JPH08119131A - Road surface slip condition detection device - Google Patents

Abstract

PURPOSE: To detect slip condition of a road surface at an early time in the beginning stage of steering before a vehicle reaches a spinning state, by estimating the slip condition between wheels and road surface by the use of a sensor of general use without needing a special sensor such as a sensor for use in detecting steering reactive force. CONSTITUTION: A circuit 39 which computes every moment a residual between a fiducial condition quantity estimated and a detected condition quantity detected and accumulates time-sequential data of fiducial residuals in accordance with a difference between slip condition of a travelling road surface and that of a fiducial road surface. Circuits 42, 43 which accumulate time-sequential data of residual models in accordance with differences between slip condition of other road surfaces having pre-determined slip and that of the fiducial road surface. The slip condition of the travelling road surface is estimated by finding which of the other road surfaces having pre-determined slip is approximate to the travelling road surface by comparing a time sequence pattern of the time-sequential data of fiducial residuals with that of the time-sequential data of residual models.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a road surface slipping state detecting device, and more particularly to a vehicle state estimating amount estimated from observed values of steering angle and motion state amount (yaw rate, lateral acceleration, etc.),
The present invention relates to a road surface slipping state detecting device for detecting a road surface slipping state from a residual difference between a vehicle motion state quantity and an observed value.

[0002]

2. Description of the Related Art As a conventional technique relating to a road surface slipping state detecting device, Japanese Patent Application Laid-Open No. 1-220485, which is applied to a 2WS (two-wheel steering) and a 4WS (four-wheel steering) vehicle,
A technique described in Japanese Patent Laid-Open No. 4-230472, which is applied to a 4WS vehicle, is known. The principle of this conventional road surface slipping state detecting device is that the magnitude of the reaction force from the road surface with respect to the steering angle becomes smaller as the road surface friction coefficient decreases, and the reaction force is reduced by a dedicated load cell or the like. The slip state of the road surface is determined based on the relationship between the detected vehicle angle and the steering angle. The control system is configured to operate the vehicle on the safe side when it is determined that the road surface is slippery, that is, the road surface has a low friction coefficient.

[0003]

However, in such a conventional road surface slipping state detecting device, since the steering reaction force with respect to the steering angle has a hysteresis characteristic, it takes one cycle or more to judge the slippage of the road surface. It is necessary to use the steered reaction force information. Therefore, for example, in JP-A-1-204865, there is a possibility that a low μ road spin will occur before the control system operates even if the road surface is slippery, that is, even if it can be detected that the friction coefficient is low.

In this respect, in the invention disclosed in Japanese Patent Laid-Open No. 4-230472, the slip state of the road surface can be determined before the driver's steering by the minute steering of the rear wheels which continues at 2-3 Hz. However, this technique requires an actuator that steers the rear wheels.

Further, in these conventional road surface slipping state detecting devices, it is necessary to provide a dedicated sensor for detecting the steering reaction force.

Further, in the conventional road surface friction coefficient measuring method (Japanese Patent Laid-Open No. 3-264846), the stability factor indicating the relationship between the steering angle and the lateral acceleration during steady circle turning changes in accordance with the road surface friction coefficient. However, when traveling on a general road, the road surface friction coefficient may not be measured because it does not always make a steady circular turn.

Therefore, the present invention provides a very general sensor without the need for a special sensor such as a sensor for detecting a steering reaction force, regardless of the vehicle control system such as 2WS or 4WS. An object of the present invention is to provide a road surface slipping state detecting device capable of estimating a slipping state between a wheel and a road surface and detecting the slipping state of the road surface early in the initial stage of steering before the vehicle reaches a spin state. And

[0008]

In order to achieve the above object, a road surface slip state detecting device of the present invention is provided with a steering amount of a vehicle,
Detecting means for detecting a state quantity relating to vehicle motion including lateral acceleration, yaw rate, vehicle speed, reference vehicle motion model on a reference road surface having a predetermined slip, and vehicle motion model on another road surface having a slip different from the reference road surface Vehicle motion model storage means for storing in advance the amount of change from the reference vehicle motion model, and at least one of lateral acceleration and yaw rate as a reference based on the reference vehicle motion model and the detected state quantity detected by the detection means. A reference state quantity estimating means for estimating as a state quantity; a reference residual for calculating a residual between the reference state quantity estimated by the reference state quantity estimating means and the detected state quantity detected by the detecting means as a reference residual. Calculating means, reference residual time series data accumulating means for accumulating time series data of the reference residual calculated by the reference residual calculating means, and the vehicle Residual model calculation means for calculating a residual model value of at least one of lateral acceleration and yaw rate for another road surface based on the amount of change of the dynamic model from the reference vehicle motion model and the detected state quantity detected by the detection means. And a residual model time series data accumulating means for accumulating time series data of residual model values calculated by the residual model calculating means, and accumulated standard residual time series data and residual model time series data. And a road surface slipping state estimating means for estimating the slipping state of the road surface by comparing the time series pattern of.

