Steering control apparatus and method

Abstract

The present disclosure relates to a steering control device and method. Specifically, the steering control device according to the present disclosure includes a model learning unit that trains a deep learning model based on normal state RWA control data, a lightweighting unit that lightweights the deep learning model, and a control unit that uses the deep learning model as a virtual motor torque sensor and predicts a failure of the steering motor based on the difference between a virtual torque value generated by the virtual motor torque sensor and a target torque value to be generated by the steering motor, using the steering information of the vehicle as input.

Description

Technology Field

[0001] The present embodiments relate to a steering control device and method. Background Technology

[0002] Recently, vehicle steering systems have been evolving into Steer-by-Wire (SbW) structures that eliminate mechanical connections. These SbW systems may include Steering Feedback Actuators (SFAs) that provide steering feedback to the driver and Road Wheel Actuators (RWAs) that generate the vehicle's actual behavior. As mechanical connections are eliminated, ensuring control reliability and safety in SbW systems has become critical; in particular, abnormalities in the RWA can have a direct impact on vehicle behavior.

[0003] In conventional SbW systems, methods have been used to monitor the status of the RWA by estimating motor torque based on motor current or diagnosing the motor output status by adding specific sensors. However, the reliability of current-based estimation can be degraded by motor degradation, temperature changes, or parameter fluctuations, and the application of additional sensors leads to increased system complexity and costs.

[0004] Furthermore, existing fault response methods focus on fail-safe responses that switch the system to protection mode only after a clear failure, such as communication interruption or loss of output, occurs. Consequently, they have limitations in predicting gradually progressing abnormal conditions in advance, such as the degradation of RWA output. This can lead to an inconsistent steering feel resulting from reduced steering performance or potential safety issues.

[0005] Therefore, there is a demand for technology capable of reliably predicting abnormal conditions, such as the degradation of RWA output, and in particular, there is a growing need for new fault prediction technology that can simultaneously ensure physical validity and prediction reliability. The problem to be solved

[0006] Against this backdrop, the present disclosure aims to provide a steering control device and method for predicting a steering motor failure by comparing a virtual torque value generated by a deep learning model capable of virtually generating the output of the steering motor with a target torque value to be generated by the steering motor. means of solving the problem

[0007] In order to solve the aforementioned problem, in one aspect, the present disclosure provides a steering control device comprising: a model learning unit that trains a deep learning model based on normal state RWA control data; a lightweighting unit that lightweights the deep learning model; and a control unit that uses the deep learning model as a virtual motor torque sensor and predicts a failure of the steering motor based on the difference between a virtual torque value generated by the virtual motor torque sensor and a target torque value to be generated by the steering motor, using the steering information of the vehicle as input.

[0008] In another aspect, the present disclosure provides a steering control method comprising a model training step for training a deep learning model based on normal state RWA control data, a lightweighting step for lightweighting the deep learning model, and a fault prediction step for using the deep learning model as a virtual motor torque sensor and predicting a failure of the steering motor based on the difference between a virtual torque value generated by the virtual motor torque sensor with steering information of the vehicle as input and a target torque value to be generated by the steering motor. Effects of the invention

[0009] According to the present disclosure, a steering control device and method can estimate the state of a steering motor and predict failure without adding a separate torque sensor by using a deep learning model as a virtual motor torque sensor.

[0010] In addition, the present disclosure can be applied without changing the sensor and communication structure of an existing vehicle by using only vehicle steering information received through the Chassis CAN.

[0011] In addition, the present disclosure can simultaneously ensure the physical validity and reliability of virtual torque values ​​by applying a Physical Artificial Intelligence (PINN) structure that includes physical equations in the loss function and an Uncertainty Quantification (UQ) structure.

