Vehicle control device

Abstract

This vehicle control device includes: a variable damper (relative-displacement suppression device) that is provided between a vehicle body and a wheel of a vehicle and changes a force for suppressing relative displacement between the vehicle body and the wheel; a vehicle-state calculation unit that is mounted on the vehicle and calculates a state of the vehicle; a state estimation unit that outputs a sprung speed or a relative speed to an input signal from the vehicle-state calculation unit on the basis of a neural network that has learned a result of estimation of the sprung speed or the relative speed in a predetermined frequency band in which a high-frequency noise reduction or a low-frequency phase-shift reduction were performed with respect to the input signal from the vehicle-state calculation unit; and a control-value calculation unit (control unit) that controls the variable damper on the basis of the output from the state estimation unit.

Description

Vehicle control device

[0001] The present disclosure relates to a vehicle control device suitable for use in, for example, a four-wheeled automobile.

[0002] Patent Literature 1 discloses a control device that appropriately reduces noise in a sensor signal and prevents degradation of control performance based on the sensor signal. This control device acquires a sensor signal that includes noise based on an output from a sensor that detects time-series data, and reduces the noise included in the sensor signal based on a recurrent neural network that is trained to learn the correspondence between a first signal that includes the noise and corresponds to the sensor signal, and a second signal that represents the first signal from which the noise has been removed.

[0003] Japanese Patent Application Laid-Open No. 2021-77030

[0004] In the prior art, the filter used for noise reduction is configured using a neural network. This allows the control device of the prior art to more appropriately reduce noise in the sensor signal and prevent a deterioration in control performance based on the sensor signal. However, in the prior art, differentiation / integration processing is performed as a preliminary step to noise reduction, and this step poses a problem in that the phase deviates from the ideal characteristics.

[0005] An object of one embodiment of the present invention is to provide a vehicle control device that is capable of appropriate state estimation.

[0006] A vehicle control device according to one embodiment of the present invention comprises a relative displacement suppression device that is provided between the body and wheels of a vehicle and that changes the force that suppresses the relative displacement between the body and the wheels; a vehicle state calculation unit that is mounted on the vehicle and calculates the state of the vehicle; a state estimation unit that outputs a sprung velocity or a relative velocity in response to an input signal from the vehicle state calculation unit based on a neural network that has learned the results of estimating a sprung velocity or a relative velocity in a predetermined frequency band in which high-frequency noise is reduced or low-frequency phase shift is reduced; and a control unit that controls the relative displacement suppression device based on the output from the state estimation unit.

[0007] According to one embodiment of the present invention, appropriate state estimation can be performed.

[0008] 10 is a block diagram showing a vehicle control device according to a first embodiment. FIG. 11 is a block diagram showing a state estimating unit in FIG. 1. FIG. 12 is an explanatory diagram showing a learning method of the neural network in FIG. 2. FIG. 13 is an explanatory diagram showing an example of a neural network of the state estimating unit. FIG. 14 is a characteristic line diagram showing the frequency characteristics of gain and phase when vertical acceleration is integrated along the frequency axis. FIG. 15 is a characteristic line diagram showing the frequency characteristics of gain and phase of a transfer function between vertical acceleration and estimated sprung velocity. FIG. 16 is a characteristic line diagram showing the time change of estimated sprung velocity. FIG. 17 is a characteristic line diagram showing the time change of sprung displacement and sprung jerk. FIG. 18 is a block diagram showing a vehicle control device according to a second embodiment. FIG. 19 is a block diagram showing a state estimating unit in FIG. 9. FIG. 10 is an explanatory diagram showing a learning method of the neural network in FIG. 11. FIG. 19 is a block diagram showing a vehicle control device according to a third embodiment. FIG. 11 is a block diagram showing a state estimating unit in FIG. 12. FIG. 19 is an explanatory diagram showing the learning method of the neural network in FIG. 13. FIG. 20 is a block diagram showing a vehicle control device according to a fourth embodiment.

[0009] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A vehicle control device according to an embodiment of the present invention will be described in detail below with reference to the accompanying drawings, taking as an example a case where the vehicle control device is applied to a four-wheeled vehicle.

[0010] Fig. 1 shows a vehicle control device 1. The vehicle control device 1 is composed of a suspension device 5 and a controller 11. In Fig. 1, for example, left and right front wheels and left and right rear wheels (hereinafter collectively referred to as wheels 3) are provided on the underside of a vehicle body 2 that constitutes the body of the vehicle. The wheels 3 are composed of tires 4, which act as springs that absorb small irregularities in the road surface.

[0011] The suspension device 5 is provided between the vehicle body 2 and the vehicle wheel 3. The suspension device 5 is composed of a suspension spring 6 (hereinafter referred to as the spring 6) and an adjustable damping force shock absorber (hereinafter referred to as the variable damper 7) provided in parallel with the spring 6 between the vehicle body 2 and the wheel 3.

[0012] 1 shows a case where one set of suspension devices 5 is provided between the vehicle body 2 and the wheels 3. However, a total of four sets of suspension devices 5 are provided individually and independently between, for example, four wheels 3 and the vehicle body 2, and only one of these sets is shown schematically in FIG.

[0013] Here, the variable damper 7 of the suspension device 5 is configured using a damping force adjustable hydraulic shock absorber interposed between the vehicle body 2 and the wheel 3. The variable damper 7 is provided between the vehicle body 2 and the wheel 3, and constitutes a relative displacement suppression device that changes the force that suppresses the relative displacement between the vehicle body 2 and the wheel 3.