[0009]

According to the present invention, the residual between the estimated reference state quantity and the detected detection state quantity is momentarily calculated as the reference residual,
Time series data of reference residuals according to the difference in slippage between the traveling road surface and the reference road surface is accumulated. Further, the residual model value is calculated according to the difference in the slip state between the other road surface having a predetermined slip and the reference road surface, and the time series data of the residual model value is accumulated. Then, by comparing the time series patterns of the reference residual model time series data and the residual model time series data, the slip state of the road surface is estimated depending on which of the other road surfaces the running road surface has a predetermined slip. To do.

When traveling on a general road surface, since the steering angle changes every moment due to the curvature of the road, etc., it does not always become a steady circular turn. However, in the present invention, the steering angle which changes every moment is a reference as a detected state quantity. Since the difference between the reference state quantity and the detected state quantity calculated based on the reference vehicle movement model and the detected state quantity is separated from the steering angle, the steering angle is There is no problem even if it changes.

According to the road surface slip detecting device of the present invention,
When the vehicle is running on a slip surface that is different from the road surface for which the vehicle motion model is set, when the steering wheel is steered, a residual error occurs in at least one of lateral acceleration and yaw rate, and the residual time series data is used as the residual model time series data. By comparing with, the slip state of the road surface can be detected at the initial steering stage regardless of whether the vehicle is a 2WS or 4WS vehicle. Therefore, it is possible to switch to the control system according to the slip state of each road surface before the spin, which contributes to safe running. Further, since it is not necessary to detect the steering reaction force, it is not necessary to provide a special sensor.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT A preferred embodiment of the present invention will now be described in detail with reference to the drawings.

FIG. 1 schematically shows a 4WS vehicle to which the present invention is applied. The 4WS vehicle of the present embodiment has a vehicle body 10 having a mass m and a yaw inertia I, a steering wheel 12 as a steering device, a front wheel tire 14 having a tire characteristic of cornering power Cf, and a rear tire having a tire characteristic of cornering power Cr. The wheel tire 16 includes a controller 18 that calculates a rear wheel control amount for the tire 16, and an actuator 62 that steers and drives the rear wheel tire 16 based on the rear wheel control amount.

The 4WS vehicle has a yaw rate sensor 20, an acceleration sensor 22, a steering angle sensor 24,
A vehicle speed sensor 26, a signal processing circuit 28, and a road surface slipping state detection device 30 are provided.

[0015] The corresponding front wheel steering angle δf is detected, and the vehicle speed sensor 26 detects the vehicle speed Vc. Then, the detection symbols of these sensors are amplified by the signal processing circuit 28.

The yaw rate generated by the steering of the 4WS vehicle and detected by the yaw rate sensor 20. You. The controller 18 controls the rear wheel tire 16 via the actuator 62 to improve the quick response. The vehicle constant values used are those measured in advance by experiments or the like. In this embodiment, constant values as shown in Table 1 described later are used.

Further, the detection signals of the respective sensors amplified in the signal processing circuit 28 are outputted to the road surface slipping state detecting device 30. Road surface slippage detection device 30
Is for detecting slippage on the road surface and for detecting a failure in the steering control system of the 4WS vehicle. Therefore,
When a sensor failure is detected, the slippage state detection of the road surface can be stopped to improve reliability. The slip state of the road surface and the failure detection result by the road surface slip state detection device 30 are input to the controller 18, and if it is detected that the actuator or the sensor is out of order, the 4WS control and the detection of the slip state of the road surface are stopped. On the other hand, if it is detected that the road surface is slippery, the 4WS control law is changed to operate the vehicle on the safe side.

In this embodiment, the road surface slipping state detecting device 3 is used.
0, failure of the acceleration sensor 22, failure of the yaw rate sensor 20, failure of the rear wheel actuator 62,
The slip condition of the road surface is detected.

In the following description, a non-slip road surface will be referred to as a reference road surface, and a slippery road surface will be referred to as a road surface different from the reference road surface or simply different road surface. Suppose there is a road surface.

FIG. 2 is a block diagram of the road surface slipping state detecting device 30, and FIG. 3 is a detailed block diagram of the road surface slipping state detecting circuit 38 in FIG.

A road surface slipping state detecting device 30 according to the present embodiment.
Includes a memory 32 storing a table of a vehicle motion model of a reference road surface, a state quantity estimating circuit 33, a vehicle slip angle estimating circuit 34, a residual calculating circuit 36, and a road surface slipping state detecting circuit 38. It is composed of.

Further, as shown in FIG. 3, the road slip state detecting circuit 38 includes a reference residual time series data storage circuit 39.
A memory 40 that stores a table of the amount of change of the vehicle motion model from the reference road surface on each different road surface, a sensor time series data storage circuit 41, a residual model time series data calculation circuit 42, and a residual model time series. Data storage circuit 43
And a road surface slipping state estimation circuit 44.

The state quantity estimating circuit 33 inputs the model and the input of the reference road surface stored in the memory 32. The slip angle estimating circuit 34 of the vehicle estimates the slip angle β of the vehicle from the detection values of the yaw rate sensor 20, the acceleration sensor 22, and the vehicle speed sensor 26 amplified by the signal processing circuit 28,
It outputs to the road surface slippage detecting device 38.

The residual calculation circuit 36 includes a yaw rate sensor 20 and an acceleration sensor 2 amplified by the signal processing circuit 28.
The residual between the detected value of 2 and the estimated value of the state quantity estimating circuit 32 is calculated and output to the road slip state detecting circuit 38.

