VEHICLE CONTROL DEVICE

Abstract

A vehicle control device is provided comprising: a variable damper (a device for suppressing relative displacement) provided between a vehicle body and a wheel of a vehicle and configured to change a force that suppresses a relative displacement between the vehicle body and the wheel; a vehicle state calculation unit attached to the vehicle, configured to calculate a state of the vehicle;a state estimator configured to output a spring velocity or relative velocity for an input signal from the vehicle state calculation unit based on a neural network trained to learn an estimation result of the spring velocity or relative velocity for the input signal from the vehicle state calculation unit in a predetermined frequency band in which noise at a high frequency or phase shift at a low frequency is reduced; and a control value calculation unit (control unit) configured to control the variable damper based on the output of the state estimator.

Description

TECHNICAL AREA The present disclosure relates to a vehicle control device, which is preferred, for example, for use in a four-wheeled vehicle. STATE OF THE ART Patent literature 1 discloses a control device that appropriately reduces the noise of a sensor signal in order to suppress the performance degradation of the control system based on the sensor signal. The control device detects a sensor signal, which is based on the output of a sensor that acquires time-series data and contains noise, and reduces the noise contained in the sensor signal based on a recurrent neural network trained to learn a correspondence relationship between a first signal containing the noise and corresponding to the sensor signal, and a second signal indicating the first signal from which the noise has been removed. State-of-the-art documentation PATENT LITERATURE PTL 1: JP 2021-77030 A PRESENTATION OF THE INVENTION TECHNICAL PROBLEM In the prior art, a filter used for noise reduction is formed from the neural network. Consequently, the prior art control device reduces the noise of the sensor signal in a more suitable manner to prevent a decrease in the control performance based on the sensor signal. However, in the prior art, differentiation / integration processing is performed as a preliminary stage to the noise reduction, and a problem exists in this stage that one phase deviates from an ideal characteristic. One object of an embodiment of the present invention is to provide a vehicle control device that is capable of performing a suitable condition assessment. SOLUTION TO THE PROBLEM A vehicle control device according to an embodiment of the present invention comprises: a device for suppressing a relative displacement, provided between a vehicle body and a wheel of a vehicle and configured to modify a force that suppresses a relative displacement between the vehicle body and the wheel; a vehicle state calculation unit attached to the vehicle, configured to calculate a state of the vehicle;a state estimator configured to output a spring velocity or relative velocity for an input signal from the vehicle state calculation unit based on a neural network trained to learn an estimation result of the spring velocity or relative velocity for the input signal from the vehicle state calculation unit in a predetermined frequency band in which noise at a high frequency or phase shift at a low frequency is reduced; and a control unit configured to control the device for suppressing a relative shift based on the output of the state estimator. According to one embodiment of the present invention, a suitable state estimation can be carried out. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a block diagram illustrating a vehicle control device according to a first embodiment of the present invention. Fig. 2 is a block diagram illustrating a state estimation unit from Fig. 1. Fig. 3 is an explanatory diagram illustrating a learning method for a neural network from Fig. 2. Fig. 4 is an explanatory diagram illustrating an example of the neural network of the state estimation unit. Fig. 5 is a characteristic diagram illustrating frequency characteristics of a gain and a phase at the time of frequency axis integration of a vertical acceleration. Fig. 6 is a characteristic diagram illustrating frequency characteristics of a gain and a phase of a transfer function between the vertical acceleration and an estimated spring velocity.Figure 7 is a characteristic diagram illustrating the temporal change of the estimated sprung velocity. Figure 8 is a characteristic diagram illustrating temporal changes in spring displacement (ground displacement) and sprung jerk (ground jerk). Figure 9 is a block diagram illustrating a vehicle control device according to a second embodiment of the present invention. Figure 10 is a block diagram illustrating a state estimation unit from Figure 9. Figure 11 is an explanatory diagram illustrating a learning method for a neural network from Figure 10. Figure 12 is a block diagram illustrating a vehicle control device according to a third embodiment of the present invention. Figure 13 is a block diagram illustrating a state estimation unit from Figure 12.Figure 14 is an explanatory diagram illustrating a learning method for a neural network of Figure 13. Figure 15 is a block diagram illustrating a vehicle control device according to a fourth embodiment of the present invention. DESCRIPTION OF THE EXECUTION FORMS With reference to the attached drawings, a vehicle control device according to the embodiments of the present invention is described in detail below by way of example of its use in a four-wheeled vehicle. Fig. 1 shows a vehicle control device 1. The vehicle control device 1 has suspension devices 5 and a control unit 11. In Fig. 1, a left and a right front wheel and a left and a right rear wheel (hereinafter collectively referred to as "wheel 3") are provided on the underside of a vehicle body 2, which forms the body of a vehicle. These wheels 3 each have a tire 4. The tire 4 acts as a spring to absorb minor irregularities in a road surface. A suspension device 5 is provided such that it is arranged between the vehicle body 2 and the wheel 3 of the vehicle. The suspension device 5 comprises a suspension spring 6 (hereinafter referred to as "spring 6") and a shock absorber with adjustable damping force (hereinafter referred to as "variable damper 7"), which is arranged parallel to the spring 6 and is provided between the vehicle body 2 and the wheel 3. Fig. 1 shows a case in which one set of suspension devices 5 is provided between the vehicle body 2 and the wheel 3. However, for example, a total of four sets of suspension devices 5 are provided individually and independently between the four wheels 3 and the vehicle body 2, with only one set being shown schematically in Fig. 1. Here, the variable damper 7 of the suspension device 5 is formed by using a hydraulic shock absorber with adjustable damping force, which is arranged such that it is positioned between the vehicle body 2 and the wheel 3. The variable damper 7 forms a device for suppressing a relative displacement, which is provided between the vehicle body 2 and the wheel 3 of the vehicle and is configured to change a force that suppresses a relative displacement between the vehicle body 2 and the wheel 3. A variable damper 7 is fitted with a variable damping force actuator 8, which consists of a valve for adjusting the damping force and the like, to continuously adjust a characteristic (i.e., a damping force characteristic) of the generated damping force from a hard characteristic to a soft characteristic. It is not always necessary for the variable damping force actuator 8 to be configured to continuously