Method for carrying out a highly accurate speed determination in a motor vehicle with the aid of a kalman filter

Abstract

The invention relates to a method for carrying out a highly accurate speed determination and for detecting slip of the wheels in a motor vehicle with the aid of a Kalman filter and data from a wheel speed sensor system, having the following steps: a) receiving data from a wheel speed sensor system which is configured to detect the wheel speed of at least one wheel of the motor vehicle; b) receiving further data, which contain information relating to the vehicle speed, from an inertial sensor system, which is configured to detect accelerations of the motor vehicle; c) carrying out an analysis of the data received in steps a) and b) in order to detect slip of the wheels of the motor vehicle using a Kalman filter; and d) processing the data received in steps a) and b) in a Kalman filter for highly accurate speed determination, wherein if slip of the wheels was detected in step c), the data received in step a) are temporarily suppressed when determining the speed.

Description

[0001] Description

[0002] title

[0003] Method for performing a highly accurate speed determination in a motor vehicle using a Kalman filter

[0004] State of the art

[0005] The invention relates to methods for performing a highly accurate speed determination in a motor vehicle using a Kalman filter. Within the framework of the method for performing a highly accurate speed determination, the invention particularly addresses the automated slip detection of wheel speed measurements by means of statistical tests, which are performed in particular in a Kalman filter, in order to generate highly accurate estimates of the vehicle speed.

[0006] Estimating the speed and its derivative (acceleration) of a vehicle using data from wheel speed sensors (WSS) and inertial measurement units (IMU) is of great importance for driving safety systems (e.g., ABS, ASR, and ESP) as well as for automated and autonomous driving functions (e.g., AEB, ACC, and lane keeping assist).

[0007] Data measured using wheel speed sensors are generally considered particularly reliable because, unlike data measured with GNSS sensors, for example, such data is obtained via a fault-protected chain of events. From the rotational speed or change in rotational speed of the wheels, the vehicle's speed and acceleration can be determined in conjunction with a few other data points.

[0008] In contrast, data obtained via GNSS has a multitude of additional dependencies that can distort the data, such as atmospheric disturbances, signal distortions, signal obscurations, etc.

[0009] Data obtained using inertial sensors has different properties than data measured with wheel speed sensors. Inertial sensors can only measure accelerations (i.e., changes in velocity), never velocity itself. Determining velocity data from inertial sensors always requires an integration step. The acceleration values ​​from the inertial sensor must be integrated with the velocity values. Therefore, the numerical accuracy of signal measurements in inertial sensor data has a significant impact on determining velocity data. Furthermore, an initial velocity value is required, from which the changes in velocity (the accelerations) measured by the inertial sensor are integrated.

[0010] A Kalman filter is frequently used for position estimation and determination based on GNSS signals. Position determination is always performed taking previous position measurements into account. The basic operating principle of this Kalman filter is as follows: Previous position measurements are updated with data from an inertial sensor to estimate the current position. The current position is determined from the previous position and integrated data from the inertial sensor. This is called the prediction step. This position is then corrected with new (current) GNSS signals to estimate the current position. This is called the innovation step. The Kalman filter typically calculates a weighting factor, which is used to balance the position determined in the prediction step with the position measured by GNSS signals.

[0011] This document describes a novel method for detecting wheel slip, which can distort the readings from wheel speed sensors used to determine vehicle speed. This method employs a Kalman filter to process the wheel speed sensor readings. The invention achieves this goal according to the features of the independent claims. Further advantageous embodiments are described in the dependent claims, the description, and particularly in the figure description. It should be noted that a person skilled in the art would combine the individual features in a technologically meaningful way to arrive at further embodiments of the invention.

[0012] This document describes a method for performing a highly accurate speed determination and detecting wheel slip in a motor vehicle using a Kalman filter, comprising the following steps: a) Receiving data from a wheel speed sensor configured to detect the wheel speed of at least one wheel of the motor vehicle; b) Receiving further data containing information about the vehicle speed from an inertial sensor configured to detect accelerations of the motor vehicle; c) Performing an analysis of the data received in steps a) and b) to detect wheel slip of the motor vehicle using a Kalman filter;and d) processing the data received in steps a) and b) in a Kalman filter for highly accurate speed determination, wherein if wheel slip was detected in step c), the data received in step a) is temporarily suppressed during speed determination.;

[0013] Basically, the method described here is about transferring the logic of plausibility checks of sensor signals, which is applied in the processing of GNSS signals together with inertial sensor signals, to the processing of signals from a wheel speed sensor system together with inertial sensor signals.

