WO2024104689A1

WO2024104689A1 - Method for evaluating redundant signals

Info

Publication number: WO2024104689A1
Application number: PCT/EP2023/078952
Authority: WO
Inventors: Douglas MARQUES FANTINI
Original assignee: Robert Bosch Gmbh
Priority date: 2022-11-14
Filing date: 2023-10-18
Publication date: 2024-05-23
Also published as: DE102022212049A1

Abstract

The invention relates to a method for evaluating redundant signals in a motor vehicle, in which method a first signal is compared with a second signal, and a result of the comparison is compared with a threshold value, the threshold value being determined depending on a statistical evaluation of the signals and depending on information concerning a driving situation of the motor vehicle.

Description

Title

Method for evaluating redundant signals

The invention relates to a method for evaluating redundant signals and an arrangement for carrying out the method.

State of the art

During operation, a large number of signals are recorded in motor vehicles, each of which contains information about a technical quantity. This information is evaluated in the vehicle. To ensure safe operation, it is necessary to check whether the signals recorded are correct or faulty. One way to do this is to use or record redundant signals. This means that a technical quantity is recorded multiple times, i.e. there are at least two signals that represent this technical quantity. The plausibility of these signals can then be checked by comparing them. This also makes it possible to detect errors that occurred during the recording and/or forwarding of these signals.

One possible use of the method presented here can be in conjunction with an inertial measuring unit, for example. An inertial measuring unit is a spatial arrangement of several inertial sensors, such as acceleration sensors and yaw rate sensors. Such a measuring unit is used, for example, as a sensory measuring unit of an inertial navigation system.

With an inertial measuring unit, six possible kinematic degrees of freedom can be measured. For this purpose, three orthogonal Vertical acceleration sensors are used to record the translational movements and three rotation rate sensors arranged orthogonally to each other are used to record rotating or circular movements. In this case, the inertial measuring unit delivers three linear acceleration values for the translational movements and three angular velocities or rates for the rotation rates.

If at least two signals are recorded for one size, these are referred to as redundant signals. Concepts for handling these redundant signals, in particular to check their plausibility, i.e. to detect possibly faulty signals, are referred to as redundancy concepts.

A currently used redundancy concept implemented in inertial measurement units is called a correlation monitor and is based on comparing measurements of the same physical quantity, such as acceleration or angular rate, from different sources. For a given pair of redundant signals, two parameters are calculated, namely delta and the gradient. Delta is the difference between the current samples. The gradient is the difference between the signal rate of change, i.e. the discrete derivative. These parameters are calculated for each signal using the current sample and the previous sample.

Once delta and gradient are calculated, they are compared to thresholds. If these thresholds are exceeded, a counter is incremented. The conditions for incrementing this counter can be combined in three ways. These are: delta only, gradient only, delta and gradient. If both conditions are used to increment the counter, they are combined using a logical OR. This means that any condition that is met will cause the counter to be incremented. However, if both conditions are met, the counter is incremented once rather than twice. The signals are assumed to be correlated until the counter reaches a limit. Once this limit is reached, the counter value is set to a hold value that is higher than the limit. A correlation flag is also set. This means that the signals are marked as uncorrelated. The counter value is held at the hold value as long as the conditions for incrementing are met. When these conditions are no longer met, the counter starts to decrement until the value reaches zero and only then is the correlation flag removed or canceled. The signals are thus marked as correlated again.

This familiar procedure is performed using the last measured values and the values immediately before. In this sense, such an operation is practically instantaneous. There is thus no memory or storage in this comparison itself. However, this is currently provided for the correlation monitor by using counters to debounce the flags.

The adjustable parameters of this implementation for a redundant signal pair are listed below:

Threshold for Delta,

Threshold for the gradient,

Conditions for incrementing, delta only, gradient only or both, counter limit,

Counter hold value.

Disclosure of the invention

Against this background, a method with the features of claim 1 and an arrangement according to claim 11 are presented. Embodiments emerge from the dependent claims and from the description. The method presented is used to evaluate redundant signals, such as those recorded with an inertial measuring unit. However, the method is not limited to this implementation.

An embodiment of the presented method is based on acquired knowledge about the following challenges that arise when operating an inertial measuring unit.

Currently, the redundancy monitoring concept shows only a low performance when tested with fault injection using vehicle measurements from the test catalog. The known implementation is so far unable to detect most of the faults injected into one of the redundant signals. This causes the signals to be uncorrelated. Additionally, potential customers in this case aim to reduce thresholds for correlation.