[0014] The variable damper 7 is provided with a variable damping force actuator 8, which is composed of a damping force adjustment valve or the like, for continuously adjusting the characteristics of the generated damping force (i.e., the damping force characteristics) from hard characteristics (hard characteristics) to soft characteristics (soft characteristics). Note that the variable damping force actuator 8 does not necessarily have to be configured to continuously adjust the damping force characteristics, and may be capable of adjusting the damping force in multiple stages, for example, two or more stages. Furthermore, the variable damper 7 may be of a pressure control type or a flow rate control type. The variable damper 7 may also be of a type that controls viscosity, such as a magnetorheological fluid or an electrorheological fluid.

[0015] The controller 11 constitutes a control device. The controller 11 is configured by, for example, a microcomputer, and serves as a control device for controlling the damping characteristics of the variable damper 7. The controller 11 is connected to, for example, a CAN 10 (Controller Area Network), which is a line network required for data communication. The controller 11 acquires data related to the vehicle behavior (hereinafter referred to as behavior information) through the CAN 10. At this time, the behavior information includes, for example, longitudinal acceleration (longitudinal G), lateral acceleration (lateral G), steering angle, yaw rate, wheel speed, etc. Therefore, this behavior information is input to the controller 11. The output side of the controller 11 is connected to the damping force variable actuator 8 of the variable damper 7.

[0016] The controller 11 also has a storage unit 12 made up of ROM, RAM, non-volatile memory, etc. The storage unit 12 of the controller 11 stores various programs, information (vehicle information), data, etc. for controlling the variable damper 7. The controller 11 estimates the sprung velocity and relative velocity as vehicle state quantities based on the behavior information. In this case, the relative velocity is the relative velocity between the sprung and unsprung parts, and is the piston velocity of the variable damper 7. The controller 11 calculates the force to be generated by the variable damper 7 (force generating mechanism) of the suspension unit 5 based on the estimated vehicle state quantities, and outputs a control signal (command current) to the variable damping force actuator 8 of the suspension unit 5.

[0017] 1 , the controller 11 includes a damper control unit 13, a weight parameter storage unit 14, and a data reading unit 15. The damper control unit 13 includes a vehicle state calculation unit 16, a state estimation unit 17, and a control value calculation unit 20. The weight parameter storage unit 14 may be a part of the storage unit 12, or may be separate from the storage unit 12.

[0018] The vehicle state calculation unit 16 is mounted on the vehicle and calculates the state of the vehicle. Specifically, the vehicle state calculation unit 16 acquires data related to the vehicle state during driving from communication data transmitted within the vehicle via the CAN 10, and calculates behavior information (vehicle state) such as wheel speed, longitudinal acceleration (longitudinal G), lateral acceleration (lateral G), steering angle, and yaw rate based on the acquired data. Note that the behavior information calculated by the vehicle state calculation unit 16 from the CAN signal includes wheel speed, longitudinal acceleration (longitudinal G), lateral acceleration (lateral G), steering angle, and yaw rate, but the present invention is not limited to this. The behavior information may be information necessary for the state estimation unit 17 to estimate the sprung velocity or relative velocity, and unnecessary information may be omitted. Furthermore, the behavior information is not limited to the above-described information, and various information may be added, for example, to improve the estimation accuracy of the sprung velocity, etc.

[0019] The state estimation unit 17 receives various types of behavior information calculated based on the CAN signal as input signals from the vehicle state calculation unit 16. The state estimation unit 17 includes a neural network that has learned the results of estimating the sprung velocity or the relative velocity for the input signals from the vehicle state calculation unit 16 in a predetermined frequency band in which high-frequency noise or low-frequency phase shift is reduced. The weight parameters of the trained neural network are stored in the weight parameter storage unit 14. Therefore, the neural network of the state estimation unit 17 is configured using the weight parameters stored in the weight parameter storage unit 14. The neural network of the state estimation unit 17 outputs the sprung velocity and the relative velocity for the input signals from the vehicle state calculation unit 16.

[0020] In this embodiment, the in-vehicle network will be described using a CAN as an example, but other in-vehicle networks such as CAN FD (CAN with Flexible Data Rate), FlexRay, or in-vehicle Ethernet may also be used.

[0021] The state estimation unit 17 reads out the weight parameters stored in the weight parameter storage unit 14 via the data reading unit 15. Based on the behavior information and weight parameters calculated by the vehicle state calculation unit 16, the state estimation unit 17 estimates data related to suspension control required by the control value calculation unit 20, specifically, instantaneous values ​​related to suspension control (sprung velocity, relative velocity).

[0022] The control value calculation unit 20 is a control unit that controls the variable damper 7 based on the output from the state estimation unit 17. 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 based on the instantaneous value input from the state estimation unit 17. Specifically, the control value calculation unit 20 calculates a suspension control value for improving the ride comfort of the vehicle based on, for example, bilinear optimal control, skyhook control, H∞ control, etc. The control value calculation unit 20 outputs a command current as a control signal to the damping force variable actuator 8 of the variable damper 7 based on the suspension control value. Note that the control value calculation unit 20 may calculate a suspension control value for improving not only the ride comfort of the vehicle but also the handling stability.