adjust the damping force characteristic; it can also be configured to adjust the damping force in several stages, for example, in two or more stages. Furthermore, the variable damper 7 can be of a pressure regulator type or a flow regulator type. The variable damper 7 can also be of a type that controls viscosity, e.g., a magnetoviscous fluid or an electroviscous fluid. The control unit 11 forms a control device. The control unit 11 consists, for example, of a microcomputer as a control device that controls the damping characteristics of the variable damper 7. The control unit 11 is connected, for example, to a Controller Area Network (CAN) 10, a network of lines required for data communication. The control unit 11 receives data about the vehicle's behavior (hereinafter referred to as "behavioral information") via the CAN 10. At this point, the behavioral information includes, for example, forward and reverse acceleration (forward G and reverse G), lateral acceleration (lateral G), steering angle, yaw rate, and wheel speed. Consequently, this behavioral information is input into the control unit 11. An output of the control unit 11 is connected to the variable damping force actuator 8 of the variable damper 7. Furthermore, the control unit 11 contains a memory unit 12 consisting of a ROM, a RAM, non-volatile memory, and the like. Various programs, information (vehicle information), data, and the like for controlling the variable dampers 7 are stored in the memory unit 12 of the control unit 11. Based on the behavioral information, the control unit 11 estimates a sprung velocity and a relative velocity as vehicle state variables. At this point, the relative velocity is the relative velocity between a sprung part and an unsprung part, and the variable damper 7 is a piston velocity.The control unit 11 determines, based on the estimated vehicle condition variables, a force to be generated by the variable damper 7 (force generation mechanism) of the suspension device 5 and outputs a control signal (an electrical command current) of the force to the actuator 8 with variable damping force of the suspension device 5. As shown in Fig. 1, the control unit 11 comprises a damper control unit 13, a weight parameter storage unit 14, and a data read unit 15. The damper control unit 13 comprises a vehicle state calculation unit 16, a state estimation unit 17, and a control value calculation unit 20. The weight parameter storage unit 14 can be part of the storage unit 12 or be separate from the storage unit 12. The vehicle state calculation unit 16 is mounted on the vehicle and calculates the vehicle's state. Specifically, the vehicle state calculation unit 16 acquires data about the vehicle's state at the time of travel from communication data transmitted via the CAN bus 10 in the vehicle and calculates the behavioral information (vehicle state), including, for example, wheel speed, forward and reverse acceleration (forward G and reverse G), lateral acceleration (lateral G), steering angle, and yaw rate, based on the acquired data. It is assumed that the behavioral information calculated by the vehicle state calculation unit 16 from the CAN signal includes wheel speed, forward and reverse acceleration (forward G and reverse G), lateral acceleration (lateral G), steering angle, and yaw rate; however, the present invention is not limited to this example.The behavioral information need only be that required by the state estimation unit 17 to estimate the spring velocity or the relative velocity; unnecessary information can be omitted. Furthermore, the behavioral information is not limited to the information described above; various other types of information can be added, for example, to improve the estimation accuracy for the spring velocity or similar purposes. The state estimation unit 17 receives as input the various types of behavioral information calculated based on the CAN signal from the vehicle state calculation unit 16. The state estimation unit 17 comprises a neural network trained to estimate a spring velocity or relative velocity for the input signal from the vehicle state calculation unit 16 within a predetermined frequency band where noise at high frequencies or phase shift at low frequencies is reduced. Weight parameters of the trained neural network are stored in the weight parameter storage unit 14. Thus, the neural network of the state estimation unit 17 is formed using the weight parameters stored in the weight parameter storage unit 14.The neural network of the state estimation unit 17 outputs the spring velocity and the relative velocity for the input signal from the vehicle state calculation unit 16. In the description of this embodiment, CAN is cited as an example of a vehicle-integrated network; however, other vehicle-integrated networks can also be used. The vehicle-integrated network could, for example, be a flexible data rate CAN (CAN FD), a FlexRay, or a vehicle-integrated Ethernet. The state estimation unit 17 reads the weight parameters stored in the weight parameter storage unit 14 via the data read unit 15. The state estimation unit 17 estimates data relating to the suspension control required in the control value calculation unit 20, in particular instantaneous values ​​(sprung velocity and relative velocity) relating to the suspension control, based on the behavior information calculated by the vehicle state calculation unit 16 and the weight parameters. The control value calculation unit 20 is a control unit that controls the variable damper 7 based on an output from the state estimation unit 17. The control value calculation unit 20 calculates a suspension control value (e.g., a target damping force) to control the damping force of the variable damper 7 of the suspension device 5 based on the instantaneous values ​​input from the state estimation unit 17. Specifically, the control value calculation unit 20 calculates the suspension control value to increase the vehicle's ride comfort, for example, based on bilinear optimal control, skyhook control, or H∞ control. Based on the suspension control value, the control value calculation unit 20 outputs an electrical command current as a control signal to the actuator 8 with the variable damping force of the variable damper 7.The control value calculation unit 20 can calculate the suspension control value in order to increase not only the driving comfort of the vehicle, but also its operational stability. As shown in Fig. 2, in this embodiment, the state estimation unit 17 accesses behavioral information from data continuously transmitted to the CAN bus 10, such as wheel speed, forward and reverse acceleration, lateral acceleration, steering angle, and yaw rate, and thus estimates the data (sprung velocity and relative velocity) related to the suspension control for each of the four wheels. In a subsequent stage, the state estimation unit 17 then transmits the determined estimation results to the control value calculation unit 20. At this point, the state estimation unit 17 estimates the sensor data (vehicle state variables) related to the suspension control for each of the four wheels. The state estimation unit 17 is, for example, implemented using a neural network.The weight parameters used in the neural network are those that were determined in advance for the vehicle in question using machine learning. Thus, the state estimation unit 17 can estimate the spring velocity and the relative velocity as vehicle state variables, which also reflect the stiffness characteristics of the overall vehicle and the like. Consequently, the control value calculation unit 20 controls the variable damper 7 based on the vehicle state variables appropriately estimated by the state estimation unit 