[0012] In addition, the present disclosure can perform predictive-based fault prediction rather than simple threshold comparison by estimating the degree of output degradation of a steering motor as a continuous value based on the difference between a virtual torque value and a target torque value, and by performing regression analysis on the temporal change trend of the degree of output degradation. Brief explanation of the drawing

[0013] FIG. 1 is a block diagram schematically illustrating a steering control system according to one embodiment. FIG. 2 is a block diagram for briefly explaining a steering control device according to one embodiment of the present disclosure. FIG. 3 is a diagram illustrating a physical artificial intelligence-based learning structure for training a deep learning model according to one embodiment. FIG. 4 is a diagram illustrating the principle by which a virtual motor torque sensor is learned by a physical equation-based loss function according to one embodiment. FIG. 5 is a diagram illustrating the SNN structure of a deep learning model according to one embodiment. FIG. 6 is a diagram illustrating the CAN-based information flow relationship between a steering control device and a vehicle according to an embodiment of the present invention. FIG. 7 is a drawing for explaining a control unit according to an embodiment of the present invention. FIG. 8 is a diagram for explaining the state of a steering motor by comparing the difference between a virtual torque value and a target torque value and the confidence interval thereof according to one embodiment. FIG. 9 is a diagram illustrating the estimation of the degree of output degradation of a steering motor as a continuous value and the prediction of failure using the trend of change according to an embodiment of the present invention. FIG. 10 is a block diagram of a steering control device according to another embodiment of the present disclosure. FIG. 11 is a flowchart illustrating a steering control method according to one embodiment of the present disclosure. Specific details for implementing the invention

[0014] Hereinafter, some embodiments of the present disclosure will be described in detail with reference to the exemplary drawings. In assigning reference numerals to the components of each drawing, the same components may have the same reference numeral as much as possible, even if they are shown in different drawings. Furthermore, in describing the embodiments, if it is determined that a detailed description of related known components or functions may obscure the essence of the technical concept, such detailed description may be omitted. Where terms such as "comprising," "having," or "consisting of" are used in this specification, other parts may be added unless "only" is used. Where a component is expressed in the singular, it may include a plural unless otherwise specified.

[0015] Additionally, terms such as first, second, A, B, (a), (b), etc., may be used to describe the components of the present disclosure. These terms are used merely to distinguish the components from other components, and the nature, order, sequence, or number of the components are not limited by such terms.

[0016] In describing the positional relationship of components, where it is stated that two or more components are "connected," "combined," or "joined," it should be understood that while the two or more components may be directly "connected," "combined," or "joined," they may also be "connected," "combined," or "joined" with other components "intervened." Here, the other components may be included in one or more of the two or more components that are "connected," "combined," or "joined" with one another.

[0017] In describing the temporal flow relationship regarding components, methods of operation, or methods of production, for example, when the temporal or sequential relationship is described using "after," "following," "next," or "before," it may include cases where the relationship is not continuous unless "immediately" or "directly" is used.

[0018] Meanwhile, where numerical values ​​or corresponding information regarding a component (e.g., levels, etc.) are mentioned, even without separate explicit notation, the numerical values ​​or corresponding information may be interpreted as including a range of error that may occur due to various factors (e.g., process factors, internal or external shocks, noise, etc.).

[0019] Hereinafter, the steering control device and method of the present disclosure will be described with reference to the attached drawings.

[0020] Below, an embodiment of a steering control system (1) that can perform the function of a steering control device (10) and performs the function of controlling the steering of a vehicle is described.

[0021] FIG. 1 is a block diagram schematically showing a steering control system (1) according to one embodiment.

[0022] Referring to FIG. 1, a steering control system according to one embodiment may include a steering control device (10), a Steering Feedback Actuator (SFA) (20) and a Road Wheel Actuator (RWA) (30), etc.

[0023] A steering control system (1) according to one embodiment may mean a system that controls the steering of a vehicle equipped with a steering control system (1) to change according to the steering angle of a steering wheel (21) operated by a driver.

[0024] This steering control system (1) may be a Steer-by-Wire (SbW) system that transmits power by transmitting and receiving electrical signals through wires, cables, etc., in a mechanical connection member (or linkage) between a steering input actuator and a steering output actuator. Below, the steering control system (1) is described based on an SbW system, but is not limited thereto.

[0025] The steering control system (1) can determine the final steering direction of the vehicle based on steering information input by the sensor and steering module mounted in the SFA (20) and output a command current to the RWA (30) so that the vehicle is steered in the steering direction.

[0026] SFA (20) may refer to a device that inputs steering information intended by the driver. As described above, this SFA (20) may include a steering wheel (21), a steering shaft (22), and a reaction motor (23). Additionally, although not illustrated, it may further include a steering gear that transmits the rotational force of the reaction motor (23) to the steering shaft (22).