[0014] The procedure is particularly advantageous if the analysis in step c) includes at least one statistical test to detect data received in step a) that is affected by slippage.

[0015] Such statistical tests are generally known in connection with Kalman filters to identify erroneous signals by checking their plausibility. However, these tests are applied somewhat differently, or rather adapted, for the procedure described here. Wheel speed signals affected by slippage are not inherently erroneous, since the true wheel speed is measured, but this differs from the vehicle speed. A vehicle speed determined based on the true wheel speeds is only erroneous if the vehicle speed is directly inferred from the wheel speeds and slippage is present. Therefore, wheel speed sensor signals affected by slippage are defined as "erroneous" for the purposes of these tests.

[0016] Kalman filters are regularly used to estimate vehicle speed and acceleration. They can be used to describe the motion of vehicles, spacecraft, and aircraft. A typical function implemented with Kalman filters involves sequentially generating prediction steps based on measurement data from an inertial measurement unit (IMU) and innovation steps based on position data from a GNSS system. The results of the prediction and innovation steps are sequentially maintained by the Kalman filter, thus continuously providing highly precise measurements of the vehicle's position, speed, and acceleration.

[0017] Here, it is proposed to use a Kalman filter for processing data from a wheel speed sensor (see step c)).

[0018] The analysis of whether input data is faulty, as applied in step c), is common practice, for example, when analyzing GNSS data as input. The following explanations of the execution of step c) are described for GNSS data and, within the scope of this invention, are applied to data from a wheel speed sensor system. GNSS data can be faulty, for example, if GNSS signals are (partially) blocked by obstacles. Therefore, a technique using statistical tests, such as an x² test or a z-test, is necessary to exclude faulty data. For example, the following hypothesis can be tested:

[0019] Ho: The innovation v( / c) is consistent with the innovation covariance matrix S(k). For this, the NIS metric [NIS = Normalized Innovation Square] e can be used, for example: e v = v T S~ v.

[0020] Here, v is the innovation vector.

[0021] S: the innovation covariance matrix

[0022] The Great e v The NIS metric is x²-distributed with m degrees of freedom, where m is equal to the dimension of the measurement vector. This allows a p-value to be determined for the NIS metric at time k. The p-value is defined as the probability of obtaining a test result that is at least as extreme as the observed result, assuming the null hypothesis is true. It thus measures the strength of the evidence against the null hypothesis.

[0023] The null hypothesis is rejected if p < a, where a represents the significance level, which can be, for example, 5%.

[0024] A Type I error occurs when the null hypothesis is incorrectly rejected when it is true. The probability of a Type I error is determined by the significance level α of the test.

[0025] A Type II error occurs when the null hypothesis is not rejected even though it is false. The probability of a Type II error depends, among other things, on the sample size and the actual deviation from the null hypothesis. It can be investigated, for example, through simulations.

[0026] In safety-critical vehicle systems (such as ABS, but also more complex applications of autonomous / highly automated driving), measurement data obtained solely from GNSS systems cannot be used to control the respective safety-critical system, as a certain level of safety (ASIL = Automotive Safety Integrity Level) must be ensured. Therefore, it is proposed here to use data from wheel speed sensors and process it in a Kalman filter. This allows highly precise speed data to be determined based on the wheel speed sensors, which can then be enhanced with data from inertial sensors and a Kalman filter.

[0027] In the method described here, a Kalman filter is used, for example, to verify the plausibility of measurement data of the wheel speeds VFL [FL = front left], VFR[FR = front right], VRL[L = rear left], VRR [RR = rear right].

[0028] Previously, the following other approaches existed to detect wheel speeds affected by slippage:

[0029] Model-based slip correction attempts to adjust wheel speed to account for slip. However, such models also offer only limited possibilities for representing reality. Wheel slip depends on numerous variables, such as vehicle acceleration, road surface conditions, and so on. A model can therefore only guarantee limited accuracy in its predictions. In particular, situations also arise in which a wheel is faster or slower than the vehicle, even when there is explicitly no slip.

[0030] By evaluating signals such as wheel acceleration and wheel pressure, situations in which the wheels are likely to be unstable or subject to high slippage should be detected.

[0031] In contrast to these other approaches, the method described here uses the data received in step b) to perform a prediction step in the Kalman filter, and the data received in step a) to perform an innovation step in the Kalman filter.