17. This makes it possible to control the suspension device 5 according to the vehicle's behavior and thus increase the accuracy of the suspension control. With reference to Fig. 2, Fig. 3, Fig. 4 to Fig. 5, a design and a learning procedure of the neural network of the state estimation unit 17 are described below. The state estimation unit 17 comprises the neural network, which represents artificial intelligence (AI) and has been trained. The weight parameters and behavioral information are input into the state estimation unit 17. The neural network of the state estimation unit 17 is formed using the weight parameters. As shown in Fig. 2, the state estimation unit 17 inputs the behavioral information into the neural network to estimate sensor data such as suspension velocity and relative velocity and output the estimated sensor data. In the next stage, the state estimation unit 17 outputs the suspension velocity and relative velocity as the vehicle state variables to the control value calculation unit 20. The neural network of the state estimation unit 17 is configured in the same way as, for example, a neural network of a vehicle behavior estimation unit disclosed in JP 2022-191913 A. As an example, a specific embodiment of the state estimation unit 17 will now be described with reference to Fig. 4. As shown in Fig. 4, the state estimation unit 17 is formed, for example, by the neural network. The neural network is formed from a hierarchical neural network with a three-layer configuration, in which elements of an input layer (number of elements: "i") 101, a hidden layer (number of elements: "j") 102, and an output layer (number of elements: "k") 103 are hierarchically connected. Each element of the input layer 101 is connected to each element of the hidden layer 102 with a weight W1ij (i=1 to I and j=1 to J). Each element of the hidden layer 102 is connected to each element of the output layer 103 with a weight W2jk (j=1 to J and k=1 to K).Information about the weight (hereinafter referred to as "weight parameters") is expressed as the determinant of a weight W1ij and a weight W2jk. The weight parameters are determined in advance by machine learning and stored in the weight parameter storage unit 14. In this example, the simplest neural network with a hidden layer 102, in which all elements are connected, has been shown, although this configuration is not limited to this example. For instance, the neural network could also be a neural network in which two or more layers form the hidden layer 102. As shown in Fig. 4, time-series data for the CAN signal of the left front wheel (wheel speed, forward and reverse acceleration, right and left acceleration, yaw rate, and the like) are input into the input layer 101 of the neural network. The instantaneous values ​​of the sprung velocity and the relative velocity of the suspension device 5, obtained assuming that the suspension device 5 is attached to, for example, the left front wheel of the vehicle, are output to the output layer 103. The number of elements in the hidden layer 102 is generally determined from the number of elements in the input layer 101 and the output layer 103, assuming that this number maximizes the accuracy of the state estimation by the neural network.The number of elements in output layer 103 is determined by the output specifications of the state estimation. Figure 3 illustrates a learning procedure for the neural network of the state estimation unit 17. For the machine learning of the neural network, data generated using the procedure shown in Figure 3 are provided as training data. First, the CAN signal (wheel speed, forward and reverse acceleration, lateral acceleration, steering angle, yaw rate, and the like) and the vertical accelerations of the four wheels are measured, which occur when the body of the actual vehicle vibrates. Then, each of the vertical accelerations is integrated with respect to the frequency axis to calculate the spring velocity and the relative velocity. The training data is generated from the calculated spring velocity and relative velocity together with the measured CAN signal. The training data is used to perform the machine learning for the neural network.During machine learning for the neural network, the weight parameters are adjusted to learn a correlation between the data on the calculated spring velocity and relative velocity and the data on the CAN signal. The weight parameters obtained as a result of the learning process are stored in the weight parameter storage unit 14. Figure 5 shows a characteristic of the frequency-axis integration. As indicated by the dotted lines in Figure 5, in a prior art real-time integration filter, to reduce the influence of the steepness described above and the like, for example, the gain at a frequency lower than a control band is reduced compared to an ideal integration characteristic. Furthermore, it is understood that, due to this influence, the phase advances near a spring resonant frequency, at which the control should ideally be performed. In contrast, with a frequency-axis integration characteristic, the control band lies within the frequency-axis integration range. In this case, the control band is a predetermined frequency band in which noise at a high frequency or phase shift at a low frequency is reduced. The control band is, for example, a band from approximately 0.8 Hz to approximately 10 Hz and includes the spring resonant frequency in the range of 1 Hz to 2 Hz. As shown by the dashed-dotted lines in Fig. 5, a comparison of the frequency-axis integration characteristic with that of the integration filter confirms that the phase shift is absent, while the gain is significantly reduced in regions outside the control band.By using the neural network, which has been trained to learn such training data, driving comfort can be achieved both in situations such as inclines where the sensor tilts, and on level roads. The neural network of the state estimation unit 17 uses as input a measurement result corresponding to a fixed past period in order to increase the estimation accuracy for the time series data. At this point, by changing a time width of the fixed past period according to a frequency (cutoff frequency of the frequency axis integration) of the control target, computing resources and performance can be simultaneously gained. In this embodiment, it is assumed that the state estimator 17 learns the correlation between the CAN signal and the data on the suspension velocity and relative velocity, obtained by using a vehicle for data acquisition, through machine learning. However, the present invention is not limited to this example. For instance, a vehicle model can be created according to the vehicle to which the state estimator 17 is attached, and the state estimator 17 can perform machine learning based on data acquired through a simulation using this vehicle model. With reference to Fig. 6, Fig. 7 to Fig. 8, a specific example of a state estimation result determined by the state estimation unit 17 and an effect of the increase in driving comfort by the vehicle control device 1 equipped with the state estimation unit 17 is described below. First, with reference to Fig. 6, the state estimation result obtained using the neural network of the state estimation unit 17 is described. The solid lines in Fig. 6 show frequency characteristics of a gain and a phase of a transfer function between the spring velocity, estimated using the neural network (state estimation unit 17) in a first embodiment of the present invention, and the vertical acceleration. At this point, the neural network of the state estimation unit 17 was trained to learn the spring velocity calculated by frequency axis integration