Method for evaluating sensor signals
The method corrects offset errors and activates sensitivity monitoring based on dynamic states to enhance the accuracy and reliability of sensor signals in automobiles, particularly in autonomous driving systems.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-09-12
- Publication Date
- 2026-03-27
AI Technical Summary
Existing methods for evaluating sensor signals in automobiles, particularly in autonomous driving, struggle to accurately monitor sensitivity errors in inertial measurement units due to challenges in distinguishing between offset and sensitivity deviations, especially in varying dynamic conditions.
A method that involves offset correction during low-dynamics states and dynamic state evaluation to activate sensitivity error monitoring, using thresholds to ensure accurate sensitivity error detection, including a low-pass filter, offset correction, and relative deviation monitoring.
Improves the accuracy of sensor signal evaluation by effectively correcting offset errors and activating sensitivity monitoring only when necessary, thereby enhancing the reliability of sensor data in dynamic conditions.
Smart Images
Figure 2026054562000001_ABST
Abstract
Description
[Technical Field]
[0001] The present invention relates to a method for evaluating sensor signals and an apparatus for carrying out the method. [Background technology]
[0002] Conventional technology A sensor is a device used to detect a physical quantity and convert it into an electrical signal. This signal may be further processed to control a specific function or process. In automobiles, various sensors, often used in combination, are used to record various quantities. An inertial measurement unit (IMU) is a spatial combination of one or more inertial sensors, such as an accelerometer and a yaw rate sensor.
[0003] The signal provided by the sensor can be evaluated before use to determine whether the signal contains errors, that is, whether the value being carried as information by the signal has an error.
[0004] In automobiles, the use of reliable sensor signals is required in many applications, such as functional units in the realm of autonomous driving (AD). For this reason, a redundant IMU sensor architecture is typically used to detect sensor errors, employing similar physical events. For this purpose, signal deviations from redundant signals are monitored. An example of such an application is the use of three angular velocity sensors located on the same circuit board.
[0005] Effective redundant signals can be combined, sometimes referred to as fusion, or a selection algorithm can be implemented to select the "best" signal from all possible signals in terms of functional safety integrity and signal accuracy.
[0006] A set of redundant sensors may each have different characteristics. For example, some sensors may have larger sensitivity errors or scaling errors than others. It is clear that using signals with smaller errors can improve the accuracy of the final or last signal. [Overview of the project] [Problems that the invention aims to solve]
[0007] Disclosure of the invention Based on this background, we propose a method having the features of claim 1 and an apparatus as described in claim 8. Each embodiment is derived from each dependent claim and the detailed description. [Means for solving the problem]
[0008] The proposed method is based on the following considerations.
[0009] Safety thresholds are typically defined not only for offset errors but also for sensitivity deviations, or scaling errors. Similarly, in this example, monitoring these error patterns may be essential to take appropriate action when the sensitivity error exceeds the defined safety threshold. For example, the action could be to invalidate the signal if the diagnostic threshold is exceeded above the defined threshold.
[0010] Monitoring sensitivity errors individually as independent quantities is difficult to achieve. In the following explanation, we will further describe the monitoring units, namely Monitor 1 for absolute offset deviation and Monitor 2 for relative deviation in dynamic signals, which are exclusively offsets.
[0011] Please refer to Figure 1 for this. The relative error is given by the ideal signal S. id (t) or physical signal S phys (t) is defined as follows, where, S id (t)=S phys (t) holds. That is, here, there is no static offset, detection error, CAS (cross-axis sensitivity), or noise.
[0012] Error (s i or s ref ) of the real signal is given by
Equation
Equation
[0013] By ignoring the contributions of CAS and noise,
Equation
[0014] This means that in a signal without offset, the sensitivity error is SENS_ERROR = SF - 1 calculated by.
[0015] In this case, it is ideal for the scaling coefficient to be 1, that is, for an ideal signal, there is no deviation in the sensor signal.
[0016] If an offset remains, the following dependencies apply. Approach of Monitor 2: Comparison of relative differences in average signals
[0017] Reference signal S refIt is assumed that it is possible to generate, for example, the average value of all redundant signals applied as an ideal signal without offset or scaling errors.
number
[0018] therefore,
number
number
[0019] The relative deviation of the average signal is,
number
[0020] The relative deviation of a sensor signal from a reference, for example, the relative deviation of multiple signals from that signal from the average value, depends on the sensitivity error (SF-1) of the sensor signal channel and the offset error of the sensor signal.
[0021] Monitoring sensitivity errors, i.e., monitoring the relative deviation of the sensor signal output with respect to a reference, is degraded by residual offset in the signal. The error contribution here decreases as the magnitude of the reference signal increases.
