Method and apparatus for estimating steering torque in a vehicle steering system

Abstract

The present invention relates to a computer-implemented method and apparatus (1) for estimating steering torque in a steering system of a vehicle and / or machine.

Description

[Technical Field]

[0001] The present invention relates to a method for estimating steering torque in a vehicle steering system. The present invention further relates to a device for estimating steering torque in a vehicle steering system. [Background technology]

[0002] Conventional technology The vehicle steering system is a crucial component of modern vehicles, enabling the driver to reliably control the vehicle. Traditionally, a steering wheel and / or pedal control system is used to introduce steering motion or for vehicle control. Furthermore, vehicle control using a joystick is one alternative method to the traditional steering wheel and pedal control system of a vehicle. In this case, there are various types of joysticks that can be used for vehicle control, including one-handed and two-handed joysticks.

[0003] There are various types of steering systems, including mechanical, hydraulic, electric, and electro-hydraulic steering systems.

[0004] Mechanical steering systems include, for example, rack-type steering systems, / or ball-circulating steering systems, and / or worm-type steering systems. Hydraulic steering systems use hydraulic pressure to steer the wheels. In contrast, electric steering systems use electric motors to assist and / or perform steering motions. Electro-hydraulic steering systems combine the advantages of both systems by combining a hydraulic pump with an electric motor.

[0005] In modern vehicles, so-called steer-by-wire (SbW) steering systems are also very popular. A steer-by-wire steering system is a vehicle control system in which the mechanical steering system is replaced by an electronic and / or electromechanical system. In this case, the steering wheel or other steering elements are not directly coupled to the wheels via a steering spindle. Rather, steering is controlled via sensors and / or steering actuators, which convert the driver's steering motion at the steering wheel into electronic signals and transmit them to a control unit via at least one cable or without cables. The control unit then controls the wheels via electric motors and / or other drive devices. Steer-by-wire systems can enable improved precision and adaptability in steering and contribute to reducing the space required in the vehicle.

[0006] In modern steering systems, torsion bar sensors are commonly used to measure the steering angle introduced by the driver and / or operator via the steering control elements. A torsion bar sensor is a mechanical component used to measure rotational motion and / or torque and / or rotational force. A torsion bar sensor typically consists of a metal bar or rod that is deformed by rotational motion. The deformation of the bar can be measured electronically and / or mechanically, providing information about the rotational motion and / or rotational force acting on the bar.

[0007] A typical application of torsion bar sensors is force measurement in automobiles. Here, for example, torsion bar sensors can be used to measure wheel load, monitor axle load, and / or measure steering force. Torsion bar sensors are also used in other fields such as medical technology, robotics, and industrial technology.

[0008] Torsion bar sensors are, on the one hand, additional components that need to be calibrated and / or maintained, and on the other hand, additional components that do not always necessarily output accurate measurement signals. Therefore, there is a growing effort to provide information on rotational motion and / or torque and / or rotational force through alternative or supplemental methods as well. [Overview of the Initiative] [Problems that the invention aims to solve]

[0009] Therefore, the fundamental problem of the present invention is to present an improved, and in particular hardware-optimized, method and an improved, and in particular hardware-optimized apparatus. [Means for solving the problem]

[0010] The above problem is solved by a method for estimating steering torque in a vehicle steering system, as described in claim 1. Furthermore, the above problem is solved by an apparatus for estimating steering torque in a vehicle steering system, as described in claim 10.

[0011] Disclosure of the invention In a first embodiment, a computer-implemented method for estimating steering torque in a steering system of a vehicle and / or machine is presented. The method includes at least the steps of: providing a machine learning algorithm; providing at least one input signal containing information about the momentum of at least one electric motor of an operating element of the steering system; and using the machine learning algorithm to estimate, based on the provided input signal, a steering torque applied by an operator through the operating element.

[0012] It is obvious that the steps and further optional steps according to the present invention do not necessarily have to be performed in the order listed, but may be performed in a different order. Furthermore, further intermediate steps may be provided. Each step may further include one or more substeps, and this will not deviate from the scope of the method according to the present invention.