as training data. The dashed-dotted lines in Fig. 6, on the other hand, show frequency characteristics of the gain and phase of the transfer function between the spring velocity, estimated using the prior art neural network, and the vertical acceleration. At this point, the prior art neural network was trained to learn the spring velocity obtained through integration filter processing as its training data. As shown in Fig. 6, the prior art neural network, like the integration filter characteristic, exhibits a phase lag of -90° (ideal integration characteristic) near the spring resonant frequency. In contrast, the neural network of the state estimation unit 17 was trained to learn the spring velocity calculated by frequency-axis integration, thus confirming that the phase does not lag near the spring resonant frequency. Next, with reference to Fig. 7, a state estimation result is described that is obtained when the neural network of the state estimation unit 17 is used while driving on an inclined road surface. The solid line in Fig. 7 shows a time change in the estimated spring velocity obtained by the neural network (state estimation unit 17) of the present invention. The dashed-dotted line in Fig. 7 shows a time change in the estimated spring velocity obtained by the neural network in the prior art. As shown in Fig. 7, the prior art neural network was trained to learn the spring velocity obtained by integration filter processing as training data. In the prior art, the estimation result changes considerably at a time when the inclination changes. In contrast, the neural network (state estimation unit 17) of the present invention was trained to learn the spring velocity calculated by frequency axis integration. Thus, it can be confirmed with the neural network of the present invention that even at a time when the inclination changes, no significant change in the estimation value occurs. Next, with reference to Fig. 8, the effect of the increased ride comfort achieved when controlling the vehicle using the spring velocity estimated by the state estimator 17 is described. The solid lines in Fig. 8 show temporal changes in spring displacement (ground displacement) and spring jerk (ground jerk) that occur when the estimated spring velocity obtained from the neural network of the state estimator 17 is used to perform suspension control. The dashed-dotted lines in Fig. 8 show temporal changes in spring displacement and jerk that occur when the estimated spring velocity obtained from the prior art neural network is used to perform suspension control. As shown in Fig. 8, in the prior art, even a road surface generating a low-frequency input resulted in a strong jerk. In contrast, in the neural network of the present invention, the gain at high frequency is significantly reduced by frequency axis integration. Thus, the control system implemented using the present invention makes it possible to reduce unnecessary high-frequency control caused by noise and the like by reducing the gain at high frequency. Consequently, it can be confirmed that the jerk is reduced in the control system implemented using the present invention. Thus, according to this embodiment, the vehicle control device 1 comprises the following: the variable damper 7 (device for suppressing a relative displacement), which is provided between the vehicle body 2 and the wheel 3 of the vehicle and is configured to change the force that suppresses the relative displacement between the vehicle body 2 and the wheel 3; the vehicle state calculation unit 16 attached to the vehicle, which is configured to calculate the state of the vehicle (behavioral information);the state estimation unit 17, which is configured to output the spring velocity or the relative velocity for the input signal from the vehicle state calculation unit 16 based on the neural network trained to learn the estimation result of the spring velocity or the relative velocity for the input signal from the vehicle state calculation unit 16 in the predetermined frequency band in which noise at a high frequency or the phase shift at a low frequency is reduced; and the control value calculation unit 20 (control unit), which is configured to control the variable damper 7 based on the output of the state estimation unit 17. At this point, the state estimation unit 17 can perform a suitable state estimation in situations where the sensor is tilted, e.g., on an inclined road surface (a sloping road, an embankment, etc.). This allows the vehicle control unit 1 to reduce unwanted control caused by an error in the state estimation and thus increase the vehicle's ride comfort. Furthermore, the phase shifts of the spring velocity and the relative velocity near the spring resonant frequency, as estimated by the state estimation unit 17, are reduced. Therefore, the vehicle control unit 1 can increase ride comfort not only on inclined road surfaces, etc., but also on level roads, etc. Figures 9, 10 to 11 next show a second embodiment of the present invention. A feature of the second embodiment is that the vehicle state calculation unit comprises a first sensing unit, which is mounted on the vehicle to use the CAN signal, and a second sensing unit, which is different from the first sensing unit. In the second embodiment, the components that are the same as in the first embodiment mentioned above are designated with the same reference numerals, and their descriptions are omitted. Fig. 9 shows a vehicle control device 31 according to the second embodiment. The vehicle control device 31 consists of the suspension device 5 and a control unit 32. The control unit 32 of the second embodiment forms a control device. The control unit 32 consists, for example, of a microcomputer as a control device that controls the damping characteristics of the variable damper 7. The control unit 32 of the second embodiment is configured in the same way as the control unit 11 of the first embodiment. The control unit 32 acquires behavioral information from the CAN bus 10 and the vertical acceleration sensor 30. At this point, the behavioral information includes, for example, the forward and reverse acceleration (forward G and reverse G), the lateral acceleration (lateral G), the steering angle, the yaw rate, the wheel speed, and the vertical acceleration. Consequently, this behavioral information is input into the control unit 32. An output of the control unit 32 is connected to the actuator 8 with variable damping force of the variable damper 7. The control unit 32 contains a memory unit 33 consisting of a ROM, a RAM, non-volatile memory, and the like. Various programs, information (vehicle information), data, and the like for controlling the variable damper 7 are stored in the memory unit 33 of the control unit 32. Based on the behavioral information, the control unit 32 estimates the sprung velocity and the relative velocity as vehicle state variables. Based on the estimated vehicle state variables, the control unit 32 determines a force to be generated by the variable damper 7 (force generation mechanism) of the suspension device 5 and outputs a corresponding control signal (an electrical command current) to the actuator 8 with variable damping force of the suspension device 5. As shown in Fig. 9, the control unit 32 comprises a damper control unit 34, the weight parameter storage unit 14, and the data reading unit 15. The damper control unit 34 comprises a vehicle state calculation unit 35, a state estimation unit 36, and the control value calculation unit 20. The vehicle state calculation unit 35 comprises a first vehicle-mounted acquisition unit 35A, which uses the CAN signal, and a second acquisition unit 35B, which is distinct from the first acquisition unit 