressure range [P _min , P _Max ].

In step S24, the feature vector _Xt =( _x1t , _x2t , _x3t , _x4t , _x5t , _x6t ) is calculated for each of the extracted time-series waveforms of the respective time windows.
In the present invention, the power values x _{1t to} x _6t of the filter filtered wave of the time series waveform of the tire vibration, which is the component of the calculated feature vector X _t , are corrected using the data of the tire internal pressure P (step S25). .. The correction is performed for each power value x _kt (k=1 to 6).
After the correction of the power value x _kt , first, for the DRY road surface HMM that is the first model, the appearance probability K(X _t )=output probability P(X _t )×transition probability π(X _t ) for each time window After obtaining (step S26), this appearance probability K(X _t ) is multiplied for all time windows to calculate the likelihood Z1 in the DRY road surface HMM (step S27).
Next, it is determined whether or not the likelihood Z has been calculated for all models (step S28), and if not completed, the process returns to step S26 and the likelihood in the next model WET road surface HMM is calculated. Calculate Z2.
When the calculation of the likelihood Z of all five models is completed, the process proceeds to step S29, and the road surface state is determined. Specifically, the likelihoods Z1 to Z5 calculated for each road surface HMM are compared, and the road surface state corresponding to the road surface HMM having the largest likelihood is set as the road surface state of the road surface on which the tire is running.

As described above, in the second embodiment, the acceleration sensor 11 detects the tire circumferential vibration of the running tire 1, the internal pressure sensor 12 measures the tire internal pressure P, and the measured tire internal pressure P Is within the determinable internal pressure range [P _min , P _Max ], the characteristic vector X _t calculated from the time-series waveform of the tire vibration for each time window extracted by windowing by the windowing means 31. the power value x _kt of filtration wave is a component, after correcting using the data of the tire internal pressure P, the feature vector X _t (P to the corrected power value x _kt (P) components of filtration wave ) Is used to determine the road surface state, it is possible to improve the road surface state determination accuracy.
Further, when the tire internal pressure P is outside the determinable internal pressure range [P _min , P _Max ], the determination of the road surface condition is stopped, so that the erroneous determination of the road surface condition can be prevented.

In the second embodiment, the discrimination parameter to be corrected according to the tire condition is the power value x of the filtered wave of the time series waveform of the tire vibration which is the component of the feature vector X _t of the time series waveform extracted in the time window. _{Although kt} is used, the magnitude of the likelihood Z which is a discriminant function may be used as the discriminant parameter. Alternatively, the output probability P(X _t ) of the feature vector X of each road surface HMM, the transition probability π(X _t ), the number of states S, the number of components of mixed Gaussian distribution M _s , the mixing ratio of mixed components c _jsm , and Gaussian distribution It may be the weight vector of the discriminant function such as the mean vector μ, the variance covariance matrix σ of the Gaussian distribution, or the intermediate parameter of the weight vector.
The case where the discriminant parameter to be corrected according to the tire condition is the magnitude of the likelihood Z corresponds to the first discriminating step for discriminating the road surface condition from the discriminant parameter and the condition information of the present invention.
Parameters such as the output probability P(X _t ) and the transition probability π(X _t ) corrected by the tire internal pressure P are obtained by learning the road surface HMM.
Further, the proper values of the bandpass filter BP(k) frequency region f _ka −f _kb for obtaining the power value x _kt vary depending on the tire internal pressure P. Therefore, these parameters are changed according to the tire internal pressure P. Then, the accuracy of determining the road surface condition can be improved.
Further, the discrimination parameter may be the time window width. When the tire internal pressure P is high, the time window width is set to be wider, and the time width (time window width) to be windowed on the time-series waveform of the tire circumferential vibration is changed according to the tire information. It is possible to improve the state determination accuracy.
Further, by simultaneously correcting and changing a plurality of determination parameters, it is possible to further improve the determination accuracy of the road surface condition.