The reference residual time series data storage circuit 39 of the road surface slipping state detection circuit 38 uses the residual difference

The sensor time series data storage circuit 41 is also provided.
Is the same fixed time as the residual time series data Accumulate data. The residual model time series data calculation circuit 42 for calculating the residual model time series data is provided with a model change amount of each road surface stored in the memory 40 from the reference road surface. It is stored in the residual model time series data storage circuit 43 as residual model time series data of the same data length sampled at the same fixed time intervals as the series data. Further, in the residual model time series data storage circuit 43, three residual models at the time of failure of the acceleration sensor 22, the yaw rate sensor 20, and the rear wheel actuator 62 are similarly stored in advance as residual model time series data. Therefore, a total of m sets (m ≧ 4) as road surface models and failure models different from the reference road surface
The residual model of will be prepared.

When the steering control system of the 4WS vehicle is normal, the detection values of the sensors 20 and 22 and the state quantity estimating circuit 33 are detected.
Since the estimated values of 1 are approximately equal, the residual output from the residual calculation circuit 36 is approximately 0. On the other hand, when some kind of failure occurs in the steering control system or when steering is started on a different road surface, the detection signals of the sensors 20 and 22 differ from the estimated value, and a residual error occurs. Therefore, when the residual is taken out from the past time point to the present time point by dividing it into a certain time (Window) and the residual error is captured as a time series pattern, a unique residual pattern is shown according to each road surface and each failure. It will be.

The road surface slipping state detection circuit 44 is generated in the steering control system from the input residual time series data and 1 to m sets of residual model time series data corresponding to different road surfaces and failures. It is configured to specify the failure point and the degree of slippage of the road surface. That is, the road surface slipping state detection circuit 38 of the present embodiment calculates the residual time series pattern of the reference residual time series data storage circuit 39 and the residual model time series pattern of the residual model time series data storage circuit 43 as The pattern matching calculation process is performed according to a predetermined routine, and the failure that occurs is the failure of the yaw rate sensor 20, the acceleration sensor 22, or the failure of the rear wheel actuator 62 according to the residual model with the highest degree of matching. It is configured to judge whether or not the road surface slips.

In this way, according to the road surface slippage detecting apparatus of this embodiment, the degree of slippage of the road surface can be detected at the initial stage of steering. Therefore, it is possible to switch to a control system according to the degree of slippage of the road surface before the spin, which contributes to safe driving. Further, the yaw rate sensor 20 and the acceleration sensor 2 generated in the steering control system
2 and a slight failure of the rear wheel actuator 62 can be detected quickly and accurately, and the controller can be operated safely. Further, since the failure of the acceleration sensor and the yaw rate sensor used for detecting the slip state of the road surface can be detected, the reliability of the slip state detection of the road surface can be improved. However, in the present embodiment, since the failure detection is also described, the 4WS vehicle is selected, but the same can be applied to the 2WS vehicle. Also, the standard state quantity to be estimated It is possible to apply the same in the same way even if estimated.

Next, the specific structure and operation of this embodiment will be described in detail. In this embodiment, , GLR (Generalized Likeliliho
od Ratio) method is used. Details of this GLR method are described in the following documents and the like.

1) Author name: A. S. Willsky Literature name: A Survey of Design Me
hows for Failure Detecti
on in Dynamic Systems, Aut
Omatica, Vol. 12, pp601-611,
1976 2) Author name: A. S. Willsky and H.M.
L. Jones Reference name: A Generalized Likelilih
od Ratio Approach to the
Detection and Estimation
of Jumps in Linear System
ms, IEEE Trans. Auto. Contr.
Vol. AC-21. NO. 2, pp108-, 19
76 The outline of the GLR method will be described with reference to FIG. GLR
The method was developed for the purpose of fault detection. First, the fault estimation target is represented by a mathematical model and a state estimation system is constructed by a Kalman filter. Then, for each failure, the target sensor output Y and Calculate as a failure model. Therefore, in the GLR method, a failure location is obtained by statistically matching the residual time series pattern actually obtained with the residual time series pattern of each failure model.
The size of the failure and the time when the failure occurred are specified.

First, a detection system for detecting a failure / slip condition on the road surface is designed. The failure / road slip condition detection system is designed in the following procedure. The vehicle speed V _c in the following specific design is V _c = 22.22 m / s (= 80 km
/ H) was used.

1) Vehicle Modeling The vehicle equation of motion is represented by a system.

[0034]

[Equation 1]

Table 1 below shows the definitions of the symbols used in the expressions (1) and (2) and the respective constants. , V _c is the vehicle speed [m / s], δf is the front wheel steering angle [ra
d] and u are rear wheel control amounts [rad].

[0036]

[Table 1]

The rear wheel control amount u is calculated by the following equation.

[0038]

[Equation 2]

Further, the slip angle β of the vehicle is β << 1 [r
lateral acceleration within the range

[0040]

(Equation 3)

[0041]

When the motion model of the vehicle is represented by the state equation using the above equations (1) to (4), the following equation is obtained.

[0043]

[Equation 4]

However,

[0045]

(Equation 5)

The subscript T in X, U and Y represents transposition. Further, q represents system noise, w represents sensor noise, and F and G are system noise vectors.
A, B, C, and D in the equations (5) and (6) are matrices as follows.