[0023] As shown in FIG. 2 , in this embodiment, the state estimation unit 17 estimates data related to suspension control for each of the four wheels (sprung velocity and relative velocity) by referencing behavior information, such as wheel speed, longitudinal acceleration, lateral acceleration, steering angle, and yaw rate, from the data constantly transmitted to the CAN 10. The estimated results are then transferred to the downstream control value calculation unit 20. At this time, the state estimation unit 17 estimates sensor data (vehicle state quantities) related to suspension control for each of the four wheels. The state estimation unit 17 is, for example, configured with a neural network. Weight parameters used in the neural network are determined in advance for the vehicle through machine learning. Therefore, the state estimation unit 17 can estimate the sprung velocity and relative velocity as vehicle state quantities, taking into account factors such as the rigidity characteristics of the entire vehicle. As a result, the control value calculation unit 20 controls the variable damper 7 based on the vehicle state quantities appropriately estimated by the state estimation unit 17. This allows the suspension device 5 to be controlled in accordance with the vehicle behavior, thereby improving the accuracy of suspension control.

[0024] Next, the configuration and learning method of the neural network of the state estimation unit 17 will be described with reference to FIGS.

[0025] The state estimation unit 17 is an artificial intelligence (AI) and includes a trained neural network. Weight parameters and behavior information are input to the state estimation unit 17. The neural network of the state estimation unit 17 is configured using the weight parameters. As shown in FIG. 2, the state estimation unit 17 inputs behavior information to the neural network, thereby estimating sensor data such as sprung speed and relative speed, and outputs the sensor data. The state estimation unit 17 outputs the sprung speed and relative speed as vehicle state quantities to a control value calculation unit 20 in the subsequent stage. The neural network of the state estimation unit 17 is configured similarly to the neural network of the vehicle behavior estimation unit disclosed in, for example, Japanese Patent Application Laid-Open No. 2022-191913.

[0026] As an example, a specific configuration of the state estimation unit 17 will be described with reference to FIG. 4 . As shown in FIG. 4 , the state estimation unit 17 is configured, for example, as a neural network. The neural network is configured as a three-layer hierarchical neural network 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 and each element of the hidden layer 102 are connected by weights W1ij (i = 1 to I, j = 1 to J), and each element of the hidden layer 102 and each element of the output layer 103 are connected by weights W2jk (j = 1 to J, k = 1 to K). Information on these weights (hereinafter referred to as weight parameters) is expressed as a determinant of the weights W1ij and W2jk. The weight parameters are calculated in advance by machine learning and stored in the weight parameter storage unit 14. Although this example shows a neural network with a single, all-element-connected hidden layer 102, which is the simplest possible configuration, the present invention is not limited to this. For example, a neural network with two or more hidden layers 102 may also be used.

[0027] As shown in Figure 4, time-series data of, for example, the CAN signal (wheel speed, longitudinal acceleration, lateral acceleration, yaw rate, etc.) of the left front wheel is input to the input layer 101 of the neural network. For example, instantaneous values ​​of the sprung velocity and relative velocity of the suspension device 5 assumed to be attached to the left front wheel of a vehicle are output to the output layer 103. The number of elements in the hidden layer 102 is generally determined by the number of elements in the input layer 101 and the output layer 103, and is set to a number that maximizes the accuracy of state estimation by the neural network. The number of elements in the output layer 103 is determined by the output specifications of the state estimation.

[0028] FIG. 3 shows a learning method for the neural network of the state estimation unit 17. Data created using the method shown in FIG. 3 is provided as training data for the neural network machine learning. First, the CAN signals (wheel speed, longitudinal acceleration, lateral acceleration, steering angle, yaw rate, etc.) and the vertical acceleration of the four wheels are measured when the vehicle body vibrates in an actual vehicle. Then, the vertical acceleration is integrated on the frequency axis to calculate the sprung velocity and relative velocity. The calculated sprung velocity and relative velocity are combined with the measured CAN signal to create training data. This training data is used for neural network machine learning. In the neural network machine learning, weight parameters are adjusted to learn the correlation between the calculated sprung velocity and relative velocity data and the CAN signal data. The weight parameters obtained as a result of learning are stored in the weight parameter memory unit 14.

[0029] The frequency axis integral characteristic is shown in Figure 5. As shown by the dotted line in Figure 5, in the real-time integral filter used in the prior art, in order to reduce the influence of the gradient and the like described above, the gain of frequencies lower than the control band is reduced relative to the ideal integral characteristic. It can also be seen that this influence causes the phase to advance near the sprung resonance frequency where control is originally desired.

[0030] In contrast, in the characteristics of the frequency axis integral, the control band is the frequency axis integral range. In this case, the control band is a predetermined frequency band in which high-frequency noise is reduced or low-frequency phase shift is reduced. The control band is, for example, a band of approximately 0.8 to 10 Hz, and includes the sprung resonance frequency of approximately 1 to 2 Hz. As shown by the two-dot chain line in Figure 5, the characteristics of the frequency axis integral show that, compared to the characteristics of the integral filter, the gain is significantly reduced in the range outside the control band, while there is no phase shift. By using a neural network trained with such training data, it is possible to achieve both ride comfort on flat roads and situations where the sensor tilts, such as on gradients.

[0031] The neural network of the state estimation unit 17 can improve the estimation accuracy for time-series data by inputting measurement results for a certain period of time in the past. At this time, by varying the time width for the certain period of time in the past according to the frequency to be controlled (the cutoff frequency of the frequency axis integral), it becomes possible to achieve both computational resources and performance.

[0032] In the present embodiment, the state estimation unit 17 performs machine learning to learn the correlation between the CAN signal acquired by the data acquisition vehicle and the data on the sprung speed and the relative speed, but the present invention is not limited to this. For example, a vehicle model corresponding to the vehicle in which the state estimation unit 17 is installed may be constructed, and the state estimation unit 17 may perform machine learning based on data acquired by a simulation using this vehicle model.