[0013] In a second embodiment, an apparatus for estimating steering torque in a steering system of a vehicle and / or machine is presented. The apparatus includes a providing device configured to provide a machine learning algorithm and at least one input signal containing information about the momentum of at least one electric motor of an operating element of the steering system, and an evaluation and calculation device configured to estimate and / or approximate the steering torque applied by an operator through the operating element based on the provided input signal by using the machine learning algorithm.

[0014] The advantage of the present invention lies in the fact that, by estimating the steering torque, the use of torsion bar sensors in the steering system in question can be essentially omitted. According to the present invention, it is possible to estimate the steering torque without necessarily having to measure it using torsion bar sensors. This makes it possible to reduce the cost of such a steering system. Furthermore, the omission of torsion bar sensors as components makes more structural space available. The method and apparatus according to the present invention have a favorable effect on the manufacturing cost of the steering system.

[0015] Essentially, the torsion bar sensor can be retained in the steering system in question and supplemented by the steering torque estimation according to the present invention. This would be preferable, for example, for validating the measurement results of the torsion bar sensor and / or for improving the measurement accuracy of the torsion bar sensor. Furthermore, by additionally and / or supplementarily estimating the steering torque, it becomes possible to install a single-channel torsion bar sensor instead of a two-channel torsion bar sensor, thereby reducing the cost of such a torsion bar sensor and the complexity of the evaluation device for the torsion bar sensor.

[0016] The electric motor is preferably a steering wheel actuator (SWA). In other words, the electric motor is preferably an operating element actuator. At least one input signal preferably includes information about at least one momentum of the operating element electric motor, which indicates, in particular, how the electric motor moves and / or behaves and / or what the current (actual) position of the electric motor's rotor is relative to the electric motor's stator. At least one input signal preferably includes multiple input data relating to the electric motor. The input data is preferably motion data of the electric motor. The electric motor may basically include a transmission, but is preferably not a steering gear of a steering system.

[0017] In the context of the present invention, a vehicle is understood to be a goods vehicle and / or a passenger car and / or a transport vehicle and / or a motor vehicle and / or a ship and / or an aircraft and / or a military vehicle and / or a crane vehicle and / or an excavation vehicle and / or a tractor and / or an endless track vehicle and / or a construction vehicle and / or a mobile hydraulic vehicle. Furthermore, the method and / or the device according to the present invention may be used, in particular, for estimating the steering torque in the steering system of industrial machines and / or tools. In other words, the method and / or the device according to the present invention may be used, for example, in other applications other than vehicles, such as for which a steering input for controlling a tool head or the like is performed. For example, the method and / or the device according to the present invention may be advantageous in the manual or semi-automatic control and / or steering of a machine by an operator and / or an automated function.

[0018] In particular, the method and / or the device according to the present invention is used in a Steer-by-Wire (SbW) type steering system. Basically, the method and / or the device according to the present invention may also be used in other steering systems.

[0019] In one embodiment, the machine learning algorithm includes an Explainable Boosting Machine (EBM) algorithm, and / or a Generalized Additive Model (GAM) algorithm, and / or a regression-based algorithm. Basically, further different machine learning algorithms, for example, a neural network and / or a gradient boosting tree algorithm and / or a k-nearest neighbor algorithm and / or a decision tree algorithm can also be used.

[0020] Regression-based algorithms are easily traceable and are particularly suitable for processing input signals and / or input data that have a linear relationship with each other. Specifically, such regression-based algorithms are configured to identify relationships between input signals and / or input data based on linear regression.

[0021] An Explainable Boosting Machine (EBM) algorithm is preferably a machine learning model configured to make predictions based on input data and / or input signals. EBM algorithms belong to a family of boosting algorithms that create stronger models by combining multiple weak learning models. Unlike many other machine learning models, EBM algorithms offer greater interpretability because they are based on decision rule structures that allow the prediction results to be displayed in a human-readable form. This means that it is easier to understand how and why the model arrived at a particular prediction. Further advantages of EBM algorithms are that they are robust to outliers and missing data and scale well when using categorical and / or numerical variables. Therefore, EBM algorithms are often used in applications that require high interpretability and accuracy.