[0047]

(Equation 6)

Substituting each numerical value of Table 1 into the equation (8) and discretizing it by the zero-order hold at the sampling period of 8 ms is as follows.

[0049]

(Equation 7)

However, k represents a time step (time). Further, although the matrices A to G in the expressions (9) and (10) use the same notation as the expressions (5) and (6) expressed in the continuous system, note that the contents are different.

The A, B, C and D matrices are as follows.

[0052]

(Equation 8)

2) Vehicle state quantity estimation system design The state quantity estimation circuit 33 uses the above equations (9) and (10),
The yaw rate and lateral acceleration of the vehicle are estimated and calculated, and the Kalman filter of the following equation is used for the estimation and calculation of the motion state.

[0054]

[Equation 9]

K in the equation (13) is the Kalman filter gain, and the Kalman filter calculates the gain using the system noise variance Q and the sensor noise variance R. First, the lateral force q (q = 2) is applied to the front wheels as system noise.
000 N) is added, the following system noise vectors F and G are assumed.

[0056]

[Equation 10]

The system noise vectors F and G are (11)
When the A matrix of the equation is used and the discretization is performed by the zero-order hold with the sampling period of 8 ms, the following equation is obtained.

[0058]

[Equation 11]

The system noise variance Q is set as follows. Q = 4 × 10 ⁶ (17) On the other hand, the noise variance R of the sensor was as follows as a result of calculation based on the actual vehicle traveling data.

[0060]

(Equation 12)

The Kalman filter gain K calculated based on the above is shown below.

[0062]

(Equation 13)

The covariance V of the residual after the update is as follows.

[0064]

[Equation 14]

3) Modeling of Faults Using the vehicle models of the above equations (9) and (10), sensors,
The actuator failure is expressed as a discrete system as follows.

[0066]

(Equation 15)

However, f is an event vector representing an actuator failure, and g is an event vector representing an actuator failure and a sensor failure. The subscript i represents a failure point, and the actuator (i = 1),
Acceleration sensor (i = 2), yaw rate sensor (i = 3)
And Further, τ represents a failure occurrence time, and ν is a scalar representing the size of the failure. Both τ and ν are unknown parameters, which will become clear when calculating the GLR value described later.

When a failure occurs, the following occurs.

[0069]

[Equation 16]

[0070] In the formula, it is a value calculated as f _i ν = 0 and g _i ν = 0 (no failure). α _i (k), β _i (k), μ _i
(K) and p _i (k) are correction terms according to the failure, and can be calculated in advance by the following formula.

[0071]

[Equation 17]

From the above, the residual pattern ρ _i (k) of the lateral acceleration and the yaw rate is created from the failure occurrence time τ as a failure model. However, the length of the residual time series pattern (Windo
w) is 40 samplings.

First, the actuator failure event vectors f ₁ and g ₁ are expressed as follows.

[0074]

(Equation 18)

Substituting the vehicle specifications into the equations (31) and (32) gives the following equation (note that f ₁ is discretized with a sampling cycle of 8 ms and a zero-order hold).

[0076]

[Formula 19]

When creating the actuator failure model ρ ₁ ,
The initial value is set as follows.

[0078]

(Equation 20)

However,

[0080]

[Equation 21]

The failure vector of the acceleration sensor and the yaw rate sensor is shown in the following equation. Acceleration sensor failure: g ₂ = [10 0] ^T (39) Yaw rate sensor failure: g ₃ = [01 1] ^T (40) The initial values of the sensor failure model p _2,3 are as follows.

[0082]

[Equation 22]

However, β _2,3 is as follows.

[0084]

(Equation 23)

The actuator failure model ρ ₁ thus obtained is shown in FIG. 5, the acceleration sensor failure model ρ ₂ is shown in FIG. 6, and the yaw rate sensor failure model ρ ₃ is shown in FIG.

4) Road surface model different from the reference road surface When running on a slippery road surface, the cornering force decreases, but this decrease is an average cornering power C.
It is assumed that f and Cr decrease. This is shown in FIG. In order to confirm this assumption, a low friction coefficient road surface (μ = 0.25-0.25
0.35) The cornering powers Cf and Cr were examined by trial and error based on the experimental results during running. As a result, it was found that Cf and Cr were reduced to 0.5 times the vehicle specifications in Table 1. FIG. 9 shows the degree of agreement between the vehicle models of the expressions (9) to (11) for the high friction coefficient road surface traveling result, and FIG. 10 shows Cf and Cr for the low friction coefficient road surface traveling result. The degree of coincidence between vehicle models was similarly examined at 0.5 times (vehicle speed is 80 km / h in each case). However, only the front wheel steering angle was used as input from the experimental data, and the rear wheel control amount u was calculated by multiplying the calculated yaw rate by 0.1 (Equation (3)). From FIGS. 9 and 10, the vehicle model shows a good degree of agreement with the experimental results for both high and low friction coefficient road surface running.

Therefore, a high friction coefficient road surface (reference road surface)
Cf and Cr set in step 2 are changed by ΔC due to changes in road surface μ.
If only f and ΔCr are changed, the vehicle is represented by the following equation (however, it is represented by a continuous system).

[0088]

[Equation 24]

[0089]

(Equation 25)

Further, when the equations (46) and (47) are expressed in a discrete system, the following equation is obtained.