Embodiment 3.
FIG. 13 is a functional block diagram of a road surface state determination device 40 according to the third embodiment. In FIG. 13, 11 is an acceleration sensor as vibration detection means, 12 is an internal pressure sensor as tire state detection means, and 13 is a tire. State determination means, 14 is a vibration waveform detection means, 31 is a windowing means, 32 is a feature vector calculation means, 33 is a feature vector correction means, 41 is a storage means, 42 is a kernel function calculation means, and 43 is a road surface state determination means. is there.
The acceleration sensor 11 and the internal pressure sensor 12 constitute a sensor unit 10A, and the tire condition determination unit 13, the vibration waveform detection unit 14, and the window mounting unit 31 to the road surface condition determination unit 43 are stored/calculated in the storage unit 40B. Make up.
Each unit that configures the storage/calculation unit 40B includes, for example, computer software and a storage device such as a RAM.
The acceleration sensor 11, the internal pressure sensor 12, the tire state determination means 13, the vibration waveform detection means 14, and the windowing means to the feature vector correction means 33, which have the same reference numerals as those in the first and second embodiments, are used in the embodiments. The acceleration sensor 11 detects the vibration in the tire circumferential direction of the running tire 1, and the internal pressure sensor 12 measures the tire internal pressure.

In addition, the tire condition determination means 13 determines whether or not the measured tire internal pressure P is within a preset determinable internal pressure range [P _min , P _Max ] from the measured tire internal pressure data, The vibration waveform detecting means 14 detects a time series waveform in which tire circumferential vibrations are arranged in time series.
As shown in FIG. 8, the windowing means 31 windows the time series waveform of the tire circumferential vibration with a preset time width (time window width), and extracts the time series waveform of the tire vibration for each time window. ..
The feature vector calculation means 32 calculates the feature vector X _t for each of the extracted time series waveforms of the respective time windows.
In this example, a time series waveform of tire vibration was obtained as a feature vector X _t through a bandpass filter of 0-1 kHz, 1-2 kHz, 2-3 kHz, 3-4 kHz, and 4-5 kHz, respectively. The vibration level (power value of the filtered wave) x _kt (k=1 to 6) in a specific frequency band was used.
Since the feature vector X _t is obtained for each time window, if the total number of time windows is N, the number of feature vectors X _t will also be N. Hereinafter, the feature vector of the window number i as X _i, the power value is a component of the X _i is described as x _ki.
The characteristic vector correction means 33 corrects the N×k power values x _ki calculated by the characteristic vector calculation means 32 by using the data of the tire internal pressure P sent from the tire condition determination means 13 to make correction. The calculated feature vector X _i (P) is obtained.
The correction method is the same as in the first embodiment, and a graph showing the relationship between the power value ratio and K _k =x _ki (P)/x _ki (P ₀ ) is prepared, and the power when the internal pressure is P is set. Correct the value x _ki .
The component x _ki (P) of the corrected feature vector X _i (P) is x _ki (P)=K _k ×x _ki .

FIG. 14 is a schematic diagram showing the input space of the feature vector X _i , in which each axis represents the vibration level a _ik of the specific frequency band that is the feature amount, and each point represents the feature vector X _i . Since the actual input space has six specific frequency bands, it becomes a seven-dimensional space when combined with the time axis, but the figure is expressed in two dimensions (horizontal axis is a ₁ and vertical axis is a ₂ ).
For example, if the vehicle is traveling on a DRY road surface, the points forming the group C can be distinguished from the group C′ composed of the feature vectors X′ _i calculated when the vehicle is traveling on the SNOW road surface. If possible, it can be determined whether the vehicle is traveling on the DRY road surface or the SNOW road surface.
The storage means 41 represents the DRY road surface and other road surfaces, the WET road surface and other road surfaces, the SNOW road surface and other road surfaces, the ICE road surface and the other road surfaces, which have been obtained in advance, as hyperplanes. The four road surface models for separation by the discriminant function f(x) are stored.
The road surface model is calculated from a time-series waveform of tire vibration obtained by running a test vehicle equipped with a tire with an acceleration sensor attached thereto on various road surfaces of DRY, WET, SNOW, and ICE at various speeds. The road surface feature vector Y _ASV (y _jk ) which is the feature vector for each time window is used as input data and is obtained by learning.
The tire size used for learning may be one kind or plural kinds.
The subscript A of the road surface feature vector Y _ASV (y _jk ) indicates DRY, WET, SNOW, and ICE. The subscript j (j=1 to M) indicates the number of time-series waveforms (window number) extracted in the time window, and the subscript k indicates the vector component. That is, y _jk =(a _j1 , a _j2 , a _j3 , a _j4 , a _j5 , a _j6 ). Further, SV is an abbreviation of support vector and represents data in the vicinity of the discrimination boundary selected by learning.
Hereinafter, the road surface feature vector Y _ASV (y _jk ) will be simply referred to as Y _ASV .