[0033] Next, with reference to FIGS. 6 to 8, a description will be given of a specific example of the state estimation result by the state estimation unit 17 and the effect of improving ride comfort by the vehicle control device 1 including the state estimation unit 17.

[0034] First, the state estimation results using the neural network of the state estimation unit 17 will be described with reference to Fig. 6. The solid line in Fig. 6 shows the frequency characteristics of the gain and phase of the transfer function between the sprung speed and the vertical acceleration estimated by the neural network (state estimation unit 17) according to the first embodiment. At this time, the neural network of the state estimation unit 17 learns the sprung speed calculated by frequency axis integration as training data.

[0035] On the other hand, the two-dot chain line in Figure 6 shows the frequency characteristics of the gain and phase of the transfer function between the sprung speed and the vertical acceleration estimated by a neural network according to the prior art, in which the neural network learns the sprung speed of the integral filter processing as training data.

[0036] As shown in Figure 6, in the neural network of the prior art, the phase is ahead of -90° (ideal integral characteristic) near the sprung resonance frequency, as is the characteristic of the integral filter. In contrast, the neural network of state estimation unit 17 learns the sprung velocity calculated by frequency axis integration, and it can be confirmed that the phase is not ahead near the sprung resonance frequency.

[0037] Next, the state estimation results when traveling on a sloped road surface using the neural network of the state estimation unit 17 will be described with reference to Fig. 7. The solid line in Fig. 7 shows the time change of the sprung velocity estimated by the neural network (state estimation unit 17) of the present invention. The two-dot chain line in Fig. 7 shows the time change of the sprung velocity estimated by the neural network of the prior art.

[0038] As shown in Figure 7, the neural network of the prior art learns the sprung speed obtained by integral filtering as training data. Therefore, in the prior art, the estimated results change significantly when the gradient changes. In contrast, the neural network of the present invention (state estimation unit 17) learns the sprung speed calculated by frequency axis integration. Therefore, it can be confirmed that the neural network of the present invention does not significantly change the estimated value even when the gradient changes.

[0039] Next, the effect of improving ride comfort when the vehicle is controlled using the sprung velocity estimated by the state estimator 17 will be described with reference to Fig. 8. The solid line in Fig. 8 shows the time changes in sprung displacement and sprung jerk when suspension control is performed using the sprung velocity estimated by the neural network of the state estimator 17. The two-dot chain line in Fig. 8 shows the time changes in sprung displacement and jerk when suspension control is performed using the sprung velocity estimated by the neural network of the prior art.

[0040] As shown in Figure 8, in the conventional technology, large jerks occur even on road surfaces with low-frequency inputs. In contrast, the neural network of the present invention significantly reduces the high-frequency gain by frequency-axis integration. Therefore, in the control using the present invention, the high-frequency gain is reduced, thereby reducing unnecessary high-frequency control caused by noise, etc. This confirms that the control using the present invention reduces jerk.

[0041] Thus, the vehicle control device 1 according to this embodiment includes a variable damper 7 (relative displacement suppression device) that is provided between the vehicle body 2 and the wheel 3 and that changes the force that suppresses the relative displacement between the vehicle body 2 and the wheel 3; a vehicle state calculation unit 16 that is mounted on the vehicle and that calculates the state (behavior information) of the vehicle; a state estimation unit 17 that outputs the sprung velocity or relative velocity in response to the input signal from the vehicle state calculation unit 16 based on a neural network that has learned the results of estimating the sprung velocity or relative velocity in a predetermined frequency band in which high-frequency noise has been reduced or low-frequency phase shift has been reduced; and a control value calculation unit 20 (control unit) that controls the variable damper 7 based on the output from the state estimation unit 17.

[0042] In this case, the state estimation unit 17 can perform appropriate state estimation in situations where the sensor is inclined, such as on an inclined road surface (slope, bank, etc.). As a result, the vehicle control device 1 can reduce unnecessary control due to errors in state estimation, thereby improving the ride comfort of the vehicle. In addition, the phase shift between the sprung velocity and relative velocity estimated by the state estimation unit 17 near the sprung resonance frequency is reduced. Therefore, the vehicle control device 1 can improve ride comfort not only on inclined road surfaces, but also on flat roads, etc.

[0043] 9 to 11 show a second embodiment. The second embodiment is characterized in that the vehicle state calculation unit has a first detection unit that uses a CAN signal mounted on the vehicle and a second detection unit that is different from the first detection unit. In the second embodiment, the same components as those in the first embodiment are denoted by the same reference numerals, and their description will be omitted.

[0044] 9 shows a vehicle control device 31 according to the second embodiment. The vehicle control device 31 is made up of a suspension device 5 and a controller 32.

[0045] The controller 32 according to the second embodiment constitutes a control device. The controller 32 is configured, for example, by a microcomputer, as a control device that controls the damping characteristics of the variable damper 7. The controller 32 according to the second embodiment is configured similarly to the controller 11 according to the first embodiment. The controller 32 acquires behavior information data from the CAN 10 and the vertical acceleration sensor 30. At this time, the behavior information includes, for example, longitudinal acceleration (longitudinal G), lateral acceleration (lateral G), steering angle, yaw rate, wheel speed, vertical acceleration, etc. Therefore, this behavior information is input to the controller 32. The output side of the controller 32 is connected to the damping force variable actuator 8 of the variable damper 7.