[0091]

(Equation 26)

Σ is a unit step function, σ = 0 when traveling on a high friction coefficient road surface (reference road surface), σ = 1 when traveling on a low friction coefficient road surface (different road surface), and τ depends on the start of steering. It represents the residual occurrence time. In addition, (4
Note that A to G and ΔA to ΔD in the equations (9) and (50) have the same notation as the equations (46) and (47), but the contents are different.

Here, the part corresponding to f _{i and} g _i as model changes as shown in equations (21) and (22) is replaced by (4
Describing the matrix changes ΔA, ΔB, ΔC, and ΔD in the equation 8), since the vector and the matrix are added,
Model change detection cannot be applied to the vector fault input type GLR. Therefore, the model change is vectorized as follows using the state quantity X and the control quantity U.

[0094]

[Equation 27]

[0095] The signal estimated by the pseudo integration of is used. That is, the slip angle β of the vehicle is estimated using the following equation.

[0096]

[Equation 28]

However, s is a Laplace operator, and T = 0.6366s (cutoff frequency f = 0.2).
5 Hz). The equation (54) is represented by a state equation and is discretized by a zero-order hold with a sampling period of 8 ms, which results in the following equation.

[0098]

[Equation 29]

Then, h in the equations (51) to (53)
Is a subscript corresponding to the degree of change in the cornering powers Cf and Cr, and is the cornering power C on the reference road surface.
f and Cr are h = 4 when they are reduced to 0.25 times, h = 5 when they are reduced to 0.5 times, and h = when they are reduced to 0.75 times.
6

ΔA _{4 to} Δ when Cf and Cr = 0.25
D ₄ is shown below.

[0101]

[Equation 30]

[0102] (52), g _h is the vector in equation (53), which vector by the following equation using the (56) - (58) below.

[0103] _{g h (k) = [ΔC} h X (k) + ΔD h U (k) 0] Therefore ^T (59), (51) - (53) equation (56) - (5
If the model change is expressed in the form of a vector by the expression (9), the vector fault input type GLR can be applied. However,
Since the equations (51) to (53) include dynamic variables such as the state quantity X and the control quantity U, a failure model ρ cannot be created in advance like a sensor / actuator failure, and the road surface is online. The residual model of, that is, the road surface model ρ will be created.

A method of creating this road surface model will be described below. First, when the cornering power changes due to traveling on a road surface different from the reference road surface, the change in the state of the vehicle and the Kalman filter and the residual error are expressed by the following equations as in the case of a failure.

[0105]

[Equation 31]

[0106] It is the value calculated by Next, α _h , β _h , μ _h , ρ _h
Is created by the following equation in the same manner as the failure model.

[0107]

[Equation 32]

That is, equations (51) to (53) and (5
6) to (59), which are vectorized from moment to moment
Dell change f_h(K, τ), g_h(K, τ), g_h(K +
Substitute (1, τ) into equations (64), (65), (67),
Online, α_h, Β_h, Μ _h, Ρ_hTo calculate.

Here, with respect to the magnitude ν of the fault, which has been parameterized in the fault model, ν = 1 since the model is created according to the change in cornering power on each road surface.
Is.

The initial values in the equations (64) to (67) are set as follows.

[0111]

[Expression 33]

As described above, the matching with the residual time series pattern is performed within the range cut out by the window. Then, the time when the residual error occurs due to the start of steering is set as a parameter τ,
In the parts relating to the model change in the equations (51) to (53), the state quantity X and the control quantity U are selected as shown in FIG. This is because the model change ΔA before the start of steering (straight ahead)
~ ΔD does not occur, and the model change component is 0 on the X and U sides.
Means to Therefore, Window = 4
In the case of 0, X of τ = k−39 to k−1 in FIG. 11,
The section where U is 0 is model change f _h (k, τ),
_{g h (k, τ),} g h (k + 1, τ) because there is a 0, (6
It is not necessary to calculate α _h , β _h , μ _h , and ρ _h in the equations 4) to (67). Further, since the tendency of the vehicle model change becomes more noticeable after the vehicle is steered to some extent, each road surface model has a steering angle δf = 0.01 rad (8.
It was decided to create it from the point when it exceeded 4 deg).

5) Method of collating residual error with failure / different road surface model: H _{0 for} no abnormality (failure / different road surface running), H _{i for} failure occurrence (however, ν and τ are unknown) or different running (However, τ is unknown)
Make a hypothesis of _h . That is, if the residual γ _N follows a Gaussian probability distribution with mean 0 and variance V,
H ₀ , H _i , and H _h are as follows.

No abnormality H ₀ : γ (k) = 0 (71) Abnormality H _i : γ (k) = ρ _i (k, τ) ν (72) H _h : γ (k) = ρ _h (k , Τ) (73) Then, γ (1), γ (2), γ (3) ..., γ
Perform a likelihood ratio test on (k). First, the likelihood ratio for a failure is expressed by the following equation using a Gaussian probability density function.

[0115]

(Equation 34)

Here,

[0117]

[Equation 35]

However, m represents the size of the residual γ. Also,

[0119]

[Equation 36]

This likelihood ratio Li is roughly the abnormality hypothesis H
It operates as follows according to _i and H ₀ .