The calculation method of each road surface characteristic vector Y _ASV is the same as that of the above-mentioned characteristic vector X _j . For example, in the case of the DRY road surface characteristic vector Y _DSV , the time series waveform of the tire vibration when traveling on the DRY road surface is windowed with the time width T. The time series waveform of the tire vibration is extracted for each time window, and the DRY road surface feature vector Y _{D is} calculated for each of the extracted time series waveforms of the respective time windows. Note that the number of dimensions of the vector y _i of the DRY road surface characteristic vector Y _D is 6 dimensions like the feature vector X _i . Then, the support vector Y _DSV is selected by learning Y _D as learning data by the support vector machine (SVM). Note that it is not necessary to store all Y _D in the storage means 41, and only the selected Y _DSV should be stored.
The WET road surface feature vector Y _WSV , the SNOW road surface feature vector Y _SSV , and the ICE road surface feature vector Y _ISV can be obtained in the same manner as the DRY road surface feature vector Y _DSV .
Here, it is important that the time width T has the same value as the time width T when the feature vector X _j is obtained. If the time width T is constant, the number M of time-series waveforms in the time window differs depending on the tire type and the vehicle speed. That is, the number M of time-series waveforms in the time window of the road surface feature vector Y _ASV does not necessarily match the number N of time-series waveforms in the time window of the feature vector X _j . For example, even if the tire types are the same, M>N when the vehicle speed when _obtaining the feature vector X _j is slower than the vehicle speed when _{obtaining the} DRY road surface feature vector Y _DSV, and M<N when it is faster. Become.

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。
このとき、内積φ（ｘ_α）φ（ｘ_β）をカーネル関数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。なお、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。
ラグランジュ乗数λは、前記の式（２）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。このとき、カーネル関数を使っているので、高次元の内積を直接求める必要がない。したがって、計算時間を大幅に縮減できる。 The road surface model is constructed by SVM using each road surface feature vector Y _A as learning data, as proposed by the present applicants in Japanese Patent Application No. 2012-176779.
FIG. 15 is a conceptual diagram showing a DRY road surface feature vector Y _DSV and a road surface feature vector Y _nDSV other than the DRY road surface in the input space, in which the black circles are the DRY road surface and the light circles are the road surfaces other than the DRY road surface. It is a feature vector.
Although both the DRY road surface feature vector and the road surface feature vector other than the DRY road surface are matrices, the DRY road surface feature vector and the road surface feature vector other than the DRY road surface are shown in FIG. 14 in order to explain how to obtain the identification boundary of the group. Each is shown as a two-dimensional vector.
The discriminating boundaries of groups are generally not linearly separable. Therefore, by using the kernel method, the road surface feature vectors Y _DSV and Y _nDSV are mapped to the high-dimensional feature space by the non-linear mapping φ to perform linear separation, so that the road surface feature vectors Y _DSV and Y _nDSV are obtained in the original input space. For non-linear classification.
Specifically, the optimum discriminant function f(x) for discriminating the data is obtained by using the data set X=(x ₁ , x ₂ ,... X _n ) and the belonging class z={1, −1}. =w ^T φ(x)-b is obtained. Here, the data are road surface feature vectors Y _Dj , Y _nDj , and the belonging class is data of the DRY road surface where z=1 is shown by χ ₁ in the figure, and z=−1 is the data of the road surface other than the DRY road surface shown by χ ₂ . The data. Further, w is a weighting coefficient, b is a constant, and f(x)=0 is an identification boundary.
The discriminant function f(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 (3) and (4).