35A. The first acquisition unit 35A acquires data about the vehicle state at the time of travel from communication data transmitted via CAN 10 in the vehicle and calculates, for example, the forward and reverse acceleration (forward G and reverse G), the lateral acceleration (lateral G), the steering angle, the yaw rate, and the wheel speed based on the acquired data. The second acquisition unit 35B calculates the vertical acceleration of the sprung part (vehicle body 2) from the vertical acceleration sensor 30.The vehicle state calculation unit 35 outputs the behavioral information about the vehicle to the state estimation unit 36, including the information (forward and reverse acceleration, lateral acceleration, steering angle, yaw rate, wheel speed and the like) that is captured from the CAN signal, as well as the information (vertical acceleration of the sprung part) that is captured by the vertical acceleration sensor 30. As shown in Fig. 10, the state estimation unit 36 ​​receives as input the various types of behavioral information calculated based on the CAN signal from the vehicle state calculation unit 35, and as input the acquisition signal from the vertical acceleration sensor 30. The state estimation unit 36 ​​comprises a neural network trained to learn the estimation result of a spring velocity or a relative velocity for an input signal from the vehicle state calculation unit 35 in a predetermined frequency band in which noise at a high frequency or phase shift at a low frequency is reduced. The weight parameters of the trained neural network are stored in the weight parameter storage unit 14. Thus, the neural network of the state estimation unit 36 ​​is formed by using the weight parameters stored in the weight parameter storage unit 14.The neural network of the state estimation unit 36 ​​outputs the spring velocity and the relative velocity for the input signal from the vehicle state calculation unit 35. The neural network of the state estimator 36 is configured in the same way as the neural network of the state estimator 17 of the first embodiment. The data described below are provided as training data for the machine learning of the neural network. Figure 11 illustrates a learning procedure for the neural network of the state estimator 36. During machine learning for the neural network of the state estimator 36, the CAN signal (wheel speed, forward and reverse acceleration, lateral acceleration, steering angle, yaw rate, and the like) is first measured at the time of vibration of the vehicle body of the actual vehicle, and the vertical acceleration of each of the four wheels is measured using the vertical acceleration sensor 30.The vertical acceleration is then integrated with respect to the frequency axis to calculate the spring velocity and the relative velocity. Training data is generated from the calculated spring velocity and relative velocity, together with the measured CAN signal and the vertical acceleration measured by the vertical acceleration sensor 30. This training data is used to perform machine learning for the neural network. During machine learning for the neural network, the weight parameters are adjusted to learn a correlation between the data on the calculated spring velocity and relative velocity, and the data on the CAN signal and the vertical acceleration. The weight parameters obtained as a learning result are stored in the weight parameter storage unit 14. Thus, essentially the same effects and results can be achieved in the second embodiment as in the first embodiment. In the second embodiment, the vehicle state calculation unit 35 comprises the first acquisition unit 35A mounted on the vehicle, which uses the CAN signal, and the second acquisition unit 35B, which differs from the first acquisition unit 35A.At this point, the state estimation unit 36 ​​learns, during the learning process for the neural network, the relationship between the input signal (the wheel speed, the forward G and the reverse G, the lateral G, the steering angle, the yaw rate, and the like) from the first sensing unit 35A and the sprung speed and the relative speed, which are calculated on the basis of the vertical acceleration from the second sensing unit 35B, and calculates a future sprung speed and relative speed from the vertical acceleration from the vehicle state estimation unit 35. In the state estimation unit 17 of the first embodiment, the learning is performed, while the CAN signal (wheel speed, forward and reverse acceleration, lateral acceleration, steering angle, yaw rate, and the like), which is commonly used in sensorless semi-active damping systems, is used as the input signal for the neural network. In contrast, the second embodiment is applied to the system with sensors, including the vertical acceleration sensor 30. Again, no differentiation and integration processing is used, and thus, in the state estimation unit 36 ​​of the second embodiment, as in the state estimation unit 17 of the first embodiment, the phase shifts of the spring velocity and the relative velocity near the spring resonant frequency are reduced.Furthermore, the condition estimation unit 36 ​​can perform a suitable condition estimation in situations where the sensors are tilted, e.g., on an inclined road surface (a sloping road, an embankment, etc.). Thus, according to the second embodiment, the vehicle control device 31 can also increase driving comfort on inclined road surfaces, etc., as well as on level roads, etc. In the second embodiment, it is assumed that the vehicle state calculation unit 35 calculates the data (vertical acceleration) required for estimating the suspension velocity and the relative velocity based on the detection signal from the vertical acceleration sensor 30, although the present invention is not limited to this. The vehicle state calculation unit can, for example, calculate the data required for estimating the suspension velocity and the relative velocity based on a detection signal from a vehicle height sensor. Furthermore, the vehicle state calculation unit can calculate the data required for estimating the suspension velocity and the relative velocity based on the detection signals from both the vertical acceleration sensor and the vehicle height sensor. In the second embodiment, it is assumed that the vehicle state calculation unit 35 calculates behavioral information from the CAN signal in addition to the acquisition signal from the vertical acceleration sensor 30, although the present invention is not limited to this. If the state estimation unit 36 ​​estimates the spring velocity and the relative velocity based on the acquisition signal of the vertical acceleration sensor 30, only the acquisition signal of the vertical acceleration sensor 30 needs to be input into the vehicle state calculation unit 35, and the input of the CAN signal can be omitted. Figures 12, 13 to 14 next show a third embodiment of the present invention. A feature of the third embodiment is that the state estimation unit predicts a future input signal from the input signal of the vehicle state calculation unit and outputs the spring velocity or the relative velocity for the future input signal in order to perform feedforward control. In the third embodiment, the components, which are the same as in the first embodiment mentioned above, are designated with the same reference numerals, and their descriptions are omitted. A camera 40 forms a unit for measuring the condition of the road surface, which is mounted on a front part of the vehicle body 2. The camera 40 measures and records the condition of the road surface (in particular, a distance and an angle to a road surface to be recorded, as well as the position and distance of a