H _i : γ (k) = ρ _i (k, τ) ν, the exp value of the equation (75) is 1, the exp value of (76) is smaller than 1, and the likelihood ratio Li is 1. Will be larger than.

H ₀ : e of the equation (76) is obtained by γ (k) = 0.
The xp value is 1, the exp value of (75) is a value smaller than 1, and the likelihood ratio Li is smaller than 1.

However, from equations (74) to (76), the magnitude ν of the failure and the time τ of failure occurrence are unknown, and therefore cannot be calculated. Therefore, first, in equation (74), (75) and (76)
The logarithmic likelihood ratio is defined as follows by substituting the expression and taking the logarithm.

Li (k, τ, ν) = 2ln (Li (k, τ, ν)) = 2ν d (k, τ, i) -ν ² a (k, τ, i) (77) where a ( k, τ, i) and d (k, τ, i) are as follows.

[0125]

(37)

Then, the magnitude ν of the fault is calculated from the equation (77) by the maximum likelihood estimation of the following equation.

[0127]

(38)

[0128]

[0129]

[Formula 39]

By substituting the equation (80) into the equation (77), the log-likelihood ratio li (k, τ, ν) is obtained by the following equation.

[0131]

(Equation 40)

Next, the likelihood ratio for different road surface running is expressed by the following equation using a Gaussian probability density function.

[0133]

[Formula 41]

Here, the denominator and the numerator term in the expression (82) are similarly described in the form where the expression ν is omitted in the expression (75) (ν = 1).

Then, similarly to the equation (77), ν is added to the equation (82).
The log-likelihood ratio obtained by substituting the equations (75) and (76) for = 1 is defined as follows.

[0136]

(Equation 42)

However, a (k, τ, h), d (k, τ, h)
Is as follows.

[0138]

[Equation 43]

However, since the failure occurrence time τ and the steering residual error occurrence time τ are unknown in the equations (81) and (83), τ is obtained by the following procedure. FIG. 12 shows the state of the failure occurrence time τ as an example. However, it is explained in FIG. In FIG. 2, the failure model ρ _i is compared with the residual γ in the next section with respect to the time k.

K−40 ≦ Window ≦ k−1 (85) That is, in FIG. 12, the failure model ρ _{i with} respect to the residual error.
It is checked whether ν is sequentially matched from j = 1 to j = 40, and when they match, τ = j. This collation is performed as follows.

First, using the failure model ρ _i calculated from j = 1 to 40 in (30), a (k, τ,
Regarding i), j = 1 to 40, j = 1 to 39, j = 1 to 3
, ..., j = 1 to 1 is calculated, and 1 ×
Make 40 vectors. Since a failure model ρ _i can be created in advance for each failure, a (k, τ, i) in equation (78) can be calculated in advance.

Next, d (k, τ, i) in the equation (78) is ρ _i
Is associated with the vector of j = 1 to 40 of γ and the vector of j = 1 to 40 of γ and the vector of j = 2 to 40 of γ with respect to the vector of j = 1 to 39 of ρ _i. Calculate and create a 1 × 40 vector.

Here, γ is the residual vector γ cut out in the range of the equation (85), and its head is the first.
That is, a (k, τ, i) and d (k, τ, i) are calculated as in the following equation (where Windows = 40).

[0144]

[Equation 44]

Then, in the search for the failure occurrence time τ, the calculated a (k, τ, i) and d (k, τ, i) are used to obtain (
81) Calculate the log-likelihood ratio li (k, τ, ν) of the equation
Create a x40 vector. The point having the largest value among the vector elements is the failure occurrence time τ.

Also, regarding different road surface running (67)
Using ρ _h of the equation, (85) to (87)
4 (a) is calculated, and a (k, τ, h) and d (k, τ, h) are calculated, and a 1 × 40 vector is created for the log-likelihood ratio lh (k, τ) of the equation (83). The point having the largest value among the vector elements is the residual generation time τ due to steering.

Therefore, as a procedure for detecting a failure / sliding state of the road surface, for each failure (actuator, acceleration sensor, yaw rate sensor), the equation (81) is represented by j = 1 to j.
Up to 40 and log likelihood ratio l in j = 1 to 40
Time τ is determined as a point at which i (k, τ, ν) and lh (k, τ) are maximized. Then, the log-likelihood ratios li (k, τ, ν) and lh (k, τ) at each time τ are set to G
Determine as LR value. Further, regarding the failure, a (k, τ, i), d (k, τ, i) at the failure occurrence time τ
The equation (80) is calculated using, and its magnitude ν is determined. Which of the faults and the degree of slip of the road surface are determined by comparing the respective GLR values calculated for the respective faults and the different road surface models and determining the one having the largest value. If it is determined that the road surface is different from the reference road surface, which road surface model (h = 4 to
The degree of road slippage is detected depending on whether the GLR value in 6) is the highest.

However, for failure detection, the time from FIG.
The failure occurrence time τ is no longer included as in the case of k = τ + 41
And the failure model ρ that matches the residual γ_iIs gone, GL
R has no meaning. That is, GL in failure detection
The applicable range of R is within the Window section. And G
The preparations for the LR design include designing the Kalman filter and
Then, for failure detection, the failure model ρ_iAnd a
(K, τ, i) is calculated in advance, and on the other hand, the slip condition detection of the road surface is performed.
Change in vehicle model ΔA_h~ ΔD_hCalculate
Will be set. However, the vehicle models used for these designs
Dell is V_cIs a function of and Kalman fill
Failure model ρ_i, Vehicle model change ΔA _h~ ΔD_hIs
V_c= 10 km / h. This
The flow chart of the design
It shows in FIG.