Here, α and β are indices of a plurality of learning data. Further, λ is a Lagrange multiplier, and λ>0.
At this time, the discriminant function f(x)=w ^T φ(x)-b can be made non-linear by replacing the inner product φ(x _α ) φ(x _β ) with the kernel function K(x _α , x _β ). Note that φ(x _α )φ(x _β ) is an inner product after mapping x _α and x _β into a high-dimensional space by the mapping φ.
The Lagrange multiplier λ can be obtained by using an optimization algorithm such as the steepest descent method or SMO (Sequential Minimal Optimization) with respect to the equation (2). At this time, since the kernel function is used, it is not necessary to directly obtain the high-dimensional inner product. Therefore, the calculation time can be greatly reduced.

In this example, a global alignment kernel function (GA kernel) is used as the kernel function K(x _α , x _β ). The GA kernel K(x _α , x _β ) is, as shown in FIG. 16 and the following equations (5) and (6), a DRY road surface feature vector x _i =Y _Di and a road surface feature vector x _j =other than the DRY road surface. It is a function consisting of the sum or product of local kernels κ _ij (x _i , x _j ) indicating the degree of similarity with Y _nDj, and it is possible to directly compare time series waveforms having different time lengths.
The local kernel κ _ij (x _i , x _j ) is obtained for each window of the time interval T.
In order to distinguish the DRY road surface and the road surface other than the DRY road surface from the discriminant function f(x), which is a separating hyperplane that separates the DRY road surface feature vector Y _{Dj from} the road surface feature vector Y _nDj other than the DRY road surface, By providing the, the DRY road surface and the road surface other than the DRY road surface can be accurately distinguished.
The margin refers to the distance from the separation hyperplane to the closest sample (support vector), and the separation hyperplane that is the discrimination boundary has f(x)=0. The DRY road surface characteristic vector Y _Dj is in the area of f(x)≧+1, and the road surface characteristic vector Y _nDj other than the DRY road surface is in the area of f(x)≦−1.
The DRY road surface model that distinguishes the DRY road surface from the other road surfaces has a support vector Y _DSV at a distance of f(x)=+1 and a support vector Y _nDSV at a distance of f(x)=−1. It is a provided input space. Wherein A Y _DSV and the _Y nDSV, generally there exist a plurality.
The same applies to the WET model that distinguishes the WET road surface from other road surfaces, the SNOW model that distinguishes the SNOW road surface from other road surfaces, and the ICE model that distinguishes the ICE road surface from the other road surfaces.

ローカルカーネルκ_ij（ｘ_i，ｘ_j）は、時間間隔Ｔの窓毎に求められる。
なお、図１７は、時間窓の数が６であるＤＲＹ路面特徴ベクトルＹ_Djと、時間窓の数が４であるＤＲＹ路面以外の路面特徴ベクトルＹ_nDjとのＧＡカーネルを求めた例である。 In this example, a global alignment kernel function (GA kernel) is used as the kernel function K(x _α , x _β ). The GA kernel K(x _α , x _β ) is, as shown in FIG. 7 and the following equations (5) and (6), a DRY road surface feature vector x _i =Y _Di and a road surface feature vector other than the DRY road surface x _j = It is a function consisting of the sum or product of local kernels κ _ij (x _i , x _j ) indicating the degree of similarity with Y _nDj, and it is possible to directly compare time series waveforms having different time lengths.

The local kernel κ _ij (x _i , x _j ) is obtained for each window of the time interval T.
Note that FIG. 17 is an example in which GA kernels of the DRY road surface feature vector Y _Dj having six time windows and the road surface feature vector Y _nDj other than the DRY road surface having four time windows are obtained.

In order to distinguish the DRY road surface and the road surface other than the DRY road surface from the discriminant function f(x), which is a separating hyperplane that separates the DRY road surface feature vector Y _{Dj from} the road surface feature vector Y _nDj other than the DRY road surface, By providing the, the DRY road surface and the road surface other than the DRY road surface can be accurately distinguished.
The margin refers to the distance from the separation hyperplane to the closest sample (support vector), and the separation hyperplane that is the discrimination boundary has f(x)=0. The DRY road surface characteristic vector Y _Dj is in the area of f(x)≧+1, and the road surface characteristic vector Y _nDj other than the DRY road surface is in the area of f(x)≦−1.
The DRY road surface model that distinguishes the DRY road surface from the other road surfaces has a support vector Y _DSV at a distance of f(x)=+1 and a support vector Y _nDSV at a distance of f(x)=−1. It is a provided input space. Wherein A Y _DSV and the _Y nDSV, generally there exist a plurality.
The same applies to the WET model that distinguishes the WET road surface from other road surfaces, the SNOW model that distinguishes the SNOW road surface from other road surfaces, and the ICE model that distinguishes the ICE road surface from the other road surfaces.

