Method for controlling a vehicle
Patent Information
- Application Number
- EP2023808687
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2022-11-18
- Filing Date
- 2023-11-09
- Publication Date
- 2025-09-24
AI Technical Summary
Existing methods for determining the speed of electric bicycles, such as using a magnet and reed sensor, can be faulty, leading to inaccurate torque support and potential speed exceedance, as sensor defects may go undetected.
A method involving multiple measurements of speed and acceleration using a speed sensor and machine learning models to determine vehicle speed and detect sensor errors by comparing measured and calculated speeds, allowing for vehicle control adjustments based on these comparisons.
This approach enables accurate speed determination and simple, cost-effective detection of sensor errors, ensuring reliable vehicle control by differentiating between faulty and functional sensors, reducing false positives and improving overall system reliability.
Smart Images
Figure 1.1
Abstract
Description
[0001] Description
[0002] title
[0003] Method for controlling a vehicle
[0004] The invention relates to a method for controlling a vehicle, in particular a single-track vehicle such as an electric bicycle.
[0005] The invention further relates to a vehicle, in particular a single-track vehicle such as an electric bicycle, with a speed sensor.
[0006] Although generally applicable to vehicles, the following invention is explained using electric bicycles.
[0007] State of the art
[0008] To determine the speed of electric bicycles, it has become known to place a magnet on the rear wheel of the electric bicycle. Using a magnetic field sensor – a so-called reed sensor – the passing of the magnet can be detected, and the speed of the electric bicycle can be determined based on the time interval between two detected magnetic signals. It is also conceivable to use a magnetic field sensor to measure the strength of a magnetic field, instead of a reed sensor, which only triggers at a predetermined magnetic field strength. The speed can then be determined based on the course of the measured magnetic field strength.
[0009] It's possible that the electric bike is only designed to provide torque assistance up to a certain speed. Therefore, accurate speed measurement is necessary to prevent torque assistance beyond the desired speed. The speed sensor may be faulty due to sensor defects.
[0010] Disclosure of the invention
[0011] In one embodiment, the present invention provides a method for controlling a vehicle, in particular a single-track vehicle such as an electric bicycle, comprising the steps:
[0012] Measuring the speed of the vehicle at several points in time using a speed sensor,
[0013] Measuring accelerations and / or rotation rates of the vehicle in at least one direction, preferably in three directions, at the plurality of times,
[0014] Determining a speed and / or a speed change of the vehicle based on the measured accelerations and / or angular rates at the multiple points in time by a machine learning model, and
[0015] Controlling the vehicle based on a comparison of the measured speeds and the determined speeds and / or determined speed change at the multiple points in time.
[0016] In one embodiment, the present invention provides a vehicle, in particular a single-track vehicle such as an electric bicycle, with a speed sensor, as well as: a first measuring unit, designed to measure a speed of the vehicle at a plurality of points in time using the speed sensor, a second measuring unit, designed to measure accelerations and / or yaw rates of the vehicle in at least one direction, preferably in three directions, at the plurality of points in time, a determination unit, designed to determine a speed and / or a speed change of the vehicle based on the measured accelerations and / or yaw rates at the plurality of points in time by means of a machine learning model, and a control unit, designed to control the vehicle based on a comparison of the measured speeds and the determined speeds and / or determined speed change at the plurality of points in time.
[0017] One of the advantages of this is that the vehicle's speed can be determined in two different ways. If the speed signals differ, this may indicate that the speed sensor is faulty. Another advantage is that it allows for a simple and cost-effective way to detect a sensor fault, allowing the vehicle to be controlled accordingly.
[0018] The phrase "comparison of the measured speeds and the determined speeds and / or determined speed change" is to be understood in the broadest sense and refers, in particular in the description, preferably in the claims, to a mathematical and / or statistical operation in which the measured and determined speeds and / or determined speed changes are linked. For example, this can be done using a mathematical comparison - the determined speed is greater than the measured speed - or by comparing the curves of the determined and measured speeds. It is also conceivable that a cumulative deviation between the measured and determined speeds is determined over several points in time and compared with a threshold value.
[0019] The term “sensor error” is to be understood in the broadest sense and refers in particular in the description, preferably in the claims, among other things to faulty measurement signals of the sensor and / or manipulated signals of the sensor.