[0027] The steering wheel (21) can rotate between the left steering lock section and the right steering lock section with the steering shaft (22) as the axis of rotation. Here, the lock section may refer to the limit point at which the steering wheel can move. The lock section may be composed of a steering damper, etc.

[0028] The reaction motor (23) can receive a control signal (or command current) from the steering control device (10) and provide feedback torque to the steering wheel (21). In one embodiment, the reaction motor (23) receives a command current from the steering control device (10), drives at a rotational speed indicated by the command current to generate feedback torque, and can transmit the feedback torque to the steering wheel (21) through a worm and a worm wheel.

[0029] SFA (20) may include a steering angle sensor that detects the steering angle of the steering wheel, a torque sensor that detects the driver's torque, and a steering angle velocity sensor that detects the steering angle velocity of the steering wheel.

[0030] The steering control device (10) receives steering information from each sensor included in the SFA (20), calculates a control value, and outputs an electrical signal indicating the control value to the RWA (30). Here, the steering information may refer to information including at least one of the steering angle, steering angle velocity, and driver torque.

[0031] Meanwhile, the steering control device (10) receives feedback on actual power information (e.g., first rack position information) output from the RWA (30), calculates a control value, and outputs an electrical signal indicating the control value to the SFA (20) to provide steering sensation to the driver.

[0032] RWA (30) may refer to a device that drives the actual vehicle to steer. This RWA (30) may include a first steering motor (31), a first rack (32), a front wheel (33), a vehicle speed sensor, a rack position sensor, etc.

[0033] And, SFA (20) and RWA (30) may further include a motor torque sensor capable of detecting the motor torque of the reaction motor (23) and the steering motor (31).

[0034] The steering motor (31) can move the rack (32) in the axial direction. Specifically, the steering motor (31) is driven by receiving a command current from the steering control device (10) and can cause the rack (32) to move linearly in the axial direction. That is, the rack (32) can move linearly between the left lock end, which is the left movement limit point, and the right lock end, which is the right movement limit point.

[0035] The rack (32) can perform linear motion by driving the steering motor (31), and through the linear motion of the rack (32), the front wheel (33) can be steered to the left or right.

[0036] Although not illustrated, the steering control system (1) may further include a clutch, etc., capable of separating or combining the SFA (20) and RWA (30). Here, the clutch may be operated by the control of the steering control device (10).

[0037] FIG. 2 is a block diagram for briefly explaining a steering control device (10) according to one embodiment of the present disclosure.

[0038] The steering control device (10) of the present disclosure may include a model learning unit (110), a lightweighting unit (120), and a control unit (130), etc.

[0039] The steering control device (10) trains a deep learning model based on normal state RWA control data, lightens the deep learning model, uses the deep learning model as a virtual motor torque sensor, and can predict a failure of the steering motor based on the difference between the virtual torque value generated by the virtual motor torque sensor and the target torque value to be generated by the steering motor using the steering information of the vehicle as input.

[0040] In one embodiment, the steering control device (10) may be an ADAS that provides information to assist in driving the vehicle or assists in the driver's vehicle control.

[0041] Here, ADAS can refer to various types of advanced driver assistance systems, and driver assistance systems may include, for example, Autonomous Emergency Braking, Smart Parking Assistance System (SPAS), Blind Spot Detection (BSD) system, Adaptive Cruise Control (ACC) system, Lane Departure Warning System (LDWS), Lane Keeping Assist System (LKAS), and Lane Change Assist System (LCAS). However, it is not limited to these.

[0042] The model training unit (110) can train a deep learning model based on normal state RWA control data.

[0043] At this time, the RWA control data in a normal state may include vehicle steering information including vehicle speed, lateral acceleration, yaw rate, rack displacement, rack speed, steering angle, and steering motor torque collected through the Chassis CAN during vehicle operation, and may be composed only of signals installed in the existing vehicle without using additional sensors.

[0044] Deep learning models can be trained using a Physical Artificial Intelligence (PINN) structure that includes a specific physical equation in the loss function, an Explainable Artificial Intelligence (XAI) structure, and a Spiking Neural Network (SNN) structure.