[0022] The Generalized Additive Model (GAM) algorithm is a statistical model used to model the non-linear relationship between a dependent variable and one or more independent variables. Different from linear models that can only model linear relationships between variables, GAM enables the modeling of complex non-linear relationships. GAM is based on the assumption that the non-linear relationship between variables can be modeled by a combination of smooth functions. In this case, the smoothing functions may be, for example, splines and / or local regression. The estimation of the smooth functions is carried out using an optimization algorithm to find the best combination of smooth functions. The advantage of GAM is that it is very interpretable. This is because GAM can represent the relationship between the dependent variable and the independent variables in a graphical form. This will be very helpful when interpreting the results.

[0023] In one embodiment, by the GAM algorithm, one fitting function for a plurality of data points is specified based on a plurality of accumulated spline functions, particularly using at least one regularization term. For example, it may be a b-spline function. Each spline function preferably describes a respective shape function for each data point and / or each input signal. The regularization term enables, preferably, an approximation of at least one input signal and / or the data transition and / or signal transition of the input data. By a preferred GAM algorithm, non-linear data relationships between input signals and / or between input data can also be specified. Such a GAM algorithm is highly performant and still has a high traceability compared to other algorithms of machine learning.

[0024] Such a GAM algorithm can be, for example, in the following formula: g(E[y])=β0+f1(x1)+f2(x2)+···+f m (x m ) {Equation 1} It can be expressed as follows, where g(E[y]) is the filtered input signal function, β0 is the data offset for at least one input signal, and f m (x m ) is the m-th spline function for the m-th input signal.

[0025] In one embodiment, the EBM algorithm is built upon the GAM algorithm and extended or expanded by the amount of the data tree-based approach. The EBM algorithm is a data tree-based approach in which bootstrapping of input signals and / or input data and aggregation of at least one input signal or input data are performed. The EBM algorithm preferably includes gradient boosting. The EBM algorithm is preferably trained by round-robin training. In this case, preferably a set of limited data trees is used as the basis. Furthermore, during training, preferably the residual from the mean of each input signal is calculated. Scaling of the EBM algorithm is performed via a learning rate that should be considered very small.

[0026] Round-robin training is a machine learning technique in which multiple models are trained simultaneously by sequentially applying them to the same data set. In round-robin training, multiple models are trained by sequentially applying them to the same training data set in a rotation cycle. That is, each model is trained sequentially once before it is the turn of the next model. In this way, all models are trained uniformly, and overfitting, which occurs when models always refer to the same training data set, is avoided. After each training unit, the model weights are saved, and the next model starts training using these weights. This ensures that each model has a similar starting position and is not unexpectedly started with other weights. Round-robin training is particularly useful when it is difficult to find the best training approach for a particular problem, or when the training data is limited. By using multiple models, the risk of overfitting can be reduced and the overall output of the models can be improved.

[0027] In one embodiment, a machine learning algorithm is trained on training data relating to at least one input signal or motion data of an electric motor of an operating element of a steering system. The training data preferably includes information on multiple motion data of the electric motor. The motion data may be identified at least partially by sensing, for example, via a position sensor and / or attitude sensor of the electric motor that identifies the position of the rotor relative to the stator. The motion data may also be provided alternatively or complementaryly by the time derivative of a measured signal detected by sensing. Training of the machine learning algorithm is preferably carried out separately and / or in parallel for each piece of motion information relating to the motion of the electric motor.

[0028] In one embodiment, at least one input signal includes rotor position information, and / or rotor speed information, and / or rotor acceleration information, and / or rotor pressure information, and / or motor target torque information, and / or information regarding the gradient of the motor target torque, and / or motor actual torque information, and / or information regarding the gradient of the motor actual torque. Particularly preferably, each piece of information is detected as a time series, and thus a plurality of data points are provided over a predetermined time, and these data points can be evaluated by a machine learning algorithm. Based on the data points, preferably, a machine learning algorithm is also trained for each feature.