[0022] An example is shown below. - Offset error present at signal 2 mg, no sensitivity error, no misalignment. - Low dynamics situations, for example, driving on a vehicle-only road without manual steering. -Actual acceleration in the Y direction: 20 mg ⇒((Sig-Ref) / Ref-1)*100=(-18 / 20-1)*100=-10%→Hypothetical sensitivity error based on offset - High dynamics situations, for example, cornering. -Actual acceleration in the Y direction: 200 mg ⇒((Sig-Ref) / Ref-1)*100=(-198 / 200-1)*100=-1%→Hypothetical sensitivity error based on offset.
[0023] Therefore, in order to improve the quality of sensitivity monitoring, two measures are necessary: 1) Reduce the signal offset before sensitivity monitoring. 2) Evaluate the sensitivity error only when the signal dynamics are sufficiently high. It is necessary.
[0024] A method for evaluating sensor signals within a vehicle is proposed. In one configuration, the sensor signal is the sensor signal of an inertial measurement unit. Sensitivity errors are identified within the evaluation range. The method includes the following steps: namely, performing offset correction in the sensor signal when a low-dynamics state of the vehicle exists. For example, low dynamics occurs when the vehicle is stopped.
[0025] Here, the dynamics must be low so that potential scaling errors do not affect the static offset value.
[0026] Next, the system evaluates the vehicle's dynamic state based on the sensor signals and compares the dynamic state with a threshold to evaluate the activation of the monitoring unit. When the dynamics are low, the monitoring of sensitivity error is minimal. When the dynamics are high, monitoring is active.
[0027] Next, in order to evaluate the sensor signal, a step is performed to calculate the relative deviation and compare that relative deviation with the limit value.
[0028] Thresholds (Thd) are values that can be used to define specific behaviors. For example, when a signal is greater than Thd_Start, sensitivity monitoring becomes active. Other thresholds are provided, for example, when it is defined whether the vehicle is in a state of low dynamics (low dynamic characteristics) (signal < Thd_lowdynamic → the state where the vehicle is in the stationary mode).
[0029] Limiting values also serve as thresholds. Here, the limiting values are used in a form where the detected sensitivity error is set to be invalid when it is smaller than the limiting value.
[0030] In this case, typically, it is configured to characterize or mark the sensor signal accordingly.
[0031] Therefore, an algorithm for sensitivity monitoring is presented that contributes to offset correction before monitoring. The algorithm further enables states based on vehicle dynamics.
[0032] The targeted approach is two activities, namely, 1. During low-dynamics driving conditions
Number
Number
[0033] Here,
Number
[0034] Further advantages and configurations of the present invention can be obtained from the detailed description and the attached drawings.
[0035] It is understood that each of the features described above and below can be used not only in the presented combinations, but also in other combinations or alone, without departing from the scope of the present invention.
Brief Description of the Drawings
[0036] [Figure 1] It is a graph showing signal characteristics for explaining a signal error model for an AD signal. [Figure 2] It is a graph showing signal characteristics for explaining the relative difference of an offset-corrected signal. This graph shows the concept of AD signal monitoring. [Figure 3] It is a block diagram showing the structure of sensitivity detection. [Figure 4] It is a flowchart showing a possible flow of the proposed method. [Figure 5] It is a highly simplified pure schematic diagram showing a vehicle equipped with a device for implementing the method.
Modes for Carrying Out the Invention
[0037] Embodiments of the Invention The present invention will be schematically shown in the drawings in accordance with embodiments, and will be described in detail below with reference to the drawings.
[0038] In FIG. 1, in graph 10 where time is plotted on the horizontal axis 12 and signal level is plotted on the vertical axis 14, the characteristics of sensor signals S120 and S222 and signals S id 24 and S ref 26 are shown. The threshold value (TH: threshold) 30 is marked on the vertical axis 14, that is, when S ref < TH 30, that is, a temporary assumption in a low-dynamics situation
Number
Number
[0039] In FIG. 2, in order to explain the definition of signal monitoring according to the monitor 2, in a graph 50 where time is plotted on the horizontal axis 52 and signal level is plotted on the vertical axis 54, the sensor signal S160 and the reference signal S ref 62 characteristics are shown. In FIG. 2, in particular, the difference between the signals after the offset is corrected is shown. This is understood from the fact that in the first static part, the two curves, namely, the "reference signal" and the affected signal, are the same. This is because the existing offset has been removed.
[0040] On the vertical axis 54, a threshold value TH70 is indicated, that is, when S ref <TH70, that is, a situation of low dynamics is indicated. The first bidirectional arrow 72 indicates the static region, and the second bidirectional arrow 74 indicates the dynamic region.