[0045] Specifically, the deep learning model can be trained using a Physical Artificial Intelligence (PINN) structure that includes a predetermined physical equation based on the physical model of the RWA system in the loss function, an Explainable Artificial Intelligence (XAI) structure that provides the basis for prediction or uncertainty regarding the output of the model, and a Spiking Neural Network (SNN) structure that considers inference in an embedded environment.

[0046] The model learning unit (110) can train a deep learning model such that a physical equation representing the relationship between steering motor torque, gear ratio, rack inertia and damping, and disturbance force is applied as a constraint during the learning process, thereby allowing the torque value generated by the model to be physically consistent with the rack dynamics of the RWA system.

[0047] In addition, the model learning unit (110) can generate a virtual motor torque sensor that can estimate an abnormal state due to a decrease in the output of the steering motor even when fault data is not secured in advance by training a deep learning model using only normal state control data.

[0048] In addition, the model learning unit (110) can train the deep learning model to output values ​​corresponding to steering motor torque and rack reaction force, and later the difference between the output of the virtual motor torque sensor and the target torque value generated by the ECU can be used for fault prediction in the control unit (130).

[0049] FIG. 3 is a diagram illustrating a physical artificial intelligence-based learning structure for training a deep learning model according to one embodiment.

[0050] Referring to FIG. 3, the model learning unit (110) collects vehicle speed, steering angle, and rack displacement through the Chassis CAN. , lag speed , rack acceleration A deep learning model can be trained by receiving vehicle steering information, including [specific information], as input. Here, the vehicle steering information may include sensor information including vehicle speed, steering angle, lateral acceleration, yaw rate, rack displacement, and rack speed, and control information generated for steering control.

[0051] The deep learning model trained in the model training unit (110) is based on a spiking neural network (SNN) structure, and as shown in Equation 1 below, the input virtual disturbance and uncertainty terms It can be trained to output the conditional probability distribution of.

[0052] [Mathematical Formula 1]

[0053]

[0054] Accordingly, deep learning models can generate uncertainty quantification (UQ) results that include the predicted mean and variance, rather than a single estimate.

[0055] In addition, the model learning unit (110) can train the deep learning model by including a steering motor torque estimation equation as a physical constraint in the loss function as shown in Equation 2 below, so that the output of the deep learning model is consistent with the physical behavior of the RWA system.

[0056] [Mathematical Formula 2]

[0057]

[0058] FIG. 4 is a diagram illustrating the principle by which a virtual motor torque sensor is learned by a physical equation-based loss function according to one embodiment.

[0059] Referring to FIG. 4, the model learning unit (110) is the actual steering motor torque and motor torque estimated by the deep learning model Error between, disturbance torque and estimated disturbance Error between, and disturbance estimation term A deep learning model can be trained to minimize the PINN loss function, Equation 3, which includes the size of .

[0060] [Mathematical Formula 3]

[0061]

[0062] Accordingly, the virtual motor torque value generated by the deep learning model can converge to a value that satisfies the physical equations of the RWA system, including the rack's inertia, damping, and disturbance terms, rather than simply a value with high data fit.

[0063] The lightweighting unit (120) can lightweight the deep learning model that has been trained.

[0064] In one embodiment, the lightweighting unit (120) can reconfigure the computational structure of the model to reduce the number of iterations or time steps of the computational blocks that are repeatedly performed along the time axis during the inference process of the deep learning model.

[0065] FIG. 5 is a diagram illustrating the SNN structure of a deep learning model according to one embodiment.

[0066] Referring to FIG. 5, the lightweighting unit (120) can reduce the amount of computation while maintaining the same inference result by reducing the iterative computation interval or removing unnecessary time steps according to a predetermined standard when the deep learning model has a structure that performs cumulative computation along the time axis.

[0067] Specifically, the lightweighting unit (120) can identify a section in which the amount of change in the internal state accumulated at each time step in each computational layer of the deep learning model is below a predetermined threshold, and can reconstruct the model structure to merge or omit the computation corresponding to the aforementioned section. By doing so, the number of computational steps required in the inference process can be reduced, thereby shortening the inference delay time.

[0068] Accordingly, the lightweighting unit (120) can simplify the computation path to enable real-time processing in an embedded environment while maintaining the structural characteristics of the deep learning model based on time axis iterative computation, and thereby reduce power consumption and memory usage.

