Method and control unit for adjusting chassis parameters during driving operation
Patent Information
- Authority / Receiving Office
- DE · DE
- Patent Type
- Patents
- Current Assignee / Owner
- DR ING H C F PORSCHE AG
- Filing Date
- 2024-09-19
- Publication Date
- 2026-07-09
AI Technical Summary
Existing methods for dynamically adjusting chassis parameters during driving are complex and inefficient, particularly in optimizing suspension settings for varying road conditions and track types.
A method utilizing a transformer-based artificial intelligence that learns from predefined driver and chassis settings to dynamically adjust chassis parameters in real-time based on current sensor data and driving information, incorporating a self-attention layer for temporal dependency training.
Enables optimal chassis settings adaptation to real-time road conditions and track types, reducing wear and improving performance by minimizing the difference between predicted and actual settings through continuous learning.
Smart Images
Figure 00000005_0000
Abstract
Description
[0001] The present invention relates to a method for the dynamic adjustment of chassis parameters, which is based on sensor data before and during driving. Furthermore, a control unit with which the method can be implemented is presented.
[0002] Race tracks, with their straight and curved sections, require different suspension settings. These can be achieved through mechanical adjustments of suspension parameters, such as wheel camber adjustment via an actuator. A camber of 0 degrees ensures optimal tire contact patch on straightaways, while negative camber provides better grip and higher power transmission in corners. Dynamic adjustment can significantly save time and reduce wear, but given the multitude of possible settings, it is very complex to optimize.
[0003] Document CN 114741781 A discloses a method for adjusting control parameters relating to the suspension of a chassis in order to make its suspension adjustable for a user. By evaluating user ratings, a user profile is created and the control parameters relating to the suspension are adjusted based on this user profile.
[0004] German patent application DE 11 2022 001 793 T5 describes a suspension control device with a parameter storage unit and a control value calculation unit. Parameters of the parameter storage unit are compared with the current behavior of the vehicle suspension, and suspension control values are calculated based on this comparison.
[0005] The publication DE 10 2019 117 228 A1 discusses the operation of chassis actuators by means of a control unit which is designed to determine sensor data relating to a road profile, to generate input values for a neural network from this data, and to determine control signals for the chassis actuators by means of the neural network.
[0006] Against this background, an object of the present invention is to provide a method for the dynamic adjustment of chassis parameters, whereby an optimal chassis setting corresponding to the road conditions is controlled at any given time. Furthermore, a control unit with which the method can be carried out is to be presented.
[0007] To solve the aforementioned problem, a method for the dynamic adjustment of chassis parameters is proposed, in which a trained artificial intelligence, based on a transformer architecture, adjusts several chassis parameters at any given time during driving based on current sensor values and driving information. The sensor values relate to the driver's current settings for operating the vehicle. The artificial intelligence is trained by • Input values are formed with several predefined scalar driving information, several predefined driver setting vectors and several predefined chassis setting vectors, wherein the driver setting vectors and the chassis setting vectors each have a number of L vector elements on a time grid of an identical first time range, • Output values are formed with several predefined chassis setting vectors, wherein the predefined chassis setting vectors each have a number of K vector elements as output values on a second time domain continued with the same time grid, • Input and output values are generated during at least one training run, which covers the first time range and the second time range on the time grid with a number of N equal to L plus K entries, and are saved as a training setting, • the transformer architecture is formed with a self-attention layer, which allows physical dependencies in the temporally correlated entries of the artificial intelligence's training specification to be trained, and • the artificial intelligence is trained based on the training specifications.
[0008] For the input values, the driver setting vectors and the chassis setting vectors each have L vector elements in the first time range T1, which is the same for these vectors. For this purpose, the first time range T1 is subdivided into a first time grid with a time interval ΔT, where T1 = ΔT * L. Accordingly, the L vector elements are formed from i = 1 to L. The time interval ΔT corresponds, for example, to a measurement interval of the vehicle's sensors.
[0009] For the initial values, the chassis setting vectors in the second time domain T2, which is identical for these vectors, each have K vector elements. For this purpose, the second time domain T2 is subdivided into a second time grid with the same time interval ΔT as in the first time grid, where T2 = ΔT * K. Accordingly, the K vector elements are formed from i = L + 1 to N = L + K.