The kernel function calculation means 42 is the feature vector X _i (P) calculated by the feature vector calculation means 32 and corrected by the feature vector correction means 33, and the DRY model, WET model, SNOW model recorded in the storage means 41. , And each of the support vectors Y _ASV and Y _nASV (A=D, W, S, I) of the ICE model, the GA kernels K _D (X, Y), K _W (X, Y) and K _S (X , Y) and K _I (X, Y) are calculated.
As shown also in FIG. 17, the GA kernel K(X,Y) is obtained when [mathematical formula 3] above, x _i is a feature vector X _i, and x _j is a road surface feature vector Y _Aj , Y _nAj. It is a function composed of the sum or product of the local kernels κ _ij (X _i , Y _j ) and can directly compare time series waveforms having different time lengths. In the figure, x _j is an example of the road surface feature vector Y _Aj , the number of time windows of the feature vector X _i is n=5, and the number of time windows of the road surface feature vector Y _Aj is m=4.
As in this example, the number n of time-series waveforms of the time window when the feature vector X _i is obtained and the number m of time-series waveforms of the time window when the road surface feature vector Y _Aj (or Y _nAj ) is obtained Even if the values are different, the similarity between the feature vectors X _i and Y _Aj (or between X _i and Y _nAj ) can be obtained.

ｆ_DはＤＲＹ路面とその他の路面とを識別する識別関数、ｆ_WはＷＥＴ路面とその他の路面とを識別する識別関数、ｆ_SはＳＮＯＷ路面とその他の路面とを識別する識別関数、ｆ_IはＩＣＥ路面とその他の路面とを識別する識別関数である。
また、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、Ｎ_SSVはＳＮＯＷモデルのサポートベクトルの数、Ｎ_ISVはＩＣＥモデルのサポートベクトルの数である。
本例では、識別関数ｆ_D，ｆ_W，ｆ_S，ｆ_Iをそれぞれ計算し、計算された識別関数ｆ_Aの最も大きな値を示す識別関数から路面状態を判別する。 The road surface state determination means 43 determines the road surface state based on the values of the four identification functions f _A (x) using the kernel function K _A (X, Y) shown in the following equations (7) to (10). (A=D, W, S, I).

f _D is a discriminant function for discriminating between the DRY road surface and other road surfaces, f _W is a discriminant function for discriminating between the WET road surface and other road surfaces, and f _S is a discriminant function for discriminating between the SNOW road surface and other road surfaces, f _I Is a discriminant function for discriminating the ICE road surface from other road surfaces.
N _DSV is the number of support vectors of the DRY model, N _WSV is the number of support vectors of the WET model, N _SSV is the number of support vectors of the SNOW model, and N _ISV is the number of support vectors of the ICE model.
In this example, the discriminant functions f _D , f _W , f _S , and f _I are calculated, and the road surface state is discriminated from the discriminant function showing the largest value of the calculated discriminant function f _A.

Next, a road surface condition determination method according to the third embodiment will be described with reference to the flowchart in FIG.
First, the acceleration sensor 11 detects the tire circumferential vibration of the running tire 1, and the internal pressure sensor 12 measures the tire internal pressure (step S30).
Next, it is determined from the measured tire internal pressure data whether or not the measured tire internal pressure P is within a preset determinable internal pressure range [P _min , P _Max ] (step S31).
If the tire internal pressure P is within the determinable internal pressure range [P _min , P _Max ], the process proceeds to step S32 to detect a time-series waveform in which tire circumferential vibration output from the acceleration sensor 11 is arranged in time series. After that, the time-series waveform that is the tire vibration data is windowed with a preset time window, and the time-series waveform of the tire vibration is extracted for each time window (step S33). Let m be the number of time-series waveforms of tire vibration for each time window.
On the other hand, when the measured tire internal pressure P is P<P _min or P>P _Max , the extraction of the time series waveform of tire vibration is stopped. In addition, after a predetermined time has elapsed, it may be determined again whether or not the measured tire internal pressure P is within the determinable internal pressure range [P _min , P _Max ].