[0069] The control unit (130) uses a deep learning model as a virtual motor torque sensor and can predict a failure of the steering motor based on the difference between the torque value generated by the virtual motor torque sensor and the target torque value to be generated by the steering motor, using the steering information of the vehicle as input.

[0070] The control unit (130) can receive vehicle steering information, including vehicle speed, steering angle, lateral acceleration, yaw rate, rack displacement, and rack speed, received through the Chassis CAN, as input and perform inference of the virtual motor torque sensor.

[0071] Additionally, the control unit (130) can predict the degree of output degradation or the possibility of an abnormality occurring in the steering motor by accumulating the difference between the torque value output by the virtual motor torque sensor and the target torque value over time, or by analyzing the magnitude, rate of change, or trend of the difference.

[0072] In addition, the control unit (130) can estimate the time of failure or remaining life of the steering motor by performing regression analysis or a predictive model based on the difference between torque values, and can utilize this for preventive maintenance or safety control of the steering system.

[0073] That is, the control unit (130) can perform fault prediction by utilizing the deep learning model not as a simple estimation means, but as a virtual sensor to monitor the condition of the steering motor in real time.

[0074] FIG. 6 is a diagram illustrating the CAN-based information flow relationship between a steering control device (10) and a vehicle according to one embodiment of the present invention.

[0075] Referring to FIG. 6, the control unit (130) can receive vehicle steering information including vehicle speed, steering angle, lateral acceleration, yaw rate, rack displacement, and rack speed from the vehicle via Chassis CAN. The vehicle steering information can be generated in real time during vehicle operation and transmitted to the control unit (130) via CAN communication.

[0076] The control unit (130) can perform inference using a virtual motor torque sensor included therein, using vehicle steering information received through the Chassis CAN as input. At this time, the virtual motor torque sensor can be implemented by a deep learning model, and the deep learning model may be a model that is learned in the model learning unit (110) and lightweighted in the lightweighting unit (120).

[0077] Additionally, the control unit (130) can calculate a target torque value to be generated by the steering motor based on vehicle steering information received via CAN, and can monitor the state of the steering motor or predict the possibility of failure by using the difference between the target torque value and the virtual torque value generated by the virtual motor torque sensor.

[0078] Meanwhile, the model learning unit (110) can learn a deep learning model based on vehicle steering information and steering control data collected through the Chassis CAN during vehicle operation, and the learned deep learning model can subsequently be used as a virtual motor torque sensor in the control unit (130).

[0079] Additionally, the lightweighting unit (120) can receive a deep learning model learned in the model learning unit (110) as input and reconstruct the model structure to reduce the amount of computation during the inference process, and the lightweighted deep learning model can be used in the control unit (130) to perform real-time processing of vehicle steering information and fault prediction.

[0080] In one embodiment, the model learning unit (110), the lightweighting unit (120), and the control unit (130) do not need to be implemented as physically independent hardware devices, but can be implemented as functionally separated components within a single hardware platform.

[0081] Specifically, the model learning unit (110), the lightweighting unit (120), and the control unit (130) can operate by being separated into software modules, logical processing blocks, or execution stages while sharing the same processor, memory, and system resources. Accordingly, model learning, model lightweighting, and steering control functions can be performed on a single ECU or integrated computing device.

[0082] At this time, the model learning unit (110) performs the function of learning a deep learning model based on vehicle steering information collected during vehicle operation, and the lightweighting unit (120) can perform the function of receiving the learned deep learning model as input and reconstructing the model structure to reduce the amount of computation in the inference process. In addition, the control unit (130) can use the lightweighted deep learning model as a virtual motor torque sensor to perform the function of estimating the state of the steering motor and predicting failure using vehicle steering information as input.

[0083] That is, according to one embodiment of the present invention, the model learning unit (110), the lightweighting unit (120), and the control unit (130) are functionally separated and implemented within a single hardware, so that model learning, lightweighting, and fault prediction functions can be performed integrally without the need for additional hardware.

[0084] FIG. 7 is a drawing for explaining a control unit (130) according to an embodiment of the present invention.