[0029] In one embodiment, at least one input signal is filtered, in particular, by using a moving average and / or low-pass filter. Basically, and preferably especially, other signal filters and / or data filters can also be considered to exclude data points containing outliers and / or errors from the dataset from the outset.

[0030] In one embodiment, the operating elements include a steering wheel and / or a joystick and / or a steering lever and / or a steering switch. The operating elements may vary depending on the type of vehicle and / or machine and / or steering technology. For example, the operating elements may include a steering wheel, which is indeed the most frequently used steering operating element in a vehicle, and typically includes a frame with a grip that the driver holds and rotates to steer the vehicle. Preferably, the operating elements are (a) a steering wheel of a steer-by-wire steering system, (b) a joystick, which is an alternative to a steering wheel used in most vehicles and / or steering systems, similar to an aircraft control stick, and preferably capable of enabling intuitive control, (c) a steering lever, which is a steering operating element used particularly in commercial vehicles and / or agricultural machinery, and which can be mounted to the side of the cab or located in the center of the cab, and / or (d) a steering switch that makes the vehicle and / or machine and / or steering system operable via at least one switch provided, for example, on the console and / or dashboard.

[0031] The present invention further relates to a steering system and / or vehicle control device, comprising at least one processor and / or evaluation / control device configured to at least partially carry out the method according to the present invention in any embodiment. The control device and / or evaluation / control device does not need to carry out all the steps of the method, but can carry out only some of the steps of the method. The control device may be configured as an intelligent control unit (SCU) of the steering system, which is also referred to as a steering SCU. The control device may be configured as a control unit for a steering wheel actuator (SWA) and / or an operating element actuator. The control device may be a control unit for an electromechanical power steering system, also referred to as an EPS (Electric Power Steering) system. Alternatively, the control device may be included in or comprised of a central control unit and / or central computer of an automobile and / or machine.

[0032] The present invention further relates to a steering system for a vehicle and / or machine, comprising an operating element for applying steering torque by an operator, an electric motor connected to the operating element to transmit motion and / or torque, and an apparatus and / or control device according to the present invention.

[0033] The present invention also claims a computer program comprising program code for performing at least a part of the method according to the present invention in one of its embodiments when the computer program is executed on a computer. In other words, the present invention also claims a computer program (product) comprising instructions for causing a computer to perform the steps of the method and / or the method according to the present invention in one of its embodiments when the program is executed by a computer.

[0034] The present invention also proposes a computer-readable data carrier comprising program code of a computer program for carrying out at least a portion of the method according to the present invention in one of its embodiments when the computer program is executed on a computer. In other words, the present invention relates to a computer-readable storage medium that, when executed by a computer, includes instructions for causing a computer to carry out the steps of the method and / or the method according to the present invention in one of its embodiments.

[0035] The embodiments and variations described can be combined in any way.

[0036] Further possible embodiments, developments, and implementations of the present invention include combinations of features of the present invention described above or below in relation to the examples, which are not explicitly mentioned.

[0037] The accompanying drawings are intended to facilitate a further understanding of embodiments of the present invention. The accompanying drawings illustrate embodiments and are used in connection with the explanation of the principles and concepts of the present invention.

[0038] Many other embodiments and the advantages described above will become apparent with reference to the drawings. The elements shown in the drawings are not necessarily shown to scale relative to one another. [Brief explanation of the drawing]

[0039] [Figure 1] This is a schematic flowchart of the method according to the present invention. [Figure 2] This is a schematic flowchart of an exemplary machine learning algorithm. [Figure 3A] This is a schematic diagram for comparing the measured steering torque with the steering torque estimated by the present invention. [Figure 3B] This is a schematic diagram showing the temporal progression of the torque error in steering torque estimated by the present invention.