[0032] The method described here employs a (proprietary) Kalman filter to validate data from wheel speed sensors. This involves a prediction step using additional information regarding the vehicle speed derived from inertial sensors. An innovative step is performed using the wheel speed sensor data. The Kalman filter is used in essentially the same way as for processing GNSS signals. Instead of GNSS signals, wheel speed sensor data is processed within the Kalman filter. The method described here automates the processing of wheel speed sensor data (especially wheel rotation signals) in a Kalman filter along with data from inertial sensors. This is particularly effective with data from high-precision IMU sensors, which allow for very accurate prediction of the vehicle speed. A statistical test procedure (such as...) is then used to verify the accuracy of the data.A x2 test of the Normalized Innovation Squared metric can detect wheel slip based on the Kalman filter prediction data. By excluding slip-related signals from wheel speed sensors, the method described here achieves higher accuracies than model-based methods or methods that model slip via process noise.

[0033] Based on the results of this procedure, faulty signals, or signals that are considered faulty because they do not allow for a sufficiently reliable determination of the (actual) vehicle speed, should be excluded. In these situations, only the data received in step b) (from an inertial sensor) are used (temporarily) to determine the speed, together with the previous state of the Kalman filter (step d)). While wheel slip signals are not faulty (as described above) because the true wheel speed is measured, this speed differs from the vehicle speed. However, here, a wheel speed that deviates from the vehicle speed is defined as faulty.This allows for automated detection of these deviations using a suitable method in a Kalman filter, resulting in a highly accurate velocity estimate.

[0034] According to the procedure described here, the following null hypothesis is tested using a statistical method before each innovation step of the Kalman filter:

[0035] Ho: The innovation v( / c) is consistent with the innovation covariance matrix S(k) of the Kalman filter.

[0036] This is the case when: f(vpL, VFR, VRL, VRR) » V xThen the speed determined from the wheel speeds is equal to the vehicle speed. In the simplest case, the function f is the average of all speeds resulting from the four wheels. In reality, more complex relationships can exist. Fewer than four wheel speed sensors can also be used. Steering angle sensors can be used as well. Other sensors (e.g., sensors to determine the vehicle's yaw rate or sensors to determine forces acting on the vehicle, such as brake pressure and torque sensors) can also be used. With V x The actual vehicle speed is described here.

[0037] If f (VFL, VFR, VRL, VRR) deviates significantly from the true vehicle speed (e.g., due to wheel slip), the hypothesis Ho is rejected. In this case, the Kalman filter is not corrected using the wheel speeds. The states of the Kalman filter reflect or incorporate previous speed measurements. Therefore, the estimation of the true vehicle speed is not based solely on inertial sensors. The data from the inertial sensors are preferentially processed with the Kalman filter to achieve the estimation of the true vehicle speed.

[0038] Other conditions may also need to be met, depending on the dimension of the innovation vector v.

[0039] Based on the test of hypothesis Ho, the Kalman filter is preferentially set to utilize the result of the test of hypothesis Ho to generate the best results.

[0040] It is particularly preferred if, in step d), the data received in step a) are evaluated with regard to their quality.

[0041] In preferred implementation variants, in step d) the consideration of data received in step a) in the Kalman filter is suppressed if slack was detected in step d).

[0042] To ensure the most accurate possible execution of the prediction step, the process noise covariance matrix (Q matrix) of the Kalman filter is preferably adjusted beforehand, during the design of the Kalman filter. The Q matrix is ​​preferably set such that a high percentage of the "true" states lie within the sigma band defined by the error covariance matrix.

[0043] To adjust the process noise covariance matrix (Q matrix), the innovation step of the Kalman filter is preferably deactivated for a few seconds, and only continuous prediction is performed. To obtain the best possible starting point for the continuous prediction, the innovation step should not be performed with noisy measurement data such as that from the WSS until this point, but rather with highly accurate reference measurement data.

[0044] Operation with the innovation step switched off is achieved, for example, by adjusting the Q-matrix so that when the innovation step is repeatedly switched off for a time interval (e.g., 20 seconds), a percentage of the true states x(i) (e.g., approximately 95%) lie in the interval [x(i) - 2o, x(i) + 2o], where x(i) denotes the predicted states at time i.

[0045] In further implementation variants, the measurement noise covariance matrix (R matrix) is specifically set in advance during the design of the Kalman filter for the execution of the procedure.

[0046] The R matrix should also be initially calibrated using reference measurement data. The R matrix should be chosen such that approximately a of the measurement data are rejected by the selected statistical test, where a is the significance level.

