JP5404062B2 - Tire pressure drop detection device and method, and tire pressure drop detection program - Google Patents

Description

  The present invention relates to a tire pressure drop detection apparatus and method for detecting a tire pressure drop based on a resonance frequency of a tire of a running vehicle, and a tire pressure drop detection program.

  One of the factors that enable a car to travel safely is the tire air pressure. If the air pressure falls below an appropriate value, the steering stability and fuel consumption are deteriorated, which may cause tire bursts. For this reason, a tire pressure alarm system (TPMS) that detects a decrease in tire air pressure and issues an alarm to the driver to prompt appropriate measures is used to protect the environment and ensure the safety of the driver. It is an important technology.

  Conventional alarm devices can be classified into two types, a direct detection type and an indirect detection type. The direct detection type directly measures tire air pressure by incorporating a pressure sensor inside the tire wheel. While it is possible to detect a decrease in air pressure with high accuracy, there are technical and cost issues such as the need for a dedicated wheel and the problem of fault-tolerant performance in the actual environment.

  On the other hand, the indirect detection type is a method of estimating the air pressure from tire rotation information, and can be subdivided into a dynamic load radius (DLR) method and a resonance frequency (Resonance Frequency Mechanism; RFM) method. The DLR method uses a phenomenon in which the dynamic load radius becomes smaller due to the crushed tire being crushed during driving, and as a result, it rotates faster than a normal pressure tire, and the pressure drop is reduced by comparing the rotation speeds of the four tires. This is a detection method. Since calculation processing can be performed relatively easily using only the wheel rotation speed signal obtained from the wheel speed sensor, research has been extensively conducted mainly for the purpose of detecting puncture of one wheel. However, since the rotational speeds of the wheels are merely compared, it is impossible to detect the case where the four wheels are depressurized simultaneously (natural leakage). In addition, since wheel speed differences also occur depending on traveling conditions such as turning of the vehicle, acceleration / deceleration, and load bias, there is a problem that pressure reduction cannot be detected accurately through all traveling states.

  On the other hand, the RFM method is a method for detecting a difference from the normal pressure by utilizing the change in the frequency characteristics of the wheel speed signal due to the reduced pressure. Unlike the DLR method, since it is an absolute comparison with the normal value of each wheel that has been held in advance, it can be used for simultaneous depressurization of four wheels, and is attracting attention as a better indirect detection method. However, depending on the driving conditions, there is a problem that the estimated value of the frequency in the target region is not robust to the vehicle speed and the road surface condition due to strong noise or the like. The present invention relates to a tire state detection device based on an RFM system. Hereinafter, the basic principle of this method will be described in more detail.

  When the vehicle travels, the torsional motion in the front-rear direction that appears when the tire receives a force from the road surface and the motion in the front-rear direction of the suspension cause a coupled resonance. Since this resonance phenomenon also affects the rotational movement of the wheel, information on the resonance phenomenon is also present in the wheel speed signal acquired from the wheel sensor mounted on the anti-lock braking system (ABS). included. In addition, since coupled resonance is an inherent vibration mode due to the torsional rigidity of the tire, its excited state changes only depending on changes in the air pressure that constitutes the physical characteristics of the tire, and changes in vehicle speed and road surface There is almost no dependence. That is, since the dynamics of the torsional motion of the tire changes as the air pressure decreases, when the wheel speed signal is analyzed by frequency, the peak created by the coupled resonance (resonance peak) appears on the lower frequency side during normal pressure than during normal pressure.

  FIG. 4 shows that tires with normal air pressure and tires with a pressure reduced by 25% from normal pressure are mounted on a vehicle, and each wheel acceleration signal (time difference between wheel speed signals) obtained during a certain time is calculated. It represents a power spectrum obtained by applying a Fast Fourier Transform (FFT) to (obtained).