[0040] In each drawing, unless otherwise indicated, the same reference numeral refers to the same or functionally equivalent element, component, or part. [Modes for carrying out the invention]

[0041] Figure 1 shows a schematic block diagram of a computer-implemented method and apparatus 1 for estimating steering torque in the steering system of a vehicle and / or machine.

[0042] The method may be carried out at least partially by apparatus 1 in any embodiment, for which apparatus 1 may include several components not shown in detail, e.g., one or more supplying devices and / or at least one evaluation / calculating device. It is obvious that the supplying devices may be configured together with the evaluation / calculating devices or may be configured separately from the evaluation / calculating devices. Apparatus 1 may further include a storage device and / or an output device and / or a display device and / or an input device.

[0043] According to the present invention, a computer-implemented method includes at least the following steps:

[0044] In step S1, providing the algorithm of the machine learning 100 is implemented. The algorithm of the machine learning includes, for example, an Explainable Boosting Machine (EBM) algorithm, which is schematically shown in a greatly simplified and incomplete manner in FIG. 2. Alternatively or additionally, the algorithm of the machine learning includes a Generalized Additive Model (GAM) algorithm and / or a regression-based algorithm.

[0045] In step S2, providing at least one input signal including information regarding at least one momentum of the electric motor of the operating element of the steering system is implemented. The at least one input signal includes rotor position information feat1, and / or rotor speed information feat2, and / or rotor acceleration information feat3, and / or rotor pressure information feat n-4 and / or motor target torque information feat n-3 and / or information regarding the gradient of the motor target torque feat n-2 and / or motor actual torque information feat n-1 and / or information regarding the gradient of the motor actual torque feat n and includes feat n-4 to feat n-1 which are not shown in FIG. 2 and are merely indicated by three dots (···).

[0046] In step S3, estimating the steering torque applied by the operator via the operating element, particularly based on the at least one input signal provided by the trained algorithm of the machine learning, is implemented.

[0047] In FIG. 2, the algorithm of the machine learning 100, in this embodiment the training method of the EBM algorithm, is schematically shown in a simplified manner. The basis for training is preferably n features feat1 to feat nIn the first iteration step, a small, preferably uncomplex, data tree is trained based on the first feature feat1. This data tree preferably uses only feature feat1. Boosting preferably updates the residual value of the data tree. Then, preferably, the process continues in the same manner using feature feat2. In this case, in the first iteration step, a small, preferably uncomplex, data tree is trained based on the second feature feat2. This data tree preferably uses only feature feat2. Boosting preferably updates the residual value of this data tree. Similarly, preferably, features feat3 to feat n This procedure is carried out using the following method: all features feat1 to feat n A single round-robin path is generated that passes through the data tree. Each of the generated data trees can preferably consider only one feature. In this case, the learning rate of this method is very small, so individual features feat1 to feat n The order in which they exist is preferably not important. Rather, a large number of such iteration steps are performed, for example, up to 10,000 or more. In this case, preferably, a new data tree is generated for each feature and for each iteration step.

[0048] After the completion of repetition step i, each feature feat1 to feat n For each feature, i data trees were generated, each trained based solely on that feature. Summing these data trees for each feature yields two plots, 200 and 202, for each feature. Each plot, 200 and 202, is generated by checking how each data tree estimates its respective output value for each input value of each plot.

[0049] This ultimately results in, preferably, feature feat1 or feat n n diagrams are obtained for each of the n features, and therefore the underlying data tree from training is no longer needed for further processing. Preferably, each diagram, or each underlying data tree, was trained in parallel. Thus, the resulting model, or the resulting trained algorithm, of machine learning 100 will contain n diagrams for each of the n features.

[0050] Here, based on the Generalized Additive Model (GAM) algorithm, one feature function can be substituted and / or supplemented for each feature by one previously generated diagram. Since these are modified functions, the subscript p is used instead of the subscript m in Equation 1. The subscript p preferably corresponds to the subscript n used above. g(E[y])=β0+f1(x1)+f2(x2)+···+f p (x p ) {equation 2}

[0051] Therefore, these diagrams can be incorporated into the identification of the functions associated with each feature, as schematically shown in Figure 2 by the representation of Equation 2.