[0020] The machine learning model can be configured to determine the determined speed based on regression and / or classification. In regression, the method is configured to determine an output variable, wherein the output variable is determined by the machine learning model using regression. The machine learning model can be configured as a trained machine learning system or as a machine learning system to be trained. In one embodiment, the trained machine learning model is not adapted during execution. The machine learning model can be executed locally, for example, by the electronics of the electric bicycle and / or a mobile device, and / or non-locally, for example, on a web server or a cloud.
[0021] In one embodiment, the machine learning model has a signal input through which the input variables for the classification are partially or completely provided. The provision can be effected, for example, by the electric bicycle, in particular by a sensor unit of the electric bicycle, and / or the mobile device, for example, a smartphone.
[0022] The term "regression" should be understood in particular to mean that an output variable is determined, whereby the output variable can be in the form of a physical variable or as a curve of a physical variable. The output variable can be in the form of a physical variable that can also be provided or is provided by another machine learning model or a sensor unit of the electric bicycle or the mobile device. The output variable can also be in the form of an output variable virtually derived from the input variables or its curve. The output variable can then be used to assign a class and / or to control a method, for example to control the electric bicycle.
[0023] The determined output value is preferably linked to a confidence level, where confidence is a measure that reflects the proximity of the input values to the training data used to train the machine learning model. For example, if the machine learning model was trained with training data that is very far removed from the input values, the confidence would be low.
[0024] The machine learning system can, for example, be designed as an SVM (support vector machine) algorithm, as at least one decision tree, in particular as a “random forest” method, as a Bayesian network or as a neural network, in particular as an MLP (multilayer perceptron).
[0025] The machine learning system can also be designed as a deep learning neural network, for example as a recurrent neural network (RNN), in particular as an LSTM network (long short term memory) or GRU network (gated recurrent units), as a CNN network (convolutional neural network), as a TCN network (temporal convolutional networks) or as a combination of the aforementioned networks.
[0026] Further features, advantages and further embodiments of the invention are described below or will become apparent thereby.
[0027] According to an advantageous development of the invention, the machine learning model is trained based on data from a fault-free journey of a vehicle. In particular, the machine learning model can be trained exclusively based on this data. With the embodiments of the invention, a faulty sensor can be detected. Since a potential fault does not need to be classified, the training data can be based solely on a fault-free journey. The machine learning model therefore detects faulty sensors by the fact that input data differs from the training data – which defines the fault-free state. Thus, the machine learning model can also detect previously unknown faults, since the machine learning model is not trained to find specific faults, but merely to determine a difference from fault-free data.In this way, even unknown errors can be easily detected, as can sensor faults. In particular, the machine learning model can be trained offline, i.e., outside the vehicle itself, since potential errors do not have to be included in the data.
[0028] According to an advantageous development of the invention, a sensor error is detected if a predeterminable deviation between the measured and determined speeds is detected over a period of at least 1 second, preferably at least 5 seconds, in particular at least 10 seconds. If the deviation is only detected for a short period of time, this is an indicator that the sensor is not faulty. One advantage of this is that the probability of false-positive sensor errors is reduced.
[0029] According to an advantageous development of the invention, for controlling the vehicle, a deviation value is determined based on the determined speeds and / or determined speed changes and the measured speeds, wherein the deviation value is compared with a deviation threshold value, wherein the threshold value is based on a detection rate, a maximum false positive rate, and / or a measurement period. The deviation value is a measure of the difference between the measured speeds and the determined speeds and / or determined speed changes. The deviation value is thus a direct measure of whether a sensor error is present. This value can be determined, for example, using the mean square error.It is also conceivable that the deviation threshold depends on other parameters, for example a desired detection rate for sensor errors, a desired false positive rate for sensor errors and / or a specific time period in which the differences are measured. It is also conceivable that the deviation value itself depends on the other parameters. One advantage of this is that the deviation threshold can be used as a direct measure of sensor errors. A further advantage is that the deviation threshold is adjustable; for example, a low deviation value could already be an indicator of a sensor error if a high detection rate is desired, or a high deviation value is required if a low false positive rate is desired.
[0030] According to an advantageous development of the invention, the machine learning model is provided in the form of a recurrent neural network, in particular in combination with a 1D convolution. Using the 1D convolution, patterns from a temporal progression of the yaw rate and / or acceleration signals and / or relationships between different signals such as acceleration and yaw rate can be recognized. One advantage of this is that a machine learning model can be easily created and trained. According to an advantageous development of the invention, the machine learning model determines the speed and / or the speed change based on a cadence, a driver torque, a drive unit speed, a drive torque, and / or a drive power. Parameters that describe the current driving state of the vehicle can be used as possible input variables for the machine learning model.Based on these variables, the speed or change in speed can be determined. One advantage of this is that the speed can be determined more accurately.