Furthermore, it was recognized that failure cases and thresholds are not representative. Not only the failure types and their magnitudes, but also the thresholds in the correlation monitor implementation were defined under strong influence of the last generations of inertial measurement units, where the signals were expected to be used in braking systems (ESP: Electronic Stability Program) in most cases. Inertial measurement units are focused on autonomous driving applications, which requires rethinking the approach for failure and threshold definition.

Currently, the correlation monitor on inertial measurement units is a hard-coded static solution. This means that once the sensor is flashed, the parameters for the redundancy monitoring will remain independent of the application. No setting is equivalent to hard-coded. In addition, this feature does not take into account the vehicle dynamics in the current implementation. This means that the thresholds will also remain independent of the driving scenario. No adaptability therefore means static. The inability to set the parameters and adapt them to the environment runs the risk of current trends in the automotive sensor market, particularly in autonomous driving applications.

The method presented is used to evaluate redundant signals in a motor vehicle. In the method, a first signal is compared with a second signal and a result of the comparison is compared with a threshold value. The threshold value is determined depending on a statistical evaluation of the signals and depending on information about a driving situation of the motor vehicle.

The method presented redefines the way to monitor signal redundancy. To do this, error cases that are relevant for autonomous driving scenarios can be considered together with the vehicle dynamics and signal-to-noise properties. The basis for this is to find safety-critical decisions that are carried out primarily based on signals, e.g. the signals from an inertial measuring unit, and to define the maximum permitted signal deviation. The current driving situation and the influence of noise on the integrated signals are taken into account.

In the approach presented, a similar procedure is applied to the acceleration and angular rate signals within a typical autonomous vehicle architecture. First, a random walk, namely angle or position, is estimated depending on the signal. Then this estimate is integrated into the formulation of the maximum allowed signal deviation. Finally, the measured signal deviation is compared with the defined threshold value. The resulting information is converted into a flag.

At different points in the process, external parameters are used for the calculations, allowing the OEM (original equipment manufacturer) to shape the characteristics of the target vehicle and adapt it to the respective driving situation. In addition, the signal properties are estimated, using previous samples to dynamically evaluate the quality of the estimates. An embodiment of the presented method is discussed below: Random Walk

The first step is to define the error caused by the integration of the noise signals. For the angular rate signals, only one integration step is required to obtain the random angular random walk. For the acceleration signals, it is necessary to perform the integration twice to obtain the random position random walk. In both cases, the input parameters for the calculation are the integration time, the noise characteristics of the original signal and the signal sampling frequency. Refer to Figures 1 and 2. This step only takes into account the noise in the signals to estimate the random random walk. By varying the parameters, it is possible to define a characteristic surface, see Figures 3 and 4.

Acceleration signal deviation

Using longitudinal control, i.e. accelerator and brake, as a reference, the scenario considered here assumes that the vehicle is travelling on a straight road and suddenly needs to make an emergency stop using dead reckoning. The aim is to stop the vehicle within a predefined longitudinal safety limit. In this case, this is not possible. The vehicle should either perform an emergency avoidance procedure or trigger damage mitigation systems.

After mathematically transforming the equations for linear motion and discrete integration, together with some worst-case assumptions, the maximum allowable deviation is expressed by the following equation:

Where Vo is the initial longitudinal velocity of the vehicle, Ai _on is the expected longitudinal deceleration during braking, Sii _m is the longitudinal safety limit and dt is the signal sampling time, see Figure 5.

Angular rate signal deviation

Using lateral control, i.e. steering, as a reference, the scenario considered here assumes that the vehicle is driving on a curved road and needs to keep its lane using dead reckoning. The goal is to keep the vehicle within a predefined lateral safety limit. In this case, this is not possible. The vehicle should either initiate a decoupling or perform emergency braking.

After mathematical transformation of the equations for circular motion and discrete integration, together with some worst-case assumptions, the maximum allowable deviation is expressed by the following equation.

Where Vi _on is the longitudinal velocity of the vehicle, Tdr is the coupling time, Siim is the lateral safety limit and dt is the signal sampling time, see Figure 6.