The algorithm for detecting the above-mentioned failure / sliding condition of the road surface is performed in the following procedure based on the flowchart of FIG.

(1) Initialize in increments of k (step 112). (2) A Kalman filter, a failure model ρ _i , and vehicle model changes ΔA _{h to} ΔD _h are created (step 114).

That is, according to the flowchart of FIG. 13, a Kalman filter, a failure model ρ _i , and vehicle model changes ΔA _{h to} ΔD _h are prepared as a table at intervals of a vehicle speed V _c = 10 km / h. These are selected according to the vehicle speed V _c in the following calculation.

(3) The residual Box and state Box are initialized (step 116). You. At the time of initialization, the residual box is composed of 2 × 40 0 elements, and the state box is composed of 4 × 40 elements.

(4) Each sensor signal is read (step 118). The speed V _c and the rear wheel control amount u are fetched. However, the zero point correction of each sensor signal is performed before the operation of this detection algorithm.

(5) Estimate the vehicle slip angle β (step 120).

(6) Estimate lateral acceleration and yaw rate state quantities (step 122).

[0156]

The detected values of the acceleration sensor 22 and the yaw rate sensor 20 in step 118

(8) The residual Box and the state Box are updated (step 126). Insert at the end of x. Similarly, in step 118, the steering angle δf, the rear wheel control amount u, Update.

(9) The step k value and the window value are compared (steps 128 and 130). Tick k and Windo
w is compared, and if the step k is a value equal to or smaller than Window, the step k is set to k = k + 1, and the process returns to step 118 to input the sensor signal. This is because the residual time series pattern is Wi.
This is because the length of the window has not been obtained. If the step k exceeds Windows, the process proceeds to step 132.

(10) The GLR value and the failure amount of each failure model are calculated (steps 132 to 146).

Using the failure model ρ _{i of} step 114 and the residual Box of step 124, the GLR value and the failure amount of each failure model are calculated as described above.

(11) Determine the start of steering (step 1
48, 168). Whether or not the slip state of the road surface is detected is determined based on the steering angle δf. Here, the slip state of the road surface is detected when the absolute value of the steering angle δf exceeds 0.01 rad (corresponding to a steering angle of 8.4 deg). If the absolute value of the steering angle δf is 0.01 rad or less, the slip state of the road surface is not detected and the GLR of each road surface model is not detected.
The value is set to 0 (step 168) and the process proceeds to step 170.

(12) The GLR value of each road surface model is calculated (steps 150 to 166). Using the vehicle model changes ΔA _{h to} ΔD _{h in} step 114 and the state Box in step 126, each road surface model is created, and step 126
The GLR value of each road surface model is calculated using the residual box of the above.

(13) A failure / road slip condition is determined (step 170). Of the calculated GLR values, the largest one is assumed to be the anomaly currently occurring. And
It is determined whether or not the GLR value of each failure model or each road surface model having the maximum GLR value exceeds a set threshold value, and if it exceeds, it is determined that the abnormality is present. Further, when it is determined that the road surface is different from the reference, the slip state of the road surface is estimated depending on which road surface model has the maximum GLR value (h = 4 to 6). Here, each threshold value for detecting the failure and the slipping state of the road surface is set to be 2 to 3 times the maximum value of each failure at the time of lane change on the high friction coefficient road surface and each road surface model GLR value.
This is shown in Table 2.

[0165]

[Table 2]

Then, the determination result of the failure / slip condition of the road surface is output to the controller 18, and the process returns to step 118.

Based on the above-mentioned procedure, the road surface slip condition and the failure detection simulation were carried out using the running data and the failure test data when the 4WS vehicle was running at a vehicle speed of 80 km / h and the skid road (low friction coefficient). The result of performing is explained.

FIG. 15 shows the result when the lane was changed on a road surface with a high friction coefficient. As a result, the GLR value has a value due to some modeling error. On the other hand, FIG. 16 shows the determination result of the low friction coefficient road surface, and the GLR value increased with time along with the steering, and it could be determined that the road surface was the low friction coefficient road surface exceeding the threshold value. Furthermore, the model in which the cornering powers Cf and Cr during road surface running with a high friction coefficient are multiplied by 0.5 has the maximum value. This corresponds to the cornering power value adjusted in running on the skid road shown in FIG. 8, and by the method of this embodiment, the skid road surface can be known at the start of steering, and the degree of slippage of the road surface can be detected.

FIG. 17 shows each GLR calculated based on the traveling data at the time of step failure of the acceleration sensor. As a result, the GLR value of the acceleration sensor failure becomes the maximum value, and the correct failure detection can be performed. You can see that Further, FIG. 18 shows the estimation result of the failure amount. It can be seen from FIG. 18 that the estimation of the failure amount is also good. Also,
FIG. 19 shows each GLR value calculated based on the traveling data at the time of step failure of the actuator. As a result, it can be seen that the GLR value of the actuator failure becomes the maximum value and correct failure detection can be performed. Further, FIG. 20 shows the estimation result of the failure amount. From FIG. 20, the estimation of the failure amount of the actuator failure is not as accurate as that of the acceleration sensor failure, but is almost correct on the order.