screen) in front of the vehicle. The camera 40 consists, for example, of a stereo camera. The camera 40 takes a pair of images from the left and right in order to record the condition of the road surface, including the distance and angle to an object (the road surface in front of the vehicle) to be imaged. Thus, a preview image in front of the vehicle captured by the camera 40 (i.e., preview information regarding the road surface) is output to the control unit 42 as a recording result obtained from the unit for measuring the condition of the road surface.The unit for measuring the condition of the road surface is not limited to the camera 40 consisting of the stereo camera, but can, for example, consist of a combination of a millimeter wave radar and a monoaural camera, a multitude of millimeter wave radars or the like. Fig. 12 shows a vehicle control device 41 according to the third embodiment. The vehicle control device 41 consists of the suspension device 5 and a control unit 42. The control unit 42 of the third embodiment forms the control device. The control unit 42 consists, for example, of a microcomputer as a control device that controls the damping characteristics of the variable damper 7. The control unit 42 of the third embodiment is configured in the same way as the control unit 11 of the first embodiment. The control unit 42 acquires behavioral information from the CAN bus 10 and from the camera 40. At this point, the behavioral information includes, for example, the forward and reverse acceleration (forward G and reverse G), the lateral acceleration (lateral G), the steering angle, the yaw rate, the wheel speed, and the road surface information. Consequently, this behavioral information is input into the control unit 42. An output of the control unit 42 is connected to the actuator 8 with variable damping force of the variable damper 7. The control unit 42 contains a memory unit 43 consisting of a ROM, a RAM, non-volatile memory, and the like. Various programs, information (vehicle information), data, and the like for controlling the variable dampers 7 are stored in the memory unit 43 of the control unit 42. Based on the behavioral information, the control unit 42 estimates the sprung velocity and the relative velocity as vehicle state variables. Based on the estimated vehicle state variables, the control unit 42 determines a force to be generated by the variable damper 7 (force generation mechanism) of the suspension device 5 and outputs a corresponding control signal (an electrical command current) to the actuator 8 with variable damping force of the suspension device 5. As shown in Fig. 12, the control unit 42 comprises a damper control unit 44, the weight parameter storage unit 14, and the data reading unit 15. The damper control unit 44 comprises a vehicle state calculation unit 45, a state estimation unit 46, and the control value calculation unit 20. The vehicle state calculation unit 45 comprises a first acquisition unit 45A mounted on the vehicle, which uses the CAN signal, and a second acquisition unit 45B, which is distinct from the first acquisition unit 45A. The first acquisition unit 45A acquires data about the vehicle state at the time of travel from communication data transmitted via CAN 10 in the vehicle and calculates, for example, the forward and reverse acceleration (forward G and reverse G), the lateral acceleration (lateral G), the steering angle, the yaw rate, and the wheel speed based on the acquired data. The second acquisition unit 45B calculates a road surface displacement in front of the vehicle (in the direction of travel ahead of the vehicle) from the camera 40.The vehicle condition calculation unit 45 outputs behavioral information about the vehicle to the condition estimation unit 46, including information (forward and reverse acceleration, lateral acceleration, steering angle, yaw rate, wheel speed, and the like) obtained from the CAN signal, as well as information (road surface displacement) obtained from the camera 40. As shown in Fig. 13, the various types of behavioral information calculated based on the CAN signal and the camera 40 acquisition signal are input from the vehicle state calculation unit 45 to the state estimation unit 46. The state estimation unit 46 comprises a neural network trained to estimate the spring velocity or relative velocity for the input signal from the vehicle state calculation unit 45 within a predetermined frequency band where noise at high frequencies or phase shift at low frequencies is reduced. The weight parameters of the trained neural network are stored in the weight parameter storage unit 14. Thus, the neural network of the state estimation unit 46 is formed using the weight parameters stored in the weight parameter storage unit 14.The neural network of the state estimation unit 46 outputs the spring velocity and the relative velocity for the input signal from the vehicle state calculation unit 45. The neural network of the state estimator 46 is configured in the same way as the neural network of the state estimator 17 of the first embodiment. However, the data described below are provided as training data for the machine learning of the neural network. Figure 14 illustrates a learning procedure for the neural network of the state estimator 46. During machine learning for the neural network of the state estimator 46, the CAN signal (wheel speed, forward and reverse acceleration, lateral acceleration, steering angle, yaw rate, and the like) is first measured at the time of vibration of the vehicle body of the actual vehicle, and the various sensors are used to measure the road surface displacement and vertical acceleration.The vertical acceleration at the moment the vehicle reaches a position corresponding to the measured road surface displacement is then integrated with respect to the frequency axis to calculate the sprung velocity and the relative velocity. The training data is generated from the calculated sprung and relative velocities, together with the measured CAN signal and the road surface displacement measured by camera 40. This training data is used to perform machine learning for the neural network. During machine learning for the neural network, the weight parameters are adjusted to learn the correlation between the data on the calculated sprung and relative velocities and the data on the CAN signal and the road surface displacement. The weight parameters obtained as a learning result are stored in the weight parameter storage unit 14. Thus, in the third embodiment, essentially the same effects and results can be achieved as in the first embodiment. In the third embodiment, the state estimation unit 46 predicts a future input signal from the input signal of the vehicle state calculation unit 45 and outputs the sprung velocity or the relative velocity for the future input signal. The vehicle control device 41 consequently performs the feedforward control. In the first and second embodiments, the state variables (such as wheel speed and vertical acceleration) of the vehicle body are used as input for the neural network. However, the input to the neural network of the present invention can be any input, as long as there are enough input signals with which the sprung and relative velocity to be output can be calculated by mathematical expressions.Thus, the present invention, as described in the third embodiment, is also applicable to a preview control system that controls the suspension based on the displacement of the road surface that the vehicle has not actually traversed. In the third embodiment, a case was taken as an example in which the unit for measuring the condition of the road surface is the camera 40, consisting of the stereo camera. The