[0046] The controller 32 has a storage unit 33 including a ROM, a RAM, a non-volatile memory, etc. The storage unit 33 of the controller 32 stores various programs, information (vehicle information), data, etc. for controlling the variable damper 7. The controller 32 estimates the sprung velocity and relative velocity as vehicle state quantities based on the behavior information. The controller 32 calculates the force to be generated by the variable damper 7 (force generating mechanism) of the suspension unit 5 based on the estimated vehicle state quantities, and outputs a control signal (command current) to the variable damping force actuator 8 of the suspension unit 5.

[0047] 9, the controller 32 includes a damper control unit 34, a weight parameter storage unit 14, and a data reading unit 15. The damper control unit 34 includes a vehicle state calculation unit 35, a state estimation unit 36, and a control value calculation unit 20.

[0048] The vehicle state calculation unit 35 includes a first detection unit 35A that uses a CAN signal installed in the vehicle, and a second detection unit 35B that is different from the first detection unit 35A. The first detection unit 35A acquires data related to the vehicle state while traveling from communication data transmitted within the vehicle via the CAN 10, and calculates, for example, longitudinal acceleration (longitudinal acceleration), lateral acceleration (lateral acceleration), steering angle, yaw rate, wheel speed, etc. based on the acquired data. The second detection unit 35B calculates the vertical acceleration of the sprung mass (vehicle body 2) from the vertical acceleration sensor 30. The vehicle state calculation unit 35 outputs vehicle behavior information to the state estimation unit 36, including information acquired from the CAN signal (longitudinal acceleration, lateral acceleration, steering angle, yaw rate, wheel speed, etc.) and information acquired from the vertical acceleration sensor 30 (vertical acceleration of the sprung mass).

[0049] As shown in Fig. 10 , the state estimation unit 36 ​​receives, as input signals from the vehicle state calculation unit 35, various types of behavior information calculated based on the CAN signal and the detection signal from the vertical acceleration sensor 30. The state estimation unit 36 ​​includes a neural network that has learned the results of estimating the sprung velocity or the relative velocity for the input signal from the vehicle state calculation unit 35 in a predetermined frequency band in which high-frequency noise has been reduced or low-frequency phase shift has been reduced. The weight parameters of the trained neural network are stored in the weight parameter storage unit 14. Therefore, the neural network of the state estimation unit 36 ​​is configured using the weight parameters stored in the weight parameter storage unit 14. The neural network of the state estimation unit 36 ​​outputs the sprung velocity and the relative velocity in response to the input signal from the vehicle state calculation unit 35.

[0050] The neural network of the state estimation unit 36 ​​is configured in the same manner as the neural network of the state estimation unit 17 according to the first embodiment. However, the following data is provided as training data for the machine learning of the neural network. FIG. 11 shows a training method for the neural network of the state estimation unit 36. In the machine learning of the neural network of the state estimation unit 36, first, CAN signals (wheel speed, longitudinal acceleration, lateral acceleration, steering angle, yaw rate, etc.) are measured when the body of an actual vehicle vibrates, and the vertical acceleration of each of the four wheels is measured using the vertical acceleration sensor 30. The vertical acceleration is then integrated on the frequency axis to calculate the sprung velocity and relative velocity. The calculated sprung velocity and relative velocity are combined with the measured CAN signals and the vertical acceleration measured by the vertical acceleration sensor 30 to create training data. This training data is used to perform machine learning of the neural network. In the machine learning of the neural network, weight parameters are adjusted to learn the correlation between the calculated sprung velocity and relative velocity data and the CAN signal and vertical acceleration data. The weighting parameters obtained as the learning results are stored in the weighting parameter storage unit 14 .

[0051] Thus, the second embodiment can also achieve substantially the same effects as the first embodiment. In the second embodiment, the vehicle state calculation unit 35 has a first detection unit 35A that uses a CAN signal mounted on the vehicle, and a second detection unit 35B that is different from the first detection unit 35A. In this case, the state estimation unit 36 ​​learns, in neural network learning, the relationship between the sprung velocity and the relative velocity calculated based on the input signals (wheel speed, longitudinal G, lateral G, steering angle, yaw rate, etc.) from the first detection unit 35A and the vertical acceleration from the second detection unit 35B, and calculates the future sprung velocity and relative velocity from the vertical acceleration from the vehicle state calculation unit 35.

[0052] The state estimator 17 according to the first embodiment performs learning using CAN signals (wheel speed, longitudinal acceleration, lateral acceleration, steering angle, yaw rate, etc.) commonly used in sensorless semi-active damper systems as input signals to the neural network. In contrast, the second embodiment is applied to a system with a sensor including a vertical acceleration sensor 30. Even in this case, differentiation and integration processing is not used. Therefore, the state estimator 36 according to the second embodiment reduces the phase shift between the sprung velocity and the relative velocity near the sprung resonance frequency, similar to the state estimator 17 according to the first embodiment. In addition, the state estimator 36 can perform appropriate state estimation in situations where the sensor is inclined, such as on an inclined road surface (such as a slope or bank). Therefore, the vehicle control device 31 according to the second embodiment can also improve ride comfort on inclined road surfaces and flat roads.

[0053] In the second embodiment, the vehicle state calculation unit 35 calculates the data (vertical acceleration) necessary to estimate the sprung velocity and the relative velocity based on the detection signal of the vertical acceleration sensor 30, but the present invention is not limited to this. The vehicle state calculation unit may calculate the data necessary to estimate the sprung velocity and the relative velocity based on the detection signal of a vehicle height sensor, for example. Furthermore, the vehicle state calculation unit may calculate the data necessary to estimate the sprung velocity and the relative velocity based on the detection signals of both the vertical acceleration sensor and the vehicle height sensor.