[0052] Figure 3A shows a schematic diagram for comparing the measured steering torque 300 with the torque 302 estimated by the present invention. The vertical axis plots time t [s]. The horizontal axis plots torque HWT [Nm]. It can be clearly seen that the change in the estimated steering torque 302 corresponds to approximately the change in the measured steering torque 300. Therefore, the estimation of the steering torque 302 according to the present invention is approximately equivalent to the measured steering torque 300.

[0053] Figure 3B shows a schematic diagram of the temporal progression of the torque error 304 of the steering torque 302 estimated by the present invention. The vertical axis plots time t [s]. The horizontal axis plots the torque error of torque HWT in [Nm]. It can be seen that the estimation of the steering torque 302 according to the present invention results in only a small torque error 304, that is, the steering torque can be estimated with considerable accuracy. In this way, since the steering torque 302 can be estimated with sufficient accuracy, it is reasonable to conclude that the use of a torsion bar sensor can be basically omitted.

Claims

1. A computer-implemented method for estimating steering torque in a steering system of a vehicle and / or machine, The aforementioned method, - Step (S1) of providing a machine learning algorithm, - Step (S2) of providing at least one input signal that includes information about the momentum of at least one electric motor of the operating element of the steering system, - Step (S3) of estimating the steering torque applied by the operator through the operating element based on the machine learning algorithm and the provided at least one input signal, Methods that include...

2. The aforementioned machine learning algorithms include an Explainable Boosting Machine (EBM) algorithm, and / or a Generalized Additive Model (GAM) algorithm, and / or a regression-based algorithm. The method according to claim 1.

3. The GAM algorithm identifies a single fitting function for multiple data points based on multiple accumulated spline functions, in particular using at least one regularization term. The method according to claim 2.

4. The aforementioned EBM algorithm is built upon the GAM algorithm and is extended by the amount of the data tree-based approach. The method according to claim 2 or 3.

5. The machine learning algorithm is trained on training data relating to at least one input signal of the electric motor of the operating element of the steering system. The method according to any one of claims 1 to 4.

6. The at least one input signal includes rotor position information, and / or rotor speed information, and / or rotor acceleration information, and / or rotor pressure information, and / or motor target torque information, and / or information regarding the gradient of the motor target torque, and / or motor actual torque information, and / or information regarding the gradient of the motor actual torque. The method according to any one of claims 1 to 5.

7. The at least one input signal is filtered, in particular, by using a moving average and / or low-pass filter. The method according to any one of claims 1 to 6.

8. The aforementioned operating elements include a steering wheel and / or a joystick and / or a steering lever and / or a steering switch. The method according to any one of claims 1 to 7.

9. A device comprising at least one processor and / or evaluation / control device configured to at least partially implement the method described in any one of claims 1 to 8, Steering system and / or vehicle control device.

10. A device (1) for estimating steering torque in the steering system of a vehicle and / or machine, The aforementioned device (1) is A providing device configured to provide a machine learning algorithm and at least one input signal containing information about the momentum of at least one electric motor of the steering system's operating element, An evaluation and calculation device configured to estimate and / or approximate the steering torque applied by the operator through the operating element based on the provided at least one input signal using the machine learning algorithm, Apparatus (1), including.

11. A steering system for a vehicle and / or machine, An operating element for applying steering torque by the operator, An electric motor connected to the aforementioned operating element, The apparatus according to claim 10 and / or the control device according to claim 9, A steering system, including the steering system.

12. A computer program comprising program code for performing at least a part of the method described in any one of claims 1 to 8 when the computer program is executed on a computer.

13. A computer-readable data carrier comprising program code of a computer program for carrying out at least a part of the method according to any one of claims 1 to 8 when the computer program is executed on a computer.

Images