  The component in the vicinity of 40 to 50 Hz is the vibration that occurs when the vibration in the longitudinal direction of the tire and the suspension of the vehicle resonate, and the frequency having the peak value (resonance frequency) moves to the lower frequency side due to the change in internal pressure. I understand that. Because this phenomenon appears independently from the above-mentioned characteristics, such as the type of tire or vehicle, traveling speed, road surface condition, etc., the RFM method focuses on this resonance frequency and is relatively in comparison to the reference frequency estimated at initialization. An alarm is issued when it is judged to be low. Here, it is necessary to estimate the resonance frequency from the wheel speed signal acquired from the ABS, but it is difficult to store the necessary time series data in an in-vehicle computer with limited calculation resources, so frequency analysis by FFT Is difficult to do. For this reason, in the conventional method, the resonance frequency is estimated using an online method described below.

Since the wheel speed signal is acquired as time-series data at each time, this is subjected to time-series analysis based on a Kth-order autoregressive (AR) model. Specifically, parameters θ = {a ₁ ,..., A _K } in the model represented by the following equation (1) are estimated by a Kalman filter (sequential least square method).

  Here, y (t) represents the wheel speed at time t, ε represents white noise, and K represents the order of the model (when a secondary model is assumed to express a phenomenon such as vibration, K = 2). Since the frequency corresponding to the pole of the transfer function representing the AR model is estimated as the resonance frequency, if the resonance peak can be correctly extracted by the model, the resonance frequency can be obtained accurately.

  However, such estimation methods that have been proposed in the past have the following problems. First, when traveling on a road with a smooth road surface, the force applied to the tire is weak and the resonance phenomenon is reduced, so that the influence of noise mixed independently of the target signal becomes relatively large. That is, since the signal noise (SN) ratio is deteriorated, it is difficult to accurately estimate the resonance frequency. Second, when traveling at a high speed (80 km / h or more), the signal of the high frequency component becomes strong due to an increase in the vertical vibration of the tire and the resonance peak tends to be unclear. Third, when a strong noise peak is generated in the frequency region in the vicinity where the resonance peak appears, the estimated value of the resonance frequency becomes unstable due to the influence. FIG. 5 shows the result of applying FFT to the traveling data obtained under such circumstances. Depending on the driving conditions, these problems may be superimposed, and in such a case, it is particularly difficult to stably estimate the resonance frequency, and as a result, it is difficult to make a pressure reduction determination.

  Even if the resonance frequency can be estimated accurately, there is the following problem regarding the determination of the reduced pressure. Depending on the type of tire or vehicle, the difference in resonance frequency between normal pressure and reduced pressure (hereinafter referred to as “decompression sensitivity”. However, the term “resonance frequency” in this specification refers to true coupled resonance. When it means a frequency estimated by a system such as the above AR model instead of a frequency, it means a difference in the average value of the distribution). In this case, if the pressure reduction sensitivity is smaller than the estimated variance of the resonance frequency, it is difficult to accurately determine the pressure reduction. That is, as a decompression judgment that is often used conventionally, there is a method of issuing an alarm when the difference between the reference frequency estimated at the time of initialization and the resonance frequency estimated at the current time is below a preset difference amount. However, for example, if the standard deviation of the distribution of the resonance frequency estimated at normal pressure is 1 Hz for a tire whose decompression sensitivity is 3 Hz, no matter how the difference is set, When the reference frequency is set lower due to variations, there is a possibility that an alarm has not been issued, and when it is set higher, there is a possibility of false alarm.

  The above-described two main problems (the problem that the resonance frequency estimation itself is difficult and the problem that the abnormality determination is difficult) make it difficult to put the abnormality detection system using the RFM method into practical use. A method for essentially solving this problem has not yet been developed.

  An object of the present invention is to solve these two problems and to provide a tire pressure drop detecting apparatus and method and a tire pressure drop detecting program capable of detecting a tire pressure drop with high accuracy. Specifically, these problems are solved by using a technique called a support vector machine (SVM).

  Here, SVM is a system that performs binary classification using feature quantities as input, such as handwritten character recognition and gene discrimination, and many studies have been made in the field of information science called machine learning. The literature has been published.