[0010] The training parameters are derived from the scalar driving information (i.e., various individual pieces of information, e.g., regarding road surface properties) and the predefined driver setting vectors for the first time interval and the predefined chassis setting vectors, which are generated for both the first and second time intervals and each contain a number N vector elements. The second time interval, for example, immediately follows the first. It is conceivable that the training parameters are derived from multiple training runs, with input and output values being assigned to each individual training run.
[0011] Artificial intelligence, also abbreviated as AI, is trained to infer the output values of the second time period from the input values obtained in the first time period. This is possible because, during a training phase, the output values are predetermined, and the difference between the predetermined output values and the output values obtained by the AI can be minimized. In a production phase, where the trained AI controls the chassis parameters based on current sensor values and driving information, a statement is then made about a current time period, which encompasses the next current point in time. It is conceivable that, starting from the current point in time, the current time period could also include several points in time on a time grid extending into the future.
[0012] In one embodiment of the method according to the invention, the driver settings and driver setting vectors for vehicle operating settings are selected from the following list: throttle position, brake position, steering angle, gear position.
[0013] In a further embodiment of the method according to the invention, the driving information is generated from the following list: vehicle type, engine power, route type, and road surface condition. The route type is, for example, a racetrack or roads used in general traffic, such as a country road or a motorway. The road surface condition is, for example, dry, wet, or icy. This driving information is provided, for example, via the vehicle's own sensors or a weather app.
[0014] In a further embodiment of the method according to the invention, the chassis parameters and chassis setting vectors are selected from the following list: camber, caster, trailing / leading, steering scrub radius, track.
[0015] In a further embodiment of the method according to the invention, the training specification is continuously extended during driving operation by the currently made driver settings and set chassis parameters, and the artificial intelligence is trained with the respective extended training specification.
[0016] Furthermore, a control unit is claimed, comprising several sensors for measuring driver settings for operating a vehicle and for measuring road surface properties, several actuators for dynamically adjusting chassis parameters, and a computing unit with memory on which an artificial intelligence based on a transformer architecture can be and is executed. The control unit is designed, on the one hand, to train the artificial intelligence by • Input values are formed with several predefined scalar driving information, several predefined driver setting vectors and several predefined chassis setting vectors, wherein the driver setting vectors and the chassis setting vectors each have a number of L vector elements on a time grid of an identical first time range, • Output values are formed with several predefined chassis setting vectors, wherein the predefined chassis setting vectors each have a number of K vector elements as output values on a second time domain continued with the same time grid, • Input and output values are generated during at least one training run, which covers the first time range and the second time range on the time grid with a number of N equal to L plus K entries, and are saved as a training setting, • the Transformer architecture is formed with a Self-Attention layer, thereby training physical dependencies in the temporally correlated entries of the artificial intelligence's training specification, and • the artificial intelligence is trained according to the training specifications, and on the other hand, the artificial intelligence is used to adjust several chassis parameters at any time during a driving operation based on current sensor values and driving information.
[0017] In one embodiment of the control unit according to the invention, the driver settings and driver setting vectors for vehicle operating settings are selected from the following list: throttle position, brake position, steering angle, gear position.
[0018] In a further embodiment of the control unit according to the invention, the scalar driving information is formed from the following list: vehicle type, engine power, route type, road condition.
[0019] In a further embodiment of the control unit according to the invention, the chassis parameters and chassis setting vectors are selected from the following list: camber, caster, trailing / leading, steering roll radius, track.
[0020] In a further embodiment of the control unit according to the invention, it is configured to continuously expand the training specification during driving operation by means of the currently made driver settings and set chassis parameters and to train the artificial intelligence with the respective expanded training specification.
[0021] Further advantages and embodiments of the invention will become apparent from the description and the accompanying drawing.
[0022] It is understood that the features mentioned above and those to be explained below can be used not only in the combinations specified, but also in other combinations or on their own, without leaving the scope of the present invention.
[0023] Fig. Figure 1 shows an input / output scheme for artificial intelligence in an embodiment of the method according to the invention.