[0085] Referring to FIG. 7, the control unit (130) may include a Micro Controller Unit (MCU) (131) and an Application Processor (AP) (132). The MCU and AP may be included within a single control unit and may operate in conjunction with each other, and may be configured to perform different functions.

[0086] Specifically, the MCU can perform the function of processing vehicle steering information received from the vehicle via the Chassis CAN and generating a target torque value to control the steering motor. In addition, the MCU can execute control logic according to the vehicle's driving state and be responsible for safety control of the steering system.

[0087] Meanwhile, the AP can perform inference using a lightweight deep learning model with vehicle steering information transmitted from the MCU or received via CAN as input. In this case, the deep learning model can be used as a virtual motor torque sensor, and the AP can generate an estimated value corresponding to the steering motor torque through the virtual motor torque sensor.

[0088] In addition, the AP can determine the state of the steering motor or predict the likelihood of failure based on the difference between the torque value generated by the virtual motor torque sensor and the target torque value generated by the MCU. The prediction result can be transmitted to the MCU via an internal interface, and the MCU can perform steering control or safety control actions based on the prediction result.

[0089] In one embodiment, the AP included in the control unit (130) may include an internal processing configuration for receiving vehicle steering information as input and performing state estimation and fault prediction of the steering motor.

[0090] Specifically, the AP may include a signal preprocessing unit for processing vehicle steering information received via the Chassis CAN or transmitted from the MCU. The signal preprocessing unit can convert the vehicle steering information into a form suitable for input to a subsequent deep learning model by performing time alignment, filtering, normalization, or scaling on the vehicle steering information.

[0091] Additionally, the AP may include a deep learning inference unit for estimating steady-state steering motor torque based on the output of the signal preprocessing unit. The deep learning inference unit is based on a spiking neural network (SNN) structure and may utilize an SNN-UQ-PINN model that includes a physical artificial intelligence (PINN) structure that incorporates physical equations into the loss function and a UQ structure that quantifies uncertainty regarding the output. Accordingly, the deep learning inference unit can estimate a virtual torque value corresponding to vehicle steering information with physical validity.

[0092] In addition, the AP may include a Health Indicator Regression unit for quantifying the state of the steering motor based on the difference between the virtual torque value output from the deep learning inference unit and the target torque value generated by the MCU. The Health Indicator Regression unit can calculate the degree of output degradation or the degradation index of the steering motor by performing regression analysis on the magnitude of the torque difference, the trend of change, or the cumulative characteristics over time.

[0093] In addition, the AP can transmit the results calculated by the state indicator regression unit to the MCU, and the MCU can perform steering control or safety control operations based on the results.

[0094] That is, AP can perform torque estimation of the steering motor, calculation of the degree of deterioration, and failure prediction from vehicle steering information through an internal configuration including a signal preprocessing unit, an SNN-UQ-PINN-based deep learning inference unit, and a state indicator regression unit.

[0095] FIG. 8 is a diagram for explaining the state of a steering motor by comparing the difference between a virtual torque value and a target torque value and the confidence interval thereof according to one embodiment.

[0096] The upper graph of FIG. 8 illustrates an example where the steering motor is in a normal state, showing that the difference between the virtual torque value estimated by the virtual motor torque sensor and the target torque value calculated by the control unit (130) is maintained within a relatively small range, and the confidence interval for the virtual torque value is stably formed. In this case, the error between the virtual torque value and the target torque value changes gradually over time, and the width of the confidence interval can also be maintained below a certain level.

[0097] On the other hand, the lower graph is an example showing a case where the output of the steering motor is reduced or a failure occurs, in which the difference between the virtual torque value estimated by the virtual motor torque sensor and the target torque value that the steering motor must generate increases rapidly or the variability expands significantly. At this time, the width of the confidence interval for the virtual torque value may increase or show an abnormal vibration pattern, and accordingly, the control unit (130) can determine the abnormal state of the steering motor.

[0098] In addition, the control unit (130) can predict the normal state, output degradation state, or possibility of failure of the steering motor by considering not only the difference between the virtual torque value output by the virtual motor torque sensor and the target torque value, but also the temporal change trend of the difference and whether the confidence interval of the virtual torque value is expanded.