  For example, the original work “The Nature of Statistical Learning Theory”, 1995, by SVM inventor V. Vapnik. Many literatures in Japanese have been published. For example, Eisaku Maeda, “Exciting! Support Vector Machine-Old and New Pattern Recognition Method”, Information Processing Society of Japan, 2001, Vol. 42, no. 7, pp. 676-683, Koji Tsuda, “What is Support Vector Machine”, Journal of the Institute of Electronics, Information and Communication Engineers, 2001, Vol. 83, no. 6, pp. 460-466.

  However, in situations where it is difficult to detect abnormalities in tire air pressure, performance guarantee is not obvious for applying SVM to improve the detection accuracy of TPMS, and no proposals or implementations have been made yet.

A tire pressure drop detecting device according to a first aspect of the present invention (hereinafter also simply referred to as “detecting device”) includes a rotational speed information detecting means for periodically detecting rotational speed information of tires of each wheel of a vehicle,
From the rotational speed information obtained by the rotational speed information detecting means, a frequency characteristic estimating means for estimating the frequency characteristics of the rotational speed information,
Determining means for determining a decrease in the tire air pressure based on the estimated frequency characteristics,
The frequency characteristic estimation means includes parameter estimation means for estimating a linear model parameter for a time-series signal including the rotation speed information, and
The determination means uses a support vector machine that receives a linear model parameter estimated by the parameter estimation means as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs pressure reduction determination for tires. It is characterized by including determination means.

In the detection apparatus according to the first aspect of the present invention, an SVM having high performance is used as a binary classifier. The method follows three steps: First, wheel speed signals are acquired under various driving conditions by driving experiments using an actual vehicle, and K parameters for any time of all the acquired data are calculated on the desk based on the parameter estimation means. . Next, the SVM is trained using the calculated parameters and the air pressure information of the original data as training data. Finally, a trained SVM is mounted on the detection device, and a pressure reduction determination is performed using parameters calculated at each time as input. That is, the SVM is trained in advance on the desk using existing data, and the SVM is used only for discrimination between normal pressure and reduced pressure in an actual vehicle environment.
By using the SVM, it is possible to easily perform the pressure reduction determination in consideration of the entire frequency characteristics including the resonance frequency, and the determination performance in the above-described difficult detection situation is more than using only the one-dimensional index of the resonance frequency. Especially improved. As a result, a decrease in tire air pressure can be detected with higher accuracy than in the conventional method.

A detection device according to a second aspect of the present invention includes a rotation speed information detection unit that periodically detects rotation speed information of a tire of each wheel of a vehicle,
From the rotational speed information obtained by this rotational speed information detecting means, rotational acceleration information calculating means for calculating tire rotational acceleration information;
From this rotational acceleration information, frequency characteristic estimation means for estimating the frequency characteristic of the rotational acceleration information,
Determining means for determining a decrease in the tire air pressure based on the estimated frequency characteristics,
The frequency characteristic estimation unit includes a parameter estimation unit that estimates a parameter of a linear model with respect to a time-series signal including the rotational acceleration information, and
The determination means uses a support vector machine that receives a linear model parameter estimated by the parameter estimation means as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs pressure reduction determination for tires. It is characterized by including determination means.

  Also in the detection device according to the second aspect of the present invention, the pressure reduction determination can be performed with high accuracy as in the detection device according to the first aspect. Moreover, in the detection apparatus which concerns on a 2nd viewpoint, although wheel acceleration data are used, since this wheel acceleration data changes less than wheel speed data, it can raise calculation precision.

  Since the parameter can be estimated by the least square method, an optimal solution for given data can be obtained.