[0031] According to an advantageous development of the invention, for controlling the vehicle, a classification of the determined speed is determined based on the machine learning model, in particular a classification into specific speed intervals (ordinal classification), wherein the classification of the determined speed is compared with the measured speed. In this case, the machine learning model does not directly determine the speed, but merely an interval within which the speed lies. This allows a classification of the speed to be made, for example, "low, medium, high." It is also conceivable that this classification can be compared with the measured speed and / or with a classified measured speed in order to detect a sensor error.In other words, the machine learning model can be designed to determine the detected speed based on an ordinal classification. In this case, the machine learning model does not determine a discrete value for the detected speed, but rather always a preferably predefined range of values.
[0032] According to an advantageous development of the invention, the speed is determined by the machine learning model based on regression, in particular wherein the speed determined by regression is embodied as a discrete speed value. One advantage of this is that the speed can be determined more accurately. The regression can, in particular, be used to determine a discrete speed value, for example 25 km / h, in contrast to an ordinal or qualitative classification. According to an advantageous development of the invention, the speed is determined by the machine learning model based on classification. The speed determined based on the classification can be embodied as a speed value or as a numerical or qualitative speed range, for example "20 to 25 km / h" or "high speed".One advantage of this is that the speed can be determined in an efficient manner.
[0033] According to an advantageous development of the invention, the speed sensor is designed as a pulse-based speed sensor, in particular a reed sensor. One advantage of this is that the speed sensor can be implemented cost-effectively.
[0034] Further important features and advantages of the invention emerge from the subclaims, from the drawings and from the associated description of the figures.
[0035] 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.
[0036] Preferred embodiments and embodiments of the present invention are illustrated in the drawings and are explained in more detail in the following description.
[0037] It shows in schematic form
[0038] Figure 1 shows steps of a method according to an embodiment of the present invention;
[0039] Figure 2 A vehicle according to an embodiment of the present invention, Figure 1 shows in schematic form steps of a method according to an embodiment of the present invention.
[0040] In step S1, the speed of a vehicle is measured at several points in time using a speed sensor, in particular a pulse-based one. The pulse-based speed sensor can be a reed sensor, for example.
[0041] In a further step S2, the accelerations and / or yaw rates of the vehicle are measured in at least one direction, preferably in three directions, at the multiple points in time. In particular, the accelerations and yaw rates are measured in a travel direction, a transverse direction, and a vertical direction.
[0042] In a further step S3, a speed and / or a speed change of the vehicle is determined by a machine learning model based on the measured accelerations and / or yaw rates at the multiple points in time. In addition, other parameters such as drive power or the driver's cadence can be used to determine the speed. The machine learning model can, for example, be a neural network that has the accelerations and yaw rates in the various directions as input variables and the speed as the output variable. The speed can be determined at multiple points in time. The neural network has been trained with data from error-free sensors during a journey. The machine learning model can thus determine the vehicle's current speed.If the measured speed deviates from the detected speed, this indicates that the speed sensor is faulty. Since the speed is determined using a machine learning model, influences such as sensor drift or inaccurate sensors can be detected and / or compensated for.
[0043] In a step S4, the vehicle is controlled based on a comparison of the measured speeds and the determined speeds and / or determined speed change at the multiple points in time. In particular, the vehicle is controlled based on the detection of a sensor error in the speed sensor. The measured and determined speeds can be compared, and based on this, a value for a deviation between the two speeds can be calculated, which is then compared, for example, with a threshold value for the deviation. It is also conceivable that a speed change is determined and compared with the measured speed difference, in particular the difference between two speeds measured at subsequent points in time. The threshold value for the deviation can additionally be based on other parameters, for example, a desired maximum false positive rate.
[0044] The deviation value is a measure of the probability that the sensor is faulty. For example, a threshold can be defined, and if the deviation value is greater than the threshold, the sensor is detected as defective. Furthermore, the deviation value could represent the probability that the sensor is faulty. It is also conceivable for the deviation value to be a binary value that classifies the sensor as defective or functional. In particular, the deviation value can be based on measured and determined speeds at multiple points in time, so that even longer-term deviations can be detected. The longer the difference between the determined and measured speeds, the greater the probability that the sensor is faulty. The vehicle can be controlled accordingly.For example, if the speed sensor is expected to be faulty, the vehicle's drive assistance can be controlled depending on the detected speed instead of the measured speed.