present invention is not limited to this example, and the unit for measuring the condition of the road surface can, for example, be a LiDAR (Light Detection and Ranging) camera, a combination of a millimeter-wave radar and a monoaural camera, or be formed by using a plurality of millimeter-wave radars. Furthermore, the unit for measuring the condition of the road surface is not limited to the camera 40 and can, for example, acquire map information from a server based on GPS data or information from another vehicle via vehicle-to-vehicle communication. Furthermore, in the third embodiment, it is assumed that the vehicle state calculation unit 45 calculates behavioral information from the CAN signal in addition to the acquisition signal from the camera 40, although the present invention is not limited to this. For example, if the camera and the vehicle speed sensor are connected to the vehicle state calculation unit and the vehicle state calculation unit can calculate the information required to estimate the suspension velocity and the relative velocity (the road surface displacement and the vehicle speed), it is not necessary to input the CAN signal into the vehicle state calculation unit. Next, Fig. 15 shows a fourth embodiment of the present invention. A feature of the fourth embodiment is that the state estimation unit, based on current position information from a position information acquisition unit, acquires vehicle information from the position information recorded in the recording unit. In the fourth embodiment, the components, which are identical to those in the second embodiment mentioned above, are designated with the same reference numerals, and their descriptions are omitted. A GPS receiver 50 is the position information acquisition unit that acquires the vehicle's position information. The GPS receiver 50 is mounted on the vehicle body 2 and receives signals (hereinafter referred to as "GPS signals") from the satellites of the Global Positioning System (GPS). Based on the GPS signals, the GPS receiver 50 calculates the vehicle's current position information. The GPS receiver 50 outputs the current position information to a control unit 52. The position information acquisition unit is not limited to a design that uses GPS signals. The position information acquisition unit can use a technology such as a vehicle speed sensor, a gyroscope, or a map reference to estimate the vehicle's current position information. Fig. 15 shows a vehicle control device 51 according to the fourth embodiment. The vehicle control device 51 consists of the suspension device 5 and the control unit 52. The control unit 52 of the fourth embodiment forms the control device. The control unit 52 consists, for example, of a microcomputer as a control device that controls the damping characteristics of the variable damper 7. The control unit 52 of the fourth embodiment is configured in the same way as the control unit 32 of the second embodiment. The control unit 52 acquires behavioral information from the CAN bus 10 and from the vertical acceleration sensor 30. The control unit 52 is connected to the GPS receiver 50. An output of the control unit 52 is connected to the actuator 8 with variable damping force of the variable damper 7. The control unit 52 contains a memory unit 53 consisting of a ROM, a RAM, non-volatile memory, and the like. Various programs, information (vehicle information), data, and the like for controlling the variable dampers 7 are stored in the memory unit 53 of the control unit 52. Furthermore, the memory unit 53 forms a recording unit in which position information and the like are recorded. Thus, the position information from the GPS receiver 50 and the vehicle information from the vehicle state calculation unit 35, which exhibit a correlation with the position information, are recorded in the memory unit 53. Based on behavioral information, the control unit 52 estimates the suspension velocity and the relative velocity as vehicle state variables. Furthermore, based on position information from the GPS receiver 50, the control unit 52 estimates the suspension velocity and the relative velocity as vehicle state variables. Based on these estimated vehicle state variables, the control unit 52 determines the force to be generated by the variable damper 7 (force generation mechanism) of the suspension device 5 and outputs a corresponding control signal (an electrical command current) to the actuator 8 with variable damping force of the suspension device 5. As shown in Fig. 15, the control unit 52 comprises a storage unit 53, a damper control unit 54, the weight parameter storage unit 14, and the data read unit 15. The damper control unit 54 comprises a vehicle state calculation unit 35, a state estimation unit 55, and the control value calculation unit 20. The state estimation unit 55 is configured in the same way as the state estimation unit 36 ​​of the second embodiment. The state estimation unit 55 receives input signals from the vehicle state calculation unit 35, consisting of various types of behavioral information calculated based on the CAN signal and the acquisition signal from the vertical acceleration sensor 30. The state estimation unit 55 comprises a neural network trained to learn the estimation result of the spring velocity or the relative velocity for the input signal from the vehicle state calculation unit 35 in the predetermined frequency band, in which noise at a high frequency or phase shift at a low frequency is reduced. The weight parameters of the trained neural network are stored in the weight parameter storage unit 14.Thus, the neural network of the state estimation unit 55 is formed using the weight parameters stored in the weight parameter storage unit 14. The neural network of the state estimation unit 55 outputs the spring velocity and the relative velocity for the input signal from the vehicle state calculation unit 35. Furthermore, based on the current position information from the GPS receiver 50, the state estimation unit 55 acquires vehicle information from the position information recorded in the storage unit 53 (recording unit). In particular, the state estimation unit 55 identifies the position information corresponding to the current position of the vehicle from the position information recorded in the storage unit 53 and acquires the behavioral information (forward and reverse acceleration, lateral acceleration, steering angle, yaw rate, wheel speed, vertical acceleration of the sprung part, and the like) as the vehicle information corresponding to the current position.Consequently, the neural network of the state estimation unit 55 can output the spring velocity and the relative velocity for the input signal, even if the behavioral state based on the current position information is entered there. The control value calculation unit 20 calculates a suspension control value (e.g. a target damping force) for controlling the damping force of the variable damper 7 of the suspension device 5 on the basis of the instantaneous values ​​of the sprung velocity and the relative velocity entered by the state estimation unit 55. Thus, essentially the same effects and properties can be achieved in the fourth embodiment as in the first and second embodiments. In the fourth embodiment, the vehicle has the GPS receiver 50 (position information acquisition unit), which acquires the position information about the vehicle, and the storage unit 53 (recording unit), which records position information from the GPS receiver 50 and vehicle information from the vehicle state calculation unit 35, which correlates with the position information. Furthermore, based on the current position information from the GPS receiver 50, the state