[0054] In the second embodiment, the vehicle state calculation unit 35 calculates behavior information data from the CAN signal in addition to the detection signal of the vertical acceleration sensor 30, but the present invention is not limited to this. When the state estimation unit 36 ​​estimates the sprung velocity and the relative velocity based on the detection signal of the vertical acceleration sensor 30, the vehicle state calculation unit 35 only needs to receive the detection signal of the vertical acceleration sensor 30, and the input of the CAN signal may be omitted.

[0055] 12 to 14 show a third embodiment. The third embodiment is characterized in that the state estimation unit predicts future input signals from input signals from the vehicle state calculation unit and performs feedforward control by outputting a sprung velocity or a relative velocity in response to the future input signals. In the third embodiment, the same components as those in the first embodiment are denoted by the same reference numerals, and their description will be omitted.

[0056] The camera 40 constitutes a road surface condition measurement unit provided at the front of the vehicle body 2. The camera 40 measures and detects the road surface condition ahead of the vehicle (specifically, including the distance and angle to the road surface to be detected, and the image position and distance). The camera 40 is, for example, a stereo camera. The camera 40 captures a pair of left and right images to detect the road surface condition, including the distance and angle to the object to be captured (the road surface located in front of the vehicle). Therefore, a preview image of the area ahead of the vehicle captured by the camera 40 (i.e., road surface preview information) is output to the controller 42 as the detection result of the road surface condition measurement unit. Note that the road surface condition measurement unit is not limited to the camera 40 consisting of a stereo camera, and may be, for example, a combination of a millimeter-wave radar and a mono camera, or may be composed of multiple millimeter-wave radars, etc.

[0057] 12 shows a vehicle control device 41 according to the third embodiment. The vehicle control device 41 is made up of a suspension device 5 and a controller 42.

[0058] The controller 42 according to the third embodiment constitutes a control device. The controller 42 is configured, for example, by a microcomputer, as a control device that controls the damping characteristics of the variable damper 7. The controller 42 according to the third embodiment is configured similarly to the controller 11 according to the first embodiment. The controller 42 acquires behavior information data from the CAN 10 and the camera 40. At this time, the behavior information includes, for example, longitudinal acceleration (longitudinal G), lateral acceleration (lateral G), steering angle, yaw rate, wheel speed, road surface information, etc. Therefore, this behavior information is input to the controller 42. The output side of the controller 42 is connected to the damping force variable actuator 8 of the variable damper 7.

[0059] The controller 42 has a storage unit 43 including a ROM, a RAM, a non-volatile memory, etc. The storage unit 43 of the controller 42 stores various programs, information (vehicle information), data, etc. for controlling the variable damper 7. The controller 42 estimates the sprung velocity and relative velocity as vehicle state quantities based on the behavior information. The controller 42 calculates the force to be generated by the variable damper 7 (force generating mechanism) of the suspension unit 5 based on the estimated vehicle state quantities, and outputs a control signal (command current) to the variable damping force actuator 8 of the suspension unit 5.

[0060] 12 , the controller 42 includes a damper control unit 44, a weight parameter storage unit 14, and a data reading unit 15. The damper control unit 44 includes a vehicle state calculation unit 45, a state estimation unit 46, and a control value calculation unit 20.

[0061] The vehicle state calculation unit 45 includes a first detection unit 45A that uses a CAN signal installed in the vehicle, and a second detection unit 45B that is different from the first detection unit 45A. The first detection unit 45A acquires data related to the vehicle state while traveling from communication data transmitted within the vehicle via the CAN 10, and calculates, for example, longitudinal acceleration (longitudinal acceleration), lateral acceleration (lateral acceleration), steering angle, yaw rate, wheel speed, etc. based on the acquired data. The second detection unit 45B calculates road surface displacement ahead of the vehicle (forward in the direction of travel) from the camera 40. The vehicle state calculation unit 45 outputs vehicle behavior information to the state estimation unit 46, including information acquired from the CAN signal (longitudinal acceleration, lateral acceleration, steering angle, yaw rate, wheel speed, etc.) and information acquired from the camera 40 (road surface displacement).

[0062] As shown in Fig. 13, the state estimation unit 46 receives, as input signals from the vehicle state calculation unit 45, various types of behavior information calculated based on the CAN signal and the detection signal from the camera 40. The state estimation unit 46 includes a neural network that has learned the results of estimating the sprung velocity or the relative velocity in a predetermined frequency band in which high-frequency noise or low-frequency phase shift is reduced for the input signal from the vehicle state calculation unit 45. The weight parameters of the trained neural network are stored in the weight parameter storage unit 14. Therefore, the neural network of the state estimation unit 46 is configured using the weight parameters stored in the weight parameter storage unit 14. The neural network of the state estimation unit 46 outputs the sprung velocity and the relative velocity in response to the input signal from the vehicle state calculation unit 45.