[0099] FIG. 9 is a diagram illustrating the estimation of the degree of output degradation of a steering motor as a continuous value and the prediction of failure using the trend of change according to an embodiment of the present invention.

[0100] Referring to FIG. 9, the control unit (130) can predict the failure of the steering motor by estimating the degree of output degradation of the steering motor as a continuous value based on the difference between a virtual torque value and a target torque value, and calculating the trend of change over time of the degree of output degradation through regression analysis. At this time, the degree of output degradation may be an indicator that quantitatively represents the level of deterioration of the steering motor, rather than simply whether it is normal or not.

[0101] Additionally, the control unit (130) can derive a trend of change in the degree of output reduction by accumulating or observing the degree of output reduction along the time axis. The trend of change can be calculated through regression analysis, and through this, it can be determined whether the output reduction of the steering motor progresses gradually or deteriorates rapidly.

[0102] In addition, the control unit (130) can predict the possibility of failure or the time of failure of the steering motor based on the results of regression analysis, and can utilize this for preventive maintenance or safety control of the steering system.

[0103] FIG. 10 is a block diagram of a control unit (130) according to another embodiment of the present disclosure.

[0104] In one embodiment, the control unit (130) may be implemented as an Electronic Control Unit (ECU). Referring to FIG. 10, the computer system (1000), such as the control unit (130), may include at least one element among one or more processors (1010), memory (1020), storage unit (1030), user interface input unit (1040), and user interface output unit (1050), and these may communicate with each other via a bus (1060). Additionally, the computer system (1000) may also include a network interface (1070) for connecting to a network. The processor (1010) may be a CPU or semiconductor device that executes processing instructions stored in the memory (1020) and / or storage unit (1030). The memory (1020) and storage unit (1030) may include various types of volatile / non-volatile storage media. For example, memory may include ROM (1024) and RAM (1025).

[0105] Hereinafter, a steering control method using a steering control device (10) capable of performing all of the above-described disclosures will be described.

[0106] FIG. 11 is a flowchart illustrating a steering control method according to one embodiment of the present disclosure.

[0107] Referring to FIG. 11, a steering control method according to one embodiment of the present disclosure may include a model training step (S1110) for training a deep learning model based on normal state RWA control data, a lightweighting step (S1120) for lightweighting the deep learning model, and a fault prediction step (S1130) for predicting a failure of a steering motor based on the difference between a virtual torque value generated by the virtual motor torque sensor using the steering information of the vehicle as input and a target torque value to be generated by the steering motor, wherein the deep learning model is used as a virtual motor torque sensor. Here, the deep learning model may be trained using a Physical Artificial Intelligence (PINN) structure that includes a predetermined physical equation in the loss function, an Explainable Artificial Intelligence (XAI) structure, and a Spiking Neural Network (SNN) structure. In addition, the RWA data may include a plurality of sensor information obtained through the Chassis CAN.

[0108] The lightweighting step (S1120) can reconfigure the computational structure of the model to reduce the number of iterations or time steps of computational blocks that are repeatedly performed along the time axis during the inference process of the deep learning model.

[0109] The failure prediction step (S1130) can predict the failure of the steering motor by estimating the degree of output degradation of the steering motor as a continuous value based on the difference between the virtual torque value and the target torque value, and calculating the trend of change of the degree of output degradation over time through regression analysis.

[0110] The steering control device includes a memory for storing collected data and at least one instruction, and at least one processor for executing at least one instruction. The at least one processor can generate a set of parameter variations by executing at least one instruction, generate a training data set for the set of parameter variations based on a predetermined simulation event, evaluate the validity of the training data set to select a valid training data set, compare the selected training data set with a target waveform to select training data sets having waveforms within a first error range with the target waveform, perform machine learning training on a parameter estimation model using the selected training data set as input, and estimate parameters using the parameter estimation model.

[0111] Additionally, the memory may be configured to be readable and writable by one or more processors. The memory may store various forms of data, signals, messages, information, programs, codes, instructions, and / or commands. The memory may store the AI ​​models of the present disclosure and data associated therewith. For example, it may store various AI models and their parameters, training / validation data used for model learning, and / or input and result data used for AI-based applications (natural language processing, image / video processing, multimodal processing, etc.). The memory may be composed of ROM, RAM, EPROM, flash memory, hard drive, registers, cache memory, computer-readable storage media, etc., and / or combinations thereof.