estimation unit 55 acquires the vertical acceleration of the sprung part and similar vehicle information from the position information recorded in the storage unit 53.Consequently, the neural network of the state estimation unit 55 can output the spring velocity and the relative velocity for the input signal, even if the behavioral state based on the current position information is input there. Thus, in the fourth embodiment, the control of the vehicle by using the position information from the GPS receiver 50 can continue even if an abnormality exists in the vehicle state estimation unit 35. Furthermore, in the fourth embodiment, the vehicle is controlled by using the position information from the GPS receiver 50, making it possible to implement forward-feedback control, in which the vehicle state is predicted in advance. In the fourth embodiment, a case was taken as an example in which vehicle control using position information is applied to the second embodiment, although the present invention is not limited to this case. Vehicle control using position information can be applied to the first and third embodiments. In the fourth embodiment, it is assumed that the position information and the vehicle information are stored in the storage unit 53 of the control unit 52, although the present invention is not limited to this configuration. The position information and the vehicle information can be stored in a cloud, an external database, or the like, and downloaded from the cloud or the like via a communication network as needed. In the description of the first embodiment, a case was taken as an example in which the control unit 11 acquires behavioral information about the vehicle, including wheel speed, via the CAN bus 10. However, the present invention is not limited to this example. The control unit 11 can, for instance, acquire the values ​​from various sensors directly from the sensors themselves. Furthermore, the control unit 11 can acquire behavioral information from another control unit or the like. This configuration is applicable to the second to fourth embodiments. Each embodiment has been described by way of example using a case in which the force-generating mechanism is the variable damper 7 formed from the semi-active damper. The present invention is not limited to this case, and an active damper (an electric or hydraulic actuator) can be used as the force-generating mechanism. Each embodiment has been described by way of example using a case in which the force-generating mechanism, which generates an adjustable force between the side of the vehicle body 2 and the side of the wheel 3, is formed from the variable damper 7, which consists of a hydraulic shock absorber with adjustable damping force.The present invention is not limited to this case, and as an alternative to the hydraulic shock absorber, the force generation mechanism can consist, for example, of an air suspension, a stabilizer (a kinetic suspension), an electromagnetic suspension or the like. Each embodiment has been described by way of example using a case in which a vehicle control device is used in a four-wheeled vehicle. However, the present invention is not limited to this example and can also be applied, for example, to two-wheeled or three-wheeled vehicles, trucks, buses, or the like, which may be work vehicles or transport vehicles. The present invention is not limited to the embodiments described above and includes various further examples of modification. For example, although the embodiments described above are described in detail to clearly describe the present invention, the present invention is not necessarily limited to an embodiment that includes all the described embodiments. Furthermore, a part of the embodiment of a particular embodiment can be replaced by the embodiment of another embodiment, and the embodiment of another embodiment can also be added to the embodiment of a particular embodiment. In addition, another embodiment can be added to or deleted from any of the embodiments, or a part thereof can be replaced. The present application claims priority based on Japanese patent application No. 2023-142300, filed on September 1, 2023. All disclosed contents, including the specification, scope of claims, drawings, and abstract of Japanese patent application No. 2023-142300 filed on September 1, 2023, are incorporated herein in full by reference. REFERENCE MARK LIST 1, 31, 41, 51 Vehicle control device, 2 Vehicle body, 3 Wheel, 5 Suspension device, 7 Variable damper (device for suppressing relative displacement), 8 Variable damping actuator, 11, 32, 42, 52 Control unit, 12, 33, 43, 53 Storage unit (recording unit), 16, 35, 45 Vehicle condition calculation unit, 17, 36, 46, 55 Condition estimation unit, 20 Control value calculation unit (control unit), 35A, 45A First acquisition unit, 35B, 45B Second acquisition unit, 50 GPS receiver (position information acquisition unit) QUOTES INCLUDED IN THE DESCRIPTION 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 JP 2021-77030 A

[0003] JP 2022-191913 A

[0024] JP 2023-142300

[0083]

Claims

Vehicle control device comprising: a device for suppressing a relative displacement provided between a vehicle body and a wheel of a vehicle and configured to modify a force that suppresses a relative displacement between the vehicle body and the wheel; a vehicle state calculation unit attached to the vehicle, configured to calculate a state of the vehicle;a state estimator configured to output a spring velocity or relative velocity for an input signal from the vehicle state calculation unit based on a neural network trained to learn an estimation result of the spring velocity or relative velocity for the input signal from the vehicle state calculation unit in a predetermined frequency band in which noise at a high frequency or phase shift at a low frequency is reduced; and a control unit configured to control the device for suppressing a relative shift based on the output of the state estimator. Vehicle control device according to claim 1, wherein the state estimation unit is configured to predict a future input signal from the input signal from the vehicle state estimation unit and outputs the spring velocity or the relative velocity for the future input signal in order to perform forward feedback control. Vehicle control device according to claim 1, wherein the vehicle state calculation unit comprises: a first sensing unit attached to the vehicle, configured to use a CAN signal; and a second sensing unit, which is different from the first sensing unit. Vehicle control device according to claim 3, wherein the state estimation unit is configured to learn, during learning for the neural network, a relationship between an input signal from the first sensing unit and a spring velocity or a relative velocity calculated on the basis of a vertical acceleration or a vehicle height value of the second sensing unit, and to calculate a future spring velocity or relative velocity from a vertical acceleration or a vehicle height value from the vehicle state estimation unit. Vehicle control device according to claim 4, wherein the vehicle comprises: a position information acquisition unit configured to acquire position information about the vehicle; and a recording unit configured to record the position information from the position information acquisition unit and vehicle information from the vehicle condition estimation unit, which has a correlation with the position information, and wherein the condition estimation unit is configured to acquire the vehicle information from the position information recorded in the recording unit based on current position information from the position information acquisition unit.

Images