[0063] The neural network of the state estimation unit 46 is configured in the same manner as the neural network of the state estimation unit 17 according to the first embodiment. However, the following data is provided as training data for the machine learning of the neural network. FIG. 14 shows a training method for the neural network of the state estimation unit 46. In the machine learning of the neural network of the state estimation unit 46, first, CAN signals (wheel speed, longitudinal acceleration, lateral acceleration, steering angle, yaw rate, etc.) are measured when the body of an actual vehicle vibrates, and road surface displacement and vertical acceleration are measured using various sensors. Then, the vertical acceleration when the vehicle reaches the position of the measured road surface displacement is integrated on the frequency axis to calculate the sprung velocity and relative velocity. The calculated sprung velocity and relative velocity are combined with the measured CAN signals and the road surface displacement measured by the camera 40 to create training data. This training data is used to perform machine learning of the neural network. In the neural network machine learning, the weight parameters are adjusted to learn the correlation between the calculated sprung velocity and relative velocity data, and the CAN signal and road surface displacement data. The weight parameters obtained as a result of the learning are stored in the weight parameter memory unit 14.

[0064] Thus, the third embodiment can achieve substantially the same effects as the first embodiment. In the third embodiment, the state estimation unit 46 predicts future input signals from input signals from the vehicle state calculation unit 45, and outputs sprung velocity or relative velocity in response to the future input signals, thereby allowing the vehicle control device 41 to perform feedforward control. In the first and second embodiments, vehicle body state quantities (wheel speed, vertical acceleration, etc.) are used as inputs to the neural network. However, in the present invention, the neural network inputs need only contain input signals that can be used to mathematically calculate the sprung velocity and relative velocity to be output. Therefore, as shown in the third embodiment, the present invention can also be applied to a preview control system that controls the suspension based on road surface deformation before the vehicle actually passes over it.

[0065] In the third embodiment, the road surface condition measuring unit is the camera 40 consisting of a stereo camera. However, the present invention is not limited to this. The road surface condition measuring unit may be, for example, a LiDAR (Light Detection and Ranging) unit, a combination of a millimeter-wave radar and a mono camera, or a unit configured with multiple millimeter-wave radars. Furthermore, the road surface condition measuring unit is not limited to the camera 40, and may be, for example, a unit that obtains map information from a server based on GPS data, or a unit that obtains information from other vehicles via vehicle-to-vehicle communication.

[0066] In the third embodiment, the vehicle state calculation unit 45 calculates behavior information data from the CAN signal in addition to the detection signal from the camera 40, but the present invention is not limited to this. For example, if a camera and a vehicle speed sensor are connected to the vehicle state calculation unit and the vehicle state calculation unit can calculate information (road surface displacement and vehicle speed) necessary to estimate the sprung speed and relative speed, there is no need to input the CAN signal to the vehicle state calculation unit.

[0067] Next, Fig. 15 shows a fourth embodiment. The fourth embodiment is characterized in that the state estimation unit acquires vehicle information from the location information recorded in the recording unit based on the current location information from the location information acquisition unit. In the fourth embodiment, the same components as those in the second embodiment described above are denoted by the same reference numerals, and their description will be omitted.

[0068] The GPS receiver 50 is a location information acquisition unit that acquires location information of the vehicle. The GPS receiver 50 is provided in the vehicle body 2 and receives signals from GPS (Global Positioning System) satellites (hereinafter referred to as GPS signals). The GPS receiver 50 calculates the current location information of the vehicle based on the GPS signals. The GPS receiver 50 outputs the current location information to the controller 52. Note that the location information acquisition unit is not limited to one that uses GPS signals. The location information acquisition unit may also estimate the current location information of the vehicle using techniques such as a vehicle speed sensor, a gyro, or map matching.

[0069] 15 shows a vehicle control device 51 according to the fourth embodiment. The vehicle control device 51 is made up of a suspension device 5 and a controller 52.

[0070] The controller 52 according to the fourth embodiment constitutes a control device. The controller 52 is configured, for example, by a microcomputer, as a control device that controls the damping characteristics of the variable damper 7. The controller 52 according to the fourth embodiment is configured similarly to the controller 32 according to the second embodiment. The controller 52 acquires behavior information data from the CAN 10 and the vertical acceleration sensor 30. The controller 52 is connected to a GPS receiver 50. The output side of the controller 32 is connected to the damping force variable actuator 8 of the variable damper 7.

[0071] The controller 52 has a storage unit 53 including a ROM, a RAM, a non-volatile memory, etc. The storage unit 53 of the controller 52 stores various programs, information (vehicle information), data, etc. for controlling the variable damper 7. In addition, the storage unit 53 also serves as a recording unit for recording position information, etc. Therefore, the storage unit 53 stores the position information from the GPS receiver 50 and the vehicle information from the vehicle state calculation unit 35 that has a correlation with the position information.

[0072] The controller 52 estimates the sprung velocity and relative velocity as vehicle state quantities based on the behavior information. In addition, the controller 52 estimates the sprung velocity and relative velocity as vehicle state quantities based on position information from the GPS receiver 50. The controller 52 calculates the force to be generated by the variable damper 7 (force generating mechanism) of the suspension unit 5 based on the estimated vehicle state quantities, and outputs a control signal (command current) to the variable damping force actuator 8 of the suspension unit 5.

[0073] 15 , the controller 52 includes a storage unit 53, a damper control unit 54, a weight parameter storage unit 14, and a data reading unit 15. The damper control unit 54 includes a vehicle state calculation unit 35, a state estimation unit 55, and a control value calculation unit 20.