[0112] The processor may be configured to perform processing related to the AI ​​model of the present disclosure by executing at least one instruction stored in memory. The processor may perform inference on input data using the AI ​​model stored in memory and, if necessary, perform training, fine-tuning, or updating of the model. The processor may be implemented as one or more CPUs, GPUs, NPUs, or other processors and may control the overall operation of the computing device to perform the operation of the AI-based application.

[0113] According to the present disclosure, a steering control device and method can estimate the state of a steering motor and predict failure without adding a separate torque sensor by using a deep learning model as a virtual motor torque sensor.

[0114] In addition, the present disclosure can be applied without changing the sensor and communication structure of an existing vehicle by using only vehicle steering information received through the Chassis CAN.

[0115] In addition, the present disclosure can simultaneously ensure the physical validity and reliability of virtual torque values ​​by applying a Physical Artificial Intelligence (PINN) structure that includes physical equations in the loss function and an Uncertainty Quantification (UQ) structure.

[0116] In addition, the present disclosure can perform predictive-based fault prediction rather than simple threshold comparison by estimating the degree of output degradation of a steering motor as a continuous value based on the difference between a virtual torque value and a target torque value, and by performing regression analysis on the temporal change trend of the degree of output degradation.

[0117] The foregoing description is merely an illustrative explanation of the technical concept of the present disclosure, and those skilled in the art to which the present disclosure pertains may make various modifications and variations within the scope of the essential characteristics of the technical concept. Furthermore, since these embodiments are intended to explain, not limit, the scope of the technical concept is not limited by these embodiments. The scope of protection of the present disclosure shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present disclosure.

Claims

Claim 1 A steering control device comprising: a model training unit that trains a deep learning model based on normal state RWA control data; a lightweighting unit that lightweights the deep learning model; and a control unit that uses the deep learning model as a virtual motor torque sensor and predicts a failure of the steering motor based on the difference between a virtual torque value generated by the virtual motor torque sensor and a target torque value to be generated by the steering motor, using vehicle steering information as input. Claim 2 In claim 1, the deep learning model is a steering control device learned using a physical artificial intelligence (PINN) structure that includes a predetermined physical equation in the loss function, an explainable artificial intelligence (XAI) structure, and a spiking neural network (SNN) structure. Claim 3 In claim 1, the lightweighting unit is a steering control device that reconstructs the computational structure of the model to reduce the number of repetitions or time steps of a computational block that is repeatedly performed along the time axis during the inference process of the deep learning model. Claim 4 A steering control device according to claim 1, wherein the control unit estimates the degree of output degradation of the steering motor as a continuous value based on the difference between the virtual torque value and the target torque value, and calculates the trend of change of the degree of output degradation over time through regression analysis to predict failure of the steering motor. Claim 5 In claim 1, the RWA control data comprises a steering control device including a plurality of sensor information obtained through a Chassis CAN. Claim 6 A steering control method comprising: a model training step for training a deep learning model based on normal state RWA control data; a lightweighting step for lightweighting the deep learning model; and a fault prediction step for using the deep learning model as a virtual motor torque sensor and predicting a failure of the steering motor based on the difference between a virtual torque value generated by the virtual motor torque sensor and a target torque value to be generated by the steering motor, using the vehicle's steering information as input. Claim 7 In claim 6, the deep learning model is a steering control method learned using a Physical Artificial Intelligence (PINN) structure that includes a predetermined physical equation in the loss function, an Explainable Artificial Intelligence (XAI) structure, and a Spiking Neural Network (SNN) structure. Claim 8 In claim 6, the lightweighting step is a steering control method that reconstructs the computational structure of the model to reduce the number of iterations or time steps of a computational block that is repeatedly performed along the time axis during the inference process of the deep learning model. Claim 9 In claim 6, the fault prediction step is a steering control method that predicts a failure of the steering motor by estimating the degree of output degradation of the steering motor as a continuous value based on the difference between the virtual torque value and the target torque value, and calculating the trend of change of the degree of output degradation over time through regression analysis. Claim 10 In claim 6, the above RWA control data is a steering control method comprising a plurality of sensor information obtained through Chassis CAN.

Images