A tire air pressure drop detection method (hereinafter also simply referred to as “detection method”) according to a third aspect of the present invention includes a detection step of periodically detecting tire rotation speed information of each wheel of a vehicle,
From the rotation speed information obtained in this detection step, an estimation process for estimating the frequency characteristics of the rotation speed information;
A determination step of determining a decrease in air pressure of the tire based on the estimated frequency characteristic,
The estimating step is configured to estimate a parameter of a linear model with respect to a time-series signal including the rotation speed information; and
The determination step uses a support vector machine that receives the linear model parameter estimated in the estimation step as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs a decompression determination for tires. It is characterized by including a process.

In the detection method according to the third aspect of the present invention, an SVM having high performance is used as a binary classifier. The method follows three steps: First, a wheel speed signal is acquired under various driving conditions by a driving experiment using an actual vehicle, and parameters for an arbitrary time of all acquired data are calculated on the desk based on the parameter estimation means. Next, the SVM is trained using the calculated parameters and the air pressure information of the original data as training data. Finally, a trained SVM is mounted on the detection device, and a pressure reduction determination is performed using parameters calculated at each time as input. That is, the SVM is trained in advance on the desk using existing data, and the SVM is used only for discrimination between normal pressure and reduced pressure in an actual vehicle environment.
By using the SVM, it is possible to easily perform the pressure reduction determination in consideration of the entire frequency characteristics including the resonance frequency, and the determination performance in the above-described difficult detection situation is more than using only the one-dimensional index of the resonance frequency. Especially improved. As a result, a decrease in tire air pressure can be detected with higher accuracy than in the conventional method.

A detection method according to a fourth aspect of the present invention includes a detection step of periodically detecting rotation speed information of a tire of each wheel of a vehicle;
From the rotational speed information obtained in this detection step, a calculation step of calculating the rotational acceleration information of the tire,
From the rotational acceleration information obtained in this calculation step, an estimation step for estimating the frequency characteristics of the rotational acceleration information,
A determination step of determining a decrease in the tire air pressure based on the estimated frequency characteristics, and
The estimation step is configured to estimate a parameter of a linear model for a time series signal including the rotational acceleration information, and
The determination step uses a support vector machine that receives the linear model parameter estimated in the estimation step as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs a decompression determination for tires. It is characterized by including a process.

Also in the detection method according to the fourth aspect of the present invention, the pressure reduction determination can be performed with high accuracy as in the detection method according to the third aspect . In the detection method according to the fourth aspect , the wheel acceleration data is used. However, since the wheel acceleration data is less changed than the wheel speed data, the calculation accuracy can be improved.

  Since the parameter can be estimated by the least square method, an optimal solution for given data can be obtained.

According to a fifth aspect of the present invention, there is provided a program for detecting a decrease in tire air pressure, a computer for detecting a decrease in tire pressure based on a resonance frequency of a tire in a running vehicle, and a rotation of a tire in each wheel of the vehicle. From the rotational speed information obtained by the rotational speed information detecting means for periodically detecting the speed information, the frequency characteristic estimating means for estimating the frequency characteristics of the rotational speed information, and the air pressure of the tire based on the estimated frequency characteristics Function as a determination means for determining a decrease in
The frequency characteristic estimation means includes parameter estimation means for estimating a linear model parameter for a time-series signal including the rotation speed information, and
The determination means uses a support vector machine that receives a linear model parameter estimated by the parameter estimation means as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs pressure reduction determination for tires. It is characterized by including determination means.

According to a sixth aspect of the present invention, there is provided a program for detecting a decrease in tire air pressure, a computer for detecting a decrease in tire pressure based on a resonance frequency of a tire in a running vehicle, and a rotation of a tire in each wheel of the vehicle. Rotational acceleration information calculating means for calculating tire rotational acceleration information from rotational speed information obtained by rotational speed information detecting means for periodically detecting speed information, and estimating frequency characteristics of the rotational acceleration information from the rotational acceleration information Functioning as frequency characteristic estimating means for determining, and determining means for determining a decrease in air pressure of the tire based on the estimated frequency characteristics;
The frequency characteristic estimation means includes parameter estimation means for estimating a linear model parameter for a time-series signal including the rotation speed information, and
The determination means uses a support vector machine that receives a linear model parameter estimated by the parameter estimation means as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs pressure reduction determination for tires. It is characterized by including determination means.