[0024] In Fig.Figure 1 shows an input / output scheme 10 for the artificial intelligence 14 in an embodiment of the method according to the invention. The artificial intelligence 14 receives scalar driving information 11, driver setting vectors 12, and chassis setting vectors 13 as input values. Scalar driving information 11 includes, for example, a vehicle type, engine power, a type of driving route such as a racetrack or a country road, or a road surface condition. The driver setting vectors 12 are settings made by the driver for vehicle controls such as the position of an accelerator pedal or the position of a brake pedal, or a steering angle, while the chassis setting vectors 13 relate, for example, toe, camber, caster, steering roll radius, or toe angle. The respective vectors have corresponding entries on a time grid over the same initial time range.The artificial intelligence 14 generates two chassis setting vectors 15 as input values over a second time period. The artificial intelligence is trained by minimizing the difference between predefined input values and the chassis setting input values obtained by the AI. QUOTES INCLUDED IN THE DESCRIPTION
[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature
[0000] CN 114741781 A
[0003] DE 11 2022 001 793 T5
[0004] DE 10 2019 117 228 A1
[0005]
Claims
[1] Method for dynamically adjusting chassis parameters, in which a trained artificial intelligence (14) based on a transformer architecture adjusts several chassis parameters at any time during driving operation based on current sensor values and driving information, wherein the sensor values relate to current driver settings made by the driver to operate the vehicle, and in which the artificial intelligence (14) is trained by • Input values (1) are formed with several predefined scalar driving information (11), several predefined driver setting vectors (12) and several predefined chassis setting vectors (13), wherein the driver setting vectors (12) and the chassis setting vectors (13) each have a number of L vector elements on a time grid of an identical first time domain, • Initial values (2) with several predefined chassis setting vectors (15) are formed, wherein the specified chassis setting vectors (15) each have a number of K vector elements on a second time domain continued with the same time grid as output values (2), • Input values (1) and output values (2) are generated during at least one training run, which covers the first time range and the second time range on the time grid with a number of N equal to L plus K entries, and are saved as a training setting, • the Transformer architecture is formed with a Self-Attention layer, which allows physical dependencies in the temporally correlated entries of the artificial intelligence's training specification (14) to be trained, and • the artificial intelligence (14) is trained according to the training specification. [2] Method according to claim 1, wherein the driver settings and driver setting vectors (12) for vehicle operating settings are selected from the following list: throttle position, brake position, steering angle, gear position. [3] Method according to one of the preceding claims, wherein the scalar driving information (11) is formed from the following list: vehicle type, engine power, route type, road condition. [4] Method according to one of the preceding claims, wherein the chassis parameters and chassis setting vectors (13) are selected from the following list: camber, caster, trailing / leading, steering scrub radius, toe. [5] Method according to one of the preceding claims, wherein the training specification is continuously extended during driving operation by the driver settings and chassis parameters currently being applied and the artificial intelligence (14) is trained with the respective extended training specification. [6] Control unit comprising several sensors for measuring driver settings for operating a vehicle and for measuring road surface characteristics, several actuators for dynamically adjusting chassis parameters and a computing unit having a memory on which an artificial intelligence (14) based on a transformer architecture can be executed and is executed, wherein the control unit is designed to a) to train the artificial intelligence by • Input values (1) are formed with several predefined scalar driving information (11), several predefined driver setting vectors (12) and several predefined chassis setting vectors (13), wherein the driver setting vectors (12) and the chassis setting vectors (13) each have a number of L vector elements on a time grid of an identical first time domain, • Initial values (2) are formed with several predefined chassis setting vectors (15), wherein the predefined chassis setting vectors (15) each have a number of K vector elements on a second time domain continued with the same time grid as initial values, • Input values (1) and output values (2) are generated during at least one training run, which covers the first time range and the second time range on the time grid with a number of N equal to L plus K entries, and are saved as a training setting, • the Transformer architecture is formed with a Self-Attention layer, thereby training physical dependencies in the temporally correlated entries of the artificial intelligence's training specification (14), and • the artificial intelligence (14) is trained using the training specification, b) and, using artificial intelligence (14), to adjust several chassis parameters at any time during a driving operation based on current sensor values and driving information. [7] Control unit according to claim 6, wherein the driver settings and driver setting vectors (12) for vehicle operating settings are selected from the following list: throttle position, brake position, steering angle, gear position. [8] Control unit according to one of claims 6 or 7, wherein the scalar driving information (11) is formed from the following list: vehicle type, engine power, route type, road condition. [9] Control unit according to one of claims 6 to 8, wherein the chassis parameters and chassis setting vectors (13) are selected from the following list: camber, caster, trailing / leading, steering roll radius, toe. [10] Control unit according to one of claims 6 to 9, which is configured to continuously extend the training specification during driving operation by means of the currently made driver settings and set chassis parameters and to train the artificial intelligence (14) with the respective extended training specification.