[0045] It is also conceivable that the speed is determined qualitatively rather than quantitatively, i.e., it is determined whether the vehicle is within a certain speed range. The determined speed range can then be compared with the measured speed, and the deviation value can be determined based on this.
[0046] The machine learning model determines the vehicle's speed and compares the determined speed with the measured speed. This only identifies the presence of faulty sensors, but not the type of error. Therefore, not only known sensor errors but also unknown errors can be detected.
[0047] Figure 2 shows in schematic form a vehicle according to an embodiment of the present invention.
[0048] The vehicle 1, here in the form of an electric bicycle, has a speed sensor 6, here in the form of a pulse-based sensor, i.e. a reed sensor.The vehicle 1 further comprises: a first measuring unit 2, designed to measure the speed of the vehicle at a plurality of points in time using the speed sensor, a second measuring unit 3, designed to measure accelerations and / or yaw rates of the vehicle in at least one direction, preferably in three directions, at the plurality of points in time, a determination unit 4, designed to determine the speed and / or the speed change of the vehicle based on the measured accelerations and / or yaw rates at the plurality of points in time using a machine learning model, and a control unit 5, designed to control the vehicle 1 based on a comparison of the measured speeds and the determined speeds and / or determined speed change at the plurality of points in time.
[0049] The vehicle 1 is particularly designed to perform steps S1 to S4 according to Figure 1. The first measuring unit 2 can be formed integrally with the speed sensor 6.
[0050] Although the present invention has been described using preferred embodiments, it is not limited thereto but can be modified in many ways.
Claims
Claims 1. A method for controlling a vehicle (1), in particular a single-track vehicle (1) such as an electric bicycle, comprising the steps: Measuring (Sl) a speed of the vehicle (1) at several times using a speed sensor (6), Measuring (S2) accelerations and / or rotation rates of the vehicle (1) in at least one direction, preferably in three directions, at the plurality of times, Determining (S3) a speed and / or a speed change of the vehicle (1) based on the measured accelerations and / or rotation rates at the multiple points in time by means of a machine learning model, and Controlling (S4) the vehicle (1) based on a comparison of the measured speeds and the determined speeds and / or determined speed change at the plurality of points in time.
2. The method according to claim 1, wherein the machine learning model is trained based on data based on a fault-free drive of a vehicle (1).
3. Method according to one of claims 1-2, wherein a sensor error is detected if a predeterminable deviation between measured and determined speeds is detected over a period of at least 1 second, preferably at least 5 seconds, in particular at least seconds.
4. Method according to one of claims 1-3, wherein for controlling the vehicle (1) a value for the deviation is determined based on the determined speeds and / or determined speed changes and the measured speeds and wherein the value of the deviation is compared with a threshold value of the deviation, wherein the threshold value is a detection rate, a maximum false positive rate and / or a time period in which the measurement is taken.
5. The method according to any one of claims 1-4, wherein the machine learning model is provided in the form of a recurrent neural network, in particular in combination with a 1D convolution.
6. The method according to any one of claims 1-5, wherein in the machine learning model the determination (S3) of the speed and / or the speed change is carried out based on a cadence, a driver torque, a drive unit speed, a drive torque and / or a drive power.
7. The method according to any one of claims 1-6, wherein the determination (S3) of the speed by the machine learning model is carried out based on a regression, in particular wherein the speed determined by regression is formed as a speed value.
8. The method according to any one of claims 1-7, wherein the determination (S3) of the speed is performed by the machine learning model based on a classification.
9. Method according to one of claims 1-8, wherein the speed sensor (6) is designed as a pulse-based speed sensor, in particular a reed sensor.
10. Vehicle, (1) in particular a single-track vehicle (1) such as an electric bicycle, with a speed sensor (6), and: a first measuring unit (2) designed to measure a speed of the vehicle (1) at several points in time using the speed sensor, a second measuring unit (3) designed to measure accelerations and / or rotation rates of the vehicle (1) in at least one direction, preferably in three directions, at the several points in time, a determination unit (4) designed to determine a speed and / or a speed change of the vehicle (1) based on the measured accelerations and / or rotation rates at the plurality of points in time by means of a machine learning model, and - a control unit (5) designed to control the vehicle (1) based on a comparison of the measured speeds and the determined speeds and / or determined speed change at the plurality of points in time.