  According to the tire pressure drop detection device and method and the tire pressure drop detection program of the present invention, even if the estimated value of the resonance frequency becomes unstable or abnormality determination becomes difficult, the tire pressure drop can be accurately detected. Can be detected.

It is a block diagram which shows one Embodiment of the detection apparatus of this invention. It is a block diagram which shows the electrical structure of the detection apparatus shown by FIG. It is a figure which shows the detection accuracy of Example 1 and Comparative Example 1. It is a figure showing distribution of power spectral density when FFT is applied to a 2-minute wheel acceleration signal. It is a figure showing distribution of the power spectrum density when FFT is applied to the wheel acceleration signal for 2 minutes which carried a high-speed run on the road where a run flat tire was attached, and the road surface is smooth. It is a figure which shows the detection accuracy of Example 2 and Comparative Example 2.

Hereinafter, embodiments of a detection apparatus and method and a tire pressure drop detection program according to the present invention will be described in detail with reference to the accompanying drawings.
As shown in FIG. 1, the detection device according to an embodiment of the present invention includes a left front wheel (FL), a right front wheel (FR), and a left rear wheel (RL) of four tires provided in a four-wheel vehicle. In order to detect the rotational speed information of the right rear wheel (RR), a normal wheel speed detecting means (rotational speed information detecting means) 1 provided in association with each tire is provided.

  The wheel speed detection means 1 generates power using rotation like a wheel speed sensor or dynamo for generating a rotation pulse using an electromagnetic pickup or the like and measuring the rotation angular speed and wheel speed from the number of pulses. It is possible to use an angular velocity sensor including that for measuring the rotational angular velocity and the wheel speed from this voltage. The output of the wheel speed detecting means 1 is given to a control unit 2 which is a computer such as ABS. The control unit 2 includes, for example, a liquid crystal display element for displaying that the tire is depressurized, a display 3 composed of a plasma display element or a CRT, and an initialization button 4 that can be operated by the driver. An alarm device 5 is connected to inform the driver of tire decompression.

  As shown in FIG. 2, the control unit 2 stores an I / O interface 2a necessary for passing signals to and from an external device, a CPU 2b that functions as a center of arithmetic processing, and a control operation program for the CPU 2b. The ROM 2c and the RAM 2d from which data is temporarily written or the written data is read when the CPU 2b performs a control operation.

  The wheel speed detection means 1 outputs a pulse signal corresponding to the number of rotations of the tire (hereinafter also referred to as “wheel speed pulse”). Then, by sampling this wheel speed pulse every predetermined sampling period ΔT (seconds), for example, ΔT = 0.005 seconds, time-series data of the wheel speed signal can be obtained. Here, it is necessary to set the sampling period to a period sufficient to observe the band in which the resonance frequency of the tire of interest appears.

  The detection apparatus according to the present embodiment includes wheel speed detection means (rotation speed information detection means) 1, frequency characteristic estimation means for estimating frequency characteristics from rotation speed information obtained by the wheel speed detection means, and estimated frequency. It is mainly comprised by the determination means which determines the fall of the air pressure of the said tire based on the characteristic. The tire pressure drop detection program causes the control unit 2 to function as a frequency characteristic estimation unit and a determination unit. The frequency characteristic estimation means includes parameter estimation means for estimating a linear model parameter for a time-series signal including the rotation speed information. In addition, the determination unit receives an input of the linear model parameter estimated by the parameter estimation unit or the resonance frequency estimated by the resonance frequency estimation unit, and outputs a binary determination value indicating normal pressure or reduced pressure. By using this, a decompression determination means for determining the decompression of the tire is included.