[0041] In FIG. 3, the sensitivity detection according to the monitor 2 is shown graphically. In the illustration, a low-pass filter (15 Hz) 100, an online offset corrector 102, three units 104, 106, 108 for forming an average value, a reference signal calculation unit (median) 110, a state calculation unit 112, and a unit 114 for threshold comparison and evaluation are shown. The inputs are the sensor signals S1120, S2122, and S3124. The other signals relate to additional information 126 and an activation signal / deactivation signal 128. The outputs are the valid signal S1130, the valid signal S2132, and the valid signal S3134.
[0042] Figure 4 shows a flowchart illustrating the possible steps of the proposed method. The method implements an algorithm that includes the following steps.
[0043] In the first step 200, all input signals are pre-filtered by a low-pass filter. Subsequently, in the second step 202, an average is formed for all signals across a specific number of datasets. Then, in the third step 204, a reference signal (median) is calculated based on the filtered redundant signals.
[0044] If the dynamics are sufficiently low, for example, when the system is stationary, an offset correction algorithm is applied to the input signal in step 206 before filtering. Low dynamics can be identified by analyzing all accelerations and angular velocities at a specific point in time. For example, in a stationary state, the angular velocity and acceleration are expected to be near 0 in the same plane, and the vertical acceleration is expected to be 1g. If these conditions are met, an offset relative to the reference is calculated and stored for each signal. Offset correction is performed continuously during normal operation. If the preconditions are met, the parameters are updated in the update cycle.
[0045] In step 208, the vehicle's dynamic state is evaluated based on the average signal, a process called state calculation. This dynamic state is then compared to a specific threshold that defines the activation of the monitoring itself. The determination is made as follows: a) The vehicle's dynamics are low, which leads to inactive sensitivity error monitoring. b) The vehicle has high dynamics, which leads to the need for active sensitivity error monitoring.
[0046] In step 210, the relative deviation described above is calculated during monitoring, and this relative deviation is compared to a specific limit value derived from the signal's safety target. If the limit value is exceeded, the signal is marked as invalid for the receiving unit. From this, safety monitoring for sensitivity errors is derived.
[0047] Figure 5 shows a pure schematic diagram of a vehicle, collectively denoted by reference numeral 300. The vehicle is equipped with an inertial measuring unit (IMU) 302 and an apparatus for carrying out the presented method according to one embodiment, the apparatus for carrying out the method, here denoted by reference numeral 304. The apparatus 304 has an evaluation unit for carrying out the method. The apparatus 304 and / or the evaluation unit 306 are integrated as hardware and / or software. Furthermore, the apparatus 304 may be integrated into the control device of the vehicle 300, or may be configured as such a control device.
[0048] The IMU302 transmits sensor signals 310 and 312, which are evaluated according to a method for evaluating these sensor signals 310 and 312 as presented herein.
Claims
1. A method for evaluating sensor signals (20, 22, 60, 120, 122, 124, 310, 312) within a vehicle (300), Sensitivity errors were identified within the evaluation range. The above method involves the following steps, namely: The steps include: performing offset correction in the sensor signals (20, 22, 60, 120, 122, 124, 310, 312) when a low dynamics state exists in the vehicle (300); The steps include evaluating the dynamic state of the vehicle (300) based on the sensor signals (20, 22, 60, 120, 122, 124, 310, 312), and comparing the dynamic state with thresholds (30, 70) to evaluate the activation of the monitoring unit, In order to evaluate the sensor signals (20, 22, 60, 120, 122, 124, 310, 312), the steps include calculating the relative deviation and comparing the relative deviation with the limit value, Methods that include...
2. The method according to claim 1, wherein the low dynamics state is determined by analyzing all accelerations and angular velocities at a specific point in time.
3. The method according to claim 1 or 2, wherein the sensor signals (20, 22, 60, 120, 122, 124, 310, 312) are filtered in advance by a low-pass filter (100).
4. The method according to any one of claims 1 to 3, wherein an average value is formed in the sensor signals (20, 22, 60, 120, 122, 124, 310, 312).
5. The method according to any one of claims 1 to 4, wherein a reference signal is calculated from the sensor signals (20, 22, 60, 120, 122, 124, 310, 312).
6. The method according to any one of claims 1 to 5, wherein, if the evaluation of the sensor signals (20, 22, 60, 120, 122, 124, 310, 312) reveals that the sensor signals (20, 22, 60, 120, 122, 124, 310, 312) deviate significantly with respect to their limit values, the signals are characterized accordingly.
7. The method described above is the method according to any one of claims 1 to 6, performed in an inertial measuring unit (302).
8. A device for evaluating sensor signals (20, 22, 60, 120, 122, 124, 310, 312), An apparatus comprising an evaluation unit (306) configured to carry out the method described in any one of claims 1 to 7.
9. The apparatus according to claim 8, wherein the apparatus is configured to evaluate the sensor signals (20, 22, 60, 120, 122, 124, 310, 312) of the inertial measurement unit (302).