Debounce and final marker arrangement

The final step in this implementation of the method is to estimate the estimated values of the random walk using the threshold formulation and compare them with the measured signal deviation. The implementations for acceleration and angular rate differ slightly, but the strategy is the same. It is necessary to comparison before the result is converted into a correlation flag, see Figures 7 and 8.

If necessary, some parameters described in this procedure can be hard-coded. However, the idea is to have a generic description that can be adapted to each specific case.

The arrangement described is designed to carry out the method presented here and can be implemented, for example, in hardware and/or software. Furthermore, the arrangement can be integrated in a control unit of a vehicle or designed as such.

Further advantages and embodiments of the invention will become apparent from the description and the accompanying drawings.

It is understood that the features mentioned above and those to be explained below can be used not only in the combination specified in each case, but also in other combinations or on their own, without departing from the scope of the present invention.

Short description of the drawings

Figure 1 shows a diagram of the calculation of the random position random walk.

Figure 2 shows a diagram of the calculation of the random angle random walk.

Figure 3 shows a graph of a position random walk estimation surface.

Figure 4 shows an angle random walk estimation surface in a graph.

Figure 5 shows a graph of the maximum deviation for acceleration signals. Figure 6 shows a graph of the maximum deviation for angular rate signals.

Figure 7 shows a flow chart of an embodiment of the presented method.

Figure 8 shows a flow chart of another embodiment of the presented method.

Embodiments of the invention

The invention is illustrated schematically by means of embodiments in the drawings and is described in detail below with reference to the drawings.

The method is described below in connection with an inertial measuring unit, but is not limited to this design, but can be used anywhere in the motor vehicle, especially where redundant signals are to be evaluated.

Figure 1 shows the calculation of the random position in a diagram 10. Please note:

The vehicle uses the inertial measurement unit for dead reckoning for Tdr seconds and the only difference between the redundant signals is given by the noise parameters. The noise in the redundant signals is integrated twice. The standard deviation for the resulting position, the linear random walk, is found as a function of the noise characteristics in Aon, i.e. OA and OB in the redundant signals, Tdr and the signal sampling time dt.

Input variables 12 are the integration time Tdr, the sampling frequency 1/dt and the noise of the acceleration sensor OACC. Output variable 14 is the position random walk op _0S . A first graph 16 shows acceleration curves, a second graph 18 shows speed curves, a third graph 20 shows positions, a fourth graph 22 shows a histogram of noise, a fifth graph 24 shows a histogram of endpoints and a sixth graph 26 also shows a histogram of endpoints.

Figure 2 shows the calculation of the angle random walk in a diagram 40. Please note:

The vehicle uses the inertial measurement unit for dead reckoning for Tdr seconds and the only difference between the redundant signals is given by the noise parameters. The noise in the redundant signals is integrated once. The standard deviation for the resulting position, the angular random walk, is found as a function of the noise characteristics in yaw rate, i.e. OA and OB in the redundant signals, Tdr and the signal sampling time dt.

Input variables 42 are the integration time Tdr, the sampling frequency 1/dt and the rate noise ORate. Output variable 44 is the angular random walk OAng. A first graph 46 shows the angular rate, a second graph 48 shows angle, a third graph 50 shows a histogram of noise and a sixth graph 52 shows a histogram of end points.

Figure 3 shows a position random walk estimation surface at a sampling rate of 1000 Hz in a graph 80, on one of whose abscissas 82 the signal noise is plotted, on the other abscissa 84 the integration time and on the ordinate 86 the position random walk.

Figure 4 shows an angle random walk estimation surface at a sampling rate of 1000 Hz in a graph 100, on whose one abscissa 102 the signal noise is plotted, on whose other abscissa 104 the integration time is plotted and on whose ordinate 106 the angle random walk is plotted.

Figure 5 shows a graph 120, on one of whose abscissas 122 the initial vehicle speed, on the other abscissa 124 the expected maximum deviation and on the ordinate 126 the maximum permissible deviation is plotted, an area of maximum deviation for acceleration signals.

Figure 6 shows an area of the maximum deviation for angular rate signals in a graph 140, on one abscissa 142 of which the initial vehicle speed is plotted, on the other abscissa 144 the integration time and on the ordinate 146 the maximum permissible deviation.

Figure 7 shows a flow chart of a possible sequence of the presented method and schematically illustrates the implementation of a redundancy monitoring for acceleration signals in an arrangement 198 for carrying out the method.