The feature of the present invention is to solve the problem that the estimated value of the resonance frequency is unstable due to the influence of noise generated by the running condition of the vehicle and the problem that the method of abnormality determination is difficult by using the above-described SVM. There is to do.
In the conventional TPMS based on the resonance frequency method, only the frequency indicated by the resonance peak is used as an index for detecting an abnormality in tire air pressure. Regarding this, (1) it has been clarified from the physical characteristics of tires and previous measurement experiments that the resonance frequency depends only on changes in tire pressure and not on external factors such as speed. (2) Since the phenomenon that the resonance frequency is lowered is simple, if the characteristics of the surrounding frequency region including the resonance frequency are known, it is easy to implement the criterion. These are the main reasons.

  On the other hand, in view of the properties of the tire, there is a possibility that there may be a feature that can be an index for determining the pressure reduction in addition to the resonance frequency. For example, frequency characteristics derived from the longitudinal spring and circumferential resonance of the tire appear around 15 to 20 Hz and around 80 to 90 Hz, respectively, but it is considered that these dynamics also change when the tire air pressure decreases. The characteristics should also change as a whole.

  However, such frequency characteristics other than the resonance frequency are complicated to change as the tire air pressure decreases, and are difficult to handle as an index. In addition, the change in physical characteristics associated with a decrease in tire air pressure such as the resonance frequency is not obvious, and the possibility of being affected by external factors such as speed cannot be excluded.

  Therefore, in the present invention, in order to solve the former (the problem that it is difficult to estimate the resonance frequency) among the two problems described above, the SVM having a high performance as a binary classifier is used, thereby the resonance frequency. It is easy to make a decompression determination in consideration of the overall frequency characteristics including That is, even if features other than the resonance frequency change in a complicated manner, the trained binary classifier appropriately performs the determination, so that the determination performance in the difficult-to-detect situation described above is particularly improved. As a result, the pressure reduction determination can be performed with high accuracy.

In the present invention, first, a parameter θ = {a ₁ ,..., A _K } obtained by K-dimensional time series analysis is used as an input value, and a determination value indicating normal pressure or reduced pressure (for example, normal pressure is 1, An SVM having an output value of a binary label that assigns a reduced pressure to 0 is prepared. Here, the reason why the linear parameter of the time series model is used as the K-dimensional input to the classifier is that the estimated parameter includes all the frequency characteristic information.

  Next, SVM is learned by using, as training data, a pair of parameters estimated from data collected under various driving conditions and normal pressure or reduced pressure label corresponding thereto. In other words, parameters are estimated on the desk based on experimental data (tire rotational speed information data and rotational acceleration information data) acquired in a situation where the air pressure state is known in advance, and the estimated parameters and the air pressure state are indicated. SVM is trained using a pair of binary labels as learning data. Note that the experimental data used for learning the SVM is not collected under similar driving conditions, and it is desirable to include a plurality of extreme conditions as much as possible in order to increase the determination accuracy. After that, in the actual vehicle, using the SVM learned on the desk in advance by the above-described method, the decompression determination is performed from the parameters of the time series model estimated online as in the conventional method.

  Also, the latter (the problem that abnormality determination is difficult) of the two problems described above can be solved by using SVM in the same manner. That is, although the resonance frequency itself can be estimated appropriately, the problem that subsequent determination is difficult based on the above-described conventional method can be solved because the pressure reduction sensitivity is small.

  First, as conventionally proposed, the resonance frequency is estimated based on the aforementioned AR model. An SVM is prepared in which the estimated resonance frequency is used as an input value, and a determination value indicating normal pressure or reduced pressure (for example, a binary label in which normal pressure is assigned 1 and reduced pressure is assigned 0) is output. Next, SVM is learned by using, as training data, a pair of a resonance frequency estimated from data collected under various driving conditions and a normal pressure or a reduced pressure label corresponding thereto. In other words, the resonance frequency is estimated on the desk based on experimental data (tire rotation speed information data and rotation acceleration information data) acquired in a situation where the air pressure state is known in advance, and the estimated resonance frequency and air pressure state SVM is trained by using a pair of binary labels indicating as learning data. After that, in the actual vehicle, using the SVM learned on the desk in advance by the above-described method, the decompression determination is performed from the resonance frequency estimated online as in the conventional method.