In step S34, the feature vector X _i =(x _i1 , xi ₂ , x _i3 , x _i4 , x _i5 , x _i6 ) is calculated for each of the extracted time series waveforms of the respective time windows.
In the present invention, the power values x _{1i to} x _6i of the filter filtered wave of the time series waveform of the tire vibration, which is the component of the calculated feature vector X _t , are corrected using the data of the tire internal pressure P (step S35). .. The correction is performed for each power value x _ki (k=1 to 6).
Next, after calculating the local kernel κ _ij (X _i , Y _j ) from the corrected feature vector X _i (P) and the support vector Y _{Ak of the} road surface model recorded in the storage unit 41, the local kernel κ _ij (X _i , Y _j ) is calculated. The global alignment kernel functions K _D (X,Y), K _W (X,Y), K _S (X,Y), and K _I (X,Y) are obtained by calculating the sum of the kernels κ _ij (X _i ,Y _j ). ) Are calculated (step S36).
Next, four discriminant functions f _D (x), f _W (x), f _S (x), and f _I (x) using the kernel function K _A (X, Y) are calculated (step S37). Then, the calculated discriminant function f _A (x) is compared, and the road surface state of the discriminant function showing the largest value is discriminated as the road surface state of the road surface on which the tire 1 is running (step S38).

In the third embodiment, the discrimination parameter to be corrected according to the tire condition is the power value x of the filtered wave of the time series waveform of the tire vibration which is the component of the feature vector X _t of the time series waveform extracted in the time window. _{Although kt} is used, the output values of the discrimination functions f _D , f _W , f _S , and f _I may be used as the discrimination parameter.
The case where the discriminant parameter to be corrected according to the tire condition is the magnitude of the likelihood Z corresponds to the first discriminating step for discriminating the road surface condition from the discriminant parameter and the condition information of the present invention.
Alternatively, the belonging class z, the local kernel κ _ij (X _i , Y _j ) or the local kernel κ _ij (X _i , Y _j ) which is a parameter for obtaining the weight w of the discriminant function f(x) is calculated. It may be a constant σ for
Further, the kernel function K may be changed according to the tire internal pressure P, for example, a dynamic time warping kernel function (DTW kernel) may be used as the kernel function K.
Alternatively, it may be a parameter necessary for the learning process of the support vector machine.
Further, instead of the vibration information, the method of extracting the vibration information may be changed depending on the tire condition.
That is, also in the third embodiment, as in the second embodiment, a proper value of the bandpass filter BP(k) frequency region when calculating the power value x _kt is f _ka −f _kb , or the tire circumferential direction. Even if the time width (time window width) of the vibration time-series waveform is changed according to the tire information, the road surface condition determination accuracy can be improved.
Further, by simultaneously correcting and changing a plurality of determination parameters, it is possible to further improve the determination accuracy of the road surface condition.

Further, in the second and third embodiments, the tire condition is the tire internal pressure. However, similar to the first embodiment, the tire internal temperature may be used, or both the tire internal pressure and the tire internal temperature may be used. Alternatively, the tire outer surface temperature may be added to either or both of the tire inner pressure and the tire inner temperature.

1 tire, 2 inner liner, 3 tread, 4 wheel rim,
5 tire air chamber,
10 road surface condition determination device, 11 acceleration sensor, 12 internal pressure sensor,
13 tire condition determining means, 14 vibration waveform detecting means, 15 area signal extracting means,
16 band value calculating means, 17 band value correcting means, 18 road surface state determining means.

Claims (2)

A road surface state determination method for determining a road surface state from a time-varying waveform of vibration of a running tire detected by a vibration detection means,
Obtaining a vibration waveform of the tire,
Acquiring the tire state information,
After changing Te discrimination parameters for determining a road surface condition obtained from the vibration waveform to the state information, and an absence step to determine the road surface condition from the modified determining parameter,
The state information is one of the internal pressure of the tire, the tire internal temperature, and the tire external surface temperature, or a plurality,
The road surface condition determination method, wherein the determination parameter is a time width of a window applied to the vibration waveform . The road surface state determination method according to claim 1, wherein when the state information exceeds a preset range, the road surface state determination is not performed.

Images