[0047] In this step, it can be useful to slightly distort the measurement data of the reference system. This can be done, for example, with a noise term; a fixed offset or increase to adjust the noise terms is also possible. This allows the innovation step in the Kalman filter to be carried out even with slightly slipping wheels.

[0048] If a very accurate IMU sensor allows for good prediction, then this distortion need not be very large. The less accurate the IMU sensor, the more noise must be tolerated, as the Kalman filter then requires more frequent support through an innovation step. The R matrix can, for example, be used with the following measured values ​​for state V. ; with a normal distribution N (1 , 0.001):

[0049] V X ADMA • y, where y ~ N (1 , 0.001)

[0050] Preferably, validation and adjustment are also performed over several iterations. In one validation step, the Kalman filter is now tested with the WSS measurement data. In particular, the type II error for different slip values ​​is to be investigated using simulation. It may then be necessary to repeat the previous steps (especially the adjustment of the R matrix) and make minor adjustments.

[0051] Also described here is a device for data processing, comprising a processor configured to perform the described procedure.

[0052] The following will be described in more detail: a computer program product, comprising commands that, when executed by a computer, cause the computer to execute the described procedure.

[0053] Furthermore, a computer-readable storage medium is described, comprising instructions which, when executed by a computer, cause it to perform the described procedure.

[0054] The invention and its technical context are explained in more detail below with reference to the figures. The figures show preferred embodiments, to which the invention is not limited. It should be noted in particular that the figures, and especially the size relationships shown in the figures, are only schematic. The figures show:

[0055] Fig. 1: a flowchart of the operation of the Kalman filter when using the method described here:

[0056] The following variables are used in Fig. 1 to describe Kalman filter 1: Xk System state

[0057] Pk Covariance matrix of the system state

[0058] A transition matrix

[0059] Q Covariance matrix of process noise

[0060] W Covariance matrix of the process noise at prediction step e v normalized squared innovation matrix a significance level m dimension of the measurement vector

[0061] H observation matrix

[0062] R Covariance matrix of measurement noise

[0063] S Innovation covariance matrix v Innovation vector

[0064] K Kalman Gain Matrix

[0065] Figure 1 shows the step-by-step operation of a Kalman filter 1. For each step, there is an initial state 2, described by the two vectors Xk and Pk. After each step, k is incremented by 1, and the output data of one step of the Kalman filter becomes the input data for the next step, or rather, it describes the initial state for the next step. This is represented by the upward-pointing arrows, each indicating the increment of k according to the equation k = k + 1. k is the index that describes the individual velocity determination steps performed with the Kalman filter 1.

[0066] Starting from the initial state 2, prediction step 3 is executed. This is followed by error test 4. Error test 4 is preferably an x2 test, as described above. Depending on the result of error test 4, either a first path 7 or a second path 8 is followed. In the first path 7, it is assumed that the wheel speed sensor readings are error-free. An innovation step 5 (as usual) is performed, and a speed output is generated. In the second path 8, it is assumed that the wheel speed sensor readings are faulty. Innovation step 5 is not performed. Faulty signals from a wheel speed sensor that do not provide a reliable indication of the vehicle speed are not considered. A speed output 6 is generated without innovation step 5.

Claims

Patent claims 1. A method for performing a highly accurate speed determination and for detecting wheel slip in a motor vehicle using a Kalman filter and data from a wheel speed sensor system, comprising the following steps: a) receiving data from a wheel speed sensor system configured to detect the wheel speed of at least one wheel of the motor vehicle; b) receiving further data containing information relating to the vehicle speed from an inertial sensor system configured to detect accelerations of the motor vehicle; c) processing the data received in steps a) and b) to detect wheel slip of the motor vehicle using a Kalman filter;and d) processing the data received in steps a) and b) in the Kalman filter for highly accurate speed determination, wherein if wheel slip was detected in step c), the data received in step a) is temporarily suppressed during speed determination.; 2. Method according to claim 1, wherein the analysis in step c) comprises at least one statistical test for detecting data received in step a) that is affected by slippage.

3. Method according to one of the preceding claims, wherein a prediction step is performed in the Kalman filter using the data received in step b) and wherein an innovation step is performed in the Kalman filter using the data received in step a).

4. Device for data processing comprising a processor configured to perform the method according to any one of claims 1 to 3.

5. Computer program product comprising instructions which, when the computer program product is executed by a computer, cause the computer to execute the method according to any one of claims 1 to 3.

6. Computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to execute the method / steps of the method according to any one of claims 1 to 3.

Images