  In the present invention, it is possible to acquire a decompression determination index suitable for each traveling condition by acquiring data of an assumed traveling condition and using this data to train an SVM in advance on a desk. it can. As for the invention that solves the former of the two problems described above, even if the assumed driving conditions are complicated and diverse, the mapping from the AR parameter space to the decompression determination becomes nonlinear even if the model is nonlinear. Non-linearity can be overcome by taking measures such as using a kernel trick with a non-linear kernel. Such complex criteria cannot be implemented by human hands, but SVM automatically learns the appropriate correspondence (the relationship between the estimated parameter given as data and the air pressure state), and is accurate according to each condition. Judgment can be made.

  For the invention that solves the latter of the two problems described above, an optimal discriminant plane that maximizes generalization performance can be obtained by pre-training SVM on the desk in the same way as the former invention. Therefore, the pressure reduction determination can be reliably performed even for a tire having a low pressure reduction sensitivity.

In addition, it is more robust by adding not only the parameters obtained by time series analysis as input of SVM, the information indicating the speed and road surface condition, but also the friction slip (μS) characteristics as well as the obtained resonance frequency. Judgment can be made.
From the above, even if it is difficult to estimate the resonance frequency stably and accurately, the conventional problem can be effectively solved by the pressure reduction determination using the SVM.

[Examples and Comparative Examples]
Next, examples of the detection method of the present invention will be described, but the present invention is not limited to only the examples described below.
FIG. 3 shows a case where the pressure reduction determination is performed by applying one of the two problems described above that solves the former (Example 1), and a conventional technique using only a one-dimensional index of resonance frequency. It is a figure which shows the determination precision with a method (comparative example 1). The data sampling period is 4 ms. The vertical axis in FIG. 3 represents the frequency for each determination result. Note that the tire used in the experiment tends to have an unclear resonance peak due to its physical characteristics, and it is particularly difficult to estimate the resonance frequency.

  Data sets collected under 25 driving conditions with different speeds, road surface conditions and loads were used. Cross validation was performed using 80% of the total data for SVM learning and the remaining 20% for evaluation. Note that learning data is not used in Comparative Example 1, which is a conventional method, because SVM learning is unnecessary. As shown in FIG. 3, the application of the present invention has fewer false alarms (determined to be depressurized when it is not depressurized) and less detected (determined to be depressurized although depressurized), and the variation in the validity of the determination is smaller From this, it can be seen that an abnormality in tire air pressure is reliably detected.

FIG. 6 shows a case where the pressure reduction determination is performed by applying one of the two problems described above that solves the latter problem (Example 2), and a conventional technique using only a one-dimensional index of resonance frequency. It is a figure which shows the determination precision with a method (comparative example 2). The used data, graph notation, and evaluation procedure are the same as those in the first embodiment. Note that the tire used in the experiment has a low pressure reduction sensitivity due to its physical characteristics, and therefore it is particularly difficult to determine the pressure reduction by the conventional method.
From FIG. 6, it can be seen that by applying the present invention, it is possible to reliably detect an abnormality in tire air pressure without false alarms / undetected. This is because an optimal discrimination plane for learning two air pressure conditions is learned in advance.

1 Wheel speed detection means 2 Control unit 2a Interface 2b CPU
2c ROM
2d RAM
3 Display 4 Initialization button 5 Alarm

Claims (8)