However, in FIG. 16, the GL of the acceleration sensor failure is compared with the GLR value of the road surface model of Cf, Cr = 0.5.
The R value is very close to, and it is difficult to make a determination only by comparing the GLR values. On the other hand, in the acceleration sensor failure, the low friction coefficient road surface model and the GL of the acceleration sensor failure model
The R values can be clearly separated. Therefore, from the two calculation results, the GLR value of the road surface model and the GL of the acceleration sensor failure
If the difference between the R values is within 50, it is determined that the road surface is slippery rather than the acceleration sensor failure.

From these results, it can be expected to be used for ABS because the slip condition of the road surface can be estimated by this method. Further, the slippery road surface and each failure can be separated from each other, and the reliability of the slipping state detection of the road surface can be improved.

[0172]

As described above, according to the present invention, 2
Regardless of the vehicle control system such as WS, 4WS, etc., a very general sensor is used without requiring a special sensor such as a sensor for detecting a steering reaction force, and a slip state between a wheel and a road surface is used. It is possible to obtain an effect that the slip state of the road surface can be detected early in the early stage of steering before the vehicle reaches the spin state.

[Brief description of drawings]

FIG. 1 is a schematic diagram of an embodiment in which the present invention is applied to a 4WS vehicle.

FIG. 2 is a block diagram of a slipping state detecting device for a road surface.

FIG. 3 is a block diagram of a road surface slipping state detection circuit in FIG.

FIG. 4 is a schematic diagram for explaining the GLR method.

FIG. 5 is a diagram showing an actuator failure model ρ ₁ .

FIG. 6 is a diagram showing an acceleration sensor failure model ρ ₂ .

FIG. 7 is a diagram showing a yaw rate sensor failure model ρ ₃ .

FIG. 8 is a diagram showing a cornering power characteristic for each assumed road surface.

FIG. 9 is a diagram showing a steering angle, a lateral acceleration, and a yaw rate during traveling on a road surface having a high friction coefficient.

FIG. 10: Steering angle, lateral acceleration, and low friction coefficient when traveling on a road surface
It is a diagram showing a yaw rate.

FIG. 11 is a diagram showing a selection state of a state quantity X and a control quantity U at each τ.

FIG. 12 is a diagram showing a state in which a failure occurrence time τ is searched.

FIG. 13 is a flowchart for designing a failure / slip surface slip state detection system.

FIG. 14 is a flowchart for detecting a failure / sliding state of a road surface.

FIG. 15 is a diagram showing a GLR value and an abnormality determination result when traveling on a high friction coefficient road surface.

FIG. 16 is a diagram showing a GLR value and an abnormality determination result when traveling on a low friction coefficient road surface.

FIG. 17 is a diagram showing a GLR value and an abnormality determination result when the acceleration sensor fails.

FIG. 18 is a diagram showing an estimation result of a failure amount when an acceleration sensor fails.

FIG. 19 is a diagram showing a GLR value and an abnormality determination result when an actuator fails.

FIG. 20 is a diagram showing an estimation result of a failure amount when an actuator fails.

[Explanation of symbols]

 18 Controller 20 Yaw Rate Sensor 22 Acceleration Sensor 26 Vehicle Speed Sensor 24 Steering Angle Sensor 30 Sliding State Detection Device

Continuation of front page (51) Int.Cl. ⁶ Identification number Reference number within the agency FI Technical display location B62D 137: 00 (72) Inventor Masataka Osawa, Nagakute-cho, Aichi-gun, Aichi Prefecture 1 No. 41 Yokomichi Yokochi 1 Co., Ltd. Toyota Central Research Institute (72) Inventor Masaki Yamamoto 1 Toyota-cho, Toyota-shi, Aichi Toyota Automobile Co., Ltd.

Claims (1)

[Claims] 1. A vehicle steering amount, lateral acceleration, yaw rate,
Detecting means for detecting a state quantity related to vehicle motion including vehicle speed, reference vehicle motion model on a reference road surface having a predetermined slip, and reference vehicle motion of a vehicle motion model on another road surface having a slip different from the reference road surface At least one of lateral acceleration and yaw rate is estimated as a reference state quantity based on the vehicle movement model storage means that stores in advance the amount of change from the model, and the detected state quantity detected by the reference vehicle movement model and the detection means. Reference state quantity estimating means, reference residual calculating means for calculating a residual between the reference state quantity estimated by the reference state quantity estimating means and the detected state quantity detected by the detecting means as a reference residual, and Reference residual time series data storage means for storing time series data of the reference residual calculated by the reference residual calculation means, and a reference vehicle of the vehicle motion model Residual model calculation means for calculating a residual model value of at least one of lateral acceleration and yaw rate for another road surface based on the amount of change from the motion model and the detected state quantity detected by the detection means; A residual model time series data accumulating means for accumulating time series data of residual model values calculated by the model calculating means, and a time series pattern of the accumulated reference residual time series data and residual model time series data, And a road surface slip condition estimating means for estimating the slip condition of the road surface by comparing