Input variables are an acceleration signal (dt) A 200 and an acceleration signal (dt) B 202. First, a statistical evaluation 204 takes place, which in a step 206 includes the calculation of an absolute difference (delta), in a step 208 a calculation of a moving standard deviation based on N previous samples of the input variable A 200, in a step 210 a calculation of a moving standard deviation based on N previous samples of the input variable B 202 and in a step 212 a moving average based on the N previous samples.

Furthermore, OEM inputs 222 are taken into account for the driving situation 220, namely an integration time Tdr 230, a safety limit Sii _m 232, a maximum expected deceleration Ai _on 234 and a vehicle speed Vo 236.

Taking into account the integration time Tdr 230 and the sliding standard deviation 208, a first expected random walk 240 results. Taking into account the integration time Tdr 230 and the sliding standard deviation 210, a second expected random walk 242 results. From the two random walks 240 and 242, a worst-case random walk combination is determined in a step 250. Then, taking into account the In a step 254, a safety limit is set for the predetermined safety limit Sii _m 232 so that it includes the random walk.

Taking into account the maximum expected deceleration Ai _on and the vehicle speed Vo, the maximum permissible deviation A _max is then determined in a step 260 using equation (1). The counter is then debounced to approximately 4 in a step 262. Finally, a correlation monitoring status 270 is output.

Figure 8 shows a flow chart of a possible sequence of the presented method and schematically illustrates the implementation of a redundancy monitoring for angular rate signals in an arrangement 298 for carrying out the method.

Input variables are an acceleration signal (dt) A 300 and an acceleration signal (dt) B 302. First, a statistical evaluation 304 takes place, which in a step 306 includes the calculation of an absolute difference (delta), in a step 308 a calculation of a moving standard deviation based on N previous samples of the input variable A 300, in a step 310 a calculation of a moving standard deviation based on N previous samples of the input variable B 302 and in a step 312 a moving average based on the N previous samples.

Furthermore, OEM inputs 322 are taken into account for the driving situation 320, namely an integration time Tdr 330, a vehicle speed Vi _on 332 and a safety limit Sn _m 334.

Taking into account the integration time Tdr 330 and the moving standard deviation 308, a first expected random walk 340 results. Taking into account the integration time Tdr 330 and the moving standard deviation 310, a second expected random walk 342 results. A worst-case random walk combination is determined from the two random walks 340 and 342 in a step 350. Then, in a step 354, a value arccos is determined so that it includes the random random walk. Taking into account the vehicle speed Vi _on and the safety limit Sii _m 334, the maximum permissible deviation A _ma x is then determined in a step 360 using equation (2). The counter is then debounced to approximately 4 in a step 362. Finally, a

Correlation monitor status 370 returned.

Claims

Expectations

1. Method for evaluating redundant signals in a motor vehicle, in which a first signal is compared with a second signal, a result of the comparison is compared with a threshold value, wherein the threshold value is determined as a function of a statistical evaluation (204, 304) of the signals and as a function of information on a driving situation (220, 320) of the motor vehicle.

2. Method according to claim 1, in which, within the framework of the statistical evaluation (204, 304), a random random walk is first determined taking into account an integration of the signals over time.

3. Method according to claim 1 or 2, wherein the statistical evaluation (204, 304) comprises at least one calculation step selected from a group consisting of: calculation of an absolute difference (206), calculation of a moving standard deviation (208, 210), calculation of a moving average (212).

4. Method according to one of claims 1 to 3, wherein the information on the driving situation (220, 320) is selected from a group consisting of: integration time (230), vehicle speed (236), safety limit (232), maximum expected deceleration (234).

5. Method according to one of claims 1 to 4, which is carried out in conjunction with an inertial measuring unit.

6. The method according to claim 5, wherein signals from at least one acceleration sensor are evaluated.

7. Method according to claim 5 or 6, in which signals from at least one rotation rate sensor are evaluated.

8. Method according to one of claims 1 to 7, which is used in an autonomously operated vehicle, wherein error cases relevant to autonomous driving scenarios are taken into account.

9. Method according to one of claims 1 to 8, wherein a counter is incremented when the threshold value is exceeded.

10. The method according to claim 9, wherein the counter is decremented when the threshold value is undershot.

11. Arrangement for evaluating redundant signals in a motor vehicle, which is set up to carry out a method according to one of claims 1 to 10.

12. Arrangement according to claim 11, which is designed for use in conjunction with an inertial measuring unit.