Rotational speed information detecting means for periodically detecting rotational speed information of tires of each wheel of the vehicle;
From the rotational speed information obtained by the rotational speed information detecting means, a frequency characteristic estimating means for estimating the frequency characteristics of the rotational speed information,
Determining means for determining a decrease in the tire air pressure based on the estimated frequency characteristics,
The frequency characteristic estimation means includes parameter estimation means for estimating a linear model parameter for a time-series signal including the rotation speed information, and
The determination means uses a support vector machine that receives a linear model parameter estimated by the parameter estimation means as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs pressure reduction determination for tires. A tire pressure drop detecting device including a judging means. Rotational speed information detecting means for periodically detecting rotational speed information of tires of each wheel of the vehicle;
From the rotational speed information obtained by this rotational speed information detecting means, rotational acceleration information calculating means for calculating tire rotational acceleration information;
From this rotational acceleration information, frequency characteristic estimation means for estimating the frequency characteristic of the rotational acceleration information,
Determining means for determining a decrease in the tire air pressure based on the estimated frequency characteristics,
The frequency characteristic estimation unit includes a parameter estimation unit that estimates a parameter of a linear model with respect to a time-series signal including the rotational acceleration information, and
The determination means uses a support vector machine that receives a linear model parameter estimated by the parameter estimation means as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs pressure reduction determination for tires. A tire pressure drop detecting device including a judging means. The tire pressure drop detecting device according to claim 1 or 2 , wherein the parameter estimating means is configured to estimate a parameter of a linear model by a least square method. A detection step of periodically detecting rotation speed information of tires of each wheel of the vehicle;
From the rotation speed information obtained in this detection step, an estimation process for estimating the frequency characteristics of the rotation speed information;
A determination step of determining a decrease in air pressure of the tire based on the estimated frequency characteristic,
The estimating step is configured to estimate a parameter of a linear model with respect to a time-series signal including the rotation speed information; and
The determination step uses a support vector machine that receives the linear model parameter estimated in the estimation step as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs a decompression determination for tires. A method for detecting a decrease in tire air pressure, comprising: a step. A detection step of periodically detecting rotation speed information of tires of each wheel of the vehicle;
From the rotational speed information obtained in this detection step, a calculation step of calculating the rotational acceleration information of the tire,
From the rotational acceleration information obtained in this calculation step, an estimation step for estimating the frequency characteristics of the rotational acceleration information,
A determination step of determining a decrease in the tire air pressure based on the estimated frequency characteristics, and
The estimation step is configured to estimate a parameter of a linear model for a time series signal including the rotational acceleration information, and
The determination step uses a support vector machine that receives the linear model parameter estimated in the estimation step as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs a decompression determination for tires. A method for detecting a decrease in tire air pressure, comprising: a step. The tire pressure drop detection method according to claim 4 or 5 , wherein in the step of estimating the parameter, the parameter of the linear model is estimated by a least square method. Rotation obtained by rotation speed information detecting means for periodically detecting the rotation speed information of the tires of each wheel of the vehicle based on the resonance frequency of the tires of the running vehicle based on the resonance frequency of the tires. From the speed information, function as frequency characteristic estimation means for estimating frequency characteristics of the rotational speed information, and determination means for determining a decrease in the tire air pressure based on the estimated frequency characteristics,
The frequency characteristic estimation means includes parameter estimation means for estimating a linear model parameter for a time-series signal including the rotation speed information, and
The determination means uses a support vector machine that receives a linear model parameter estimated by the parameter estimation means as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs pressure reduction determination for tires. A tire pressure drop detection program including a determination means. Rotation obtained by rotation speed information detecting means for periodically detecting the rotation speed information of the tires of each wheel of the vehicle based on the resonance frequency of the tires of the running vehicle based on the resonance frequency of the tires. Rotational acceleration information calculating means for calculating the rotational acceleration information of the tire from the speed information, frequency characteristic estimating means for estimating the frequency characteristic of the rotational acceleration information from the rotational acceleration information, and based on the estimated frequency characteristics It functions as a determination means for determining a decrease in tire air pressure,
The frequency characteristic estimation means includes parameter estimation means for estimating a linear model parameter for a time-series signal including the rotation speed information, and
The determination means uses a support vector machine that receives a linear model parameter estimated by the parameter estimation means as an input and outputs a binary determination value indicating normal pressure or reduced pressure, and performs pressure reduction determination for tires. A tire pressure drop detection program including a determination means.

Images