[0074] The state estimation unit 55 has a configuration similar to that of the state estimation unit 36 ​​according to the second embodiment. The state estimation unit 55 receives, as input signals from the vehicle state calculation unit 35, various types of behavior information calculated based on the CAN signal and the detection signal of the vertical acceleration sensor 30. The state estimation unit 55 includes a neural network that has been trained to estimate the sprung velocity or the relative velocity in a predetermined frequency band in which high-frequency noise or low-frequency phase shift is reduced for the input signal from the vehicle state calculation unit 35. The weight parameters of the trained neural network are stored in the weight parameter storage unit 14. Therefore, the neural network of the state estimation unit 55 is configured using the weight parameters stored in the weight parameter storage unit 14. The neural network of the state estimation unit 55 outputs the sprung velocity and the relative velocity in response to the input signal from the vehicle state calculation unit 35.

[0075] Furthermore, the state estimation unit 55 acquires vehicle information from the position information recorded in the storage unit 53 (recording unit) based on the current position information from the GPS receiver 50. Specifically, the state estimation unit 55 identifies position information corresponding to the current position of the vehicle from the position information recorded in the storage unit 53, and acquires behavior information (longitudinal acceleration, lateral acceleration, steering angle, yaw rate, wheel speed, sprung vertical acceleration, etc.) as vehicle information corresponding to the current position. Therefore, even when a behavior state based on the current position information is input, the neural network of the state estimation unit 55 can output the sprung velocity and relative velocity in response to this input signal.

[0076] 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 based on the instantaneous values ​​of the sprung velocity and relative velocity input from the state estimation unit 55.

[0077] Thus, the fourth embodiment can achieve substantially the same effects as the first and second embodiments. In the fourth embodiment, the vehicle includes a GPS receiver 50 (position information acquisition unit) that acquires vehicle position information and a storage unit 53 (recording unit) that records the position information from the GPS receiver 50 and vehicle information from the vehicle state calculation unit 35 that is correlated with the position information. In addition, the state estimation unit 55 acquires vehicle information, such as sprung vertical acceleration, from the position information stored in the storage unit 53 based on the current position information from the GPS receiver 50. Therefore, even when a behavior state based on the current position information is input, the neural network of the state estimation unit 55 can output the sprung velocity and relative velocity in response to this input signal. Therefore, in the fourth embodiment, by using the position information from the GPS receiver 50, vehicle control can be continued even if an abnormality occurs in the vehicle state calculation unit 35. In addition, in the fourth embodiment, the vehicle is controlled using the position information from the GPS receiver 50, enabling control that predicts the vehicle state in advance through feedforward control.

[0078] Although the fourth embodiment illustrates a case in which vehicle control using position information is applied to the second embodiment, the present invention is not limited to this. Vehicle control using position information may be applied to the first and third embodiments.

[0079] In the fourth embodiment, the location information and vehicle information are stored in the storage unit 53 of the controller 52, but the present invention is not limited to this. The location information and vehicle information may be stored in an external database such as a cloud, and may be downloaded from the cloud via a communication network as needed.

[0080] In the first embodiment, the controller 11 acquires vehicle behavior information including wheel speeds via the CAN 10, but the present invention is not limited to this. For example, the controller 11 may acquire detected values ​​of various sensors directly from the various sensors. Furthermore, the controller 11 may acquire behavior information from another controller, etc. This configuration can be applied to the second to fourth embodiments.

[0081] In the above-described embodiments, the force generating mechanism is a variable damper 7 made up of a semi-active damper. However, the present invention is not limited to this, and an active damper (either an electric actuator or a hydraulic actuator) may be used as the force generating mechanism. In the above-described embodiments, the force generating mechanism that generates an adjustable force between the vehicle body 2 side and the wheel 3 side is configured by a variable damper 7 made up of a damping force adjustable hydraulic shock absorber. However, the present invention is not limited to this, and the force generating mechanism may be configured by an air suspension, a stabilizer (kinesus), an electromagnetic suspension, or the like, in addition to a hydraulic shock absorber.

[0082] In the above embodiments, the vehicle control device used in a four-wheeled vehicle has been described as an example. However, the present invention is not limited to this and may be applied to, for example, two-wheeled vehicles and three-wheeled vehicles, as well as work vehicles and transport vehicles such as trucks and buses.

[0083] The present invention is not limited to the above-described embodiments and includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those including all of the described configurations. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

[0084] This application claims priority to Japanese Patent Application No. 2023-142300, filed September 1, 2023. The entire disclosure of Japanese Patent Application No. 2023-142300, filed September 1, 2023, including the specification, claims, drawings, and abstract, is incorporated herein by reference in its entirety.

[0085] 1, 31, 41, 51: vehicle control device, 2: vehicle body, 3: wheel, 5: suspension device, 7: variable damper (relative displacement suppression device), 8: damping force variable actuator, 11, 32, 42, 52: controller, 12, 33, 43, 53: memory unit (recording unit), 16, 35, 45: vehicle state calculation unit, 17, 36, 46, 55: state estimation unit, 20: control value calculation unit (control unit), 35A, 45A: first detection unit, 35B, 45B: second detection unit, 50: GPS receiver (position information acquisition unit)

Claims

1. A vehicle control device comprising: a relative displacement suppression device provided between a vehicle body and wheels, the relative displacement suppression device changing a force suppressing relative displacement between the vehicle body and the wheels; a vehicle state calculation unit mounted on the vehicle and calculating a state of the vehicle; a state estimation unit that outputs a sprung velocity or a relative velocity in response to an input signal from the vehicle state calculation unit based on a neural network that has learned the results of estimating a sprung velocity or a relative velocity in a predetermined frequency band in which high-frequency noise is reduced or low-frequency phase shift is reduced for an input signal from the vehicle state calculation unit; and a control unit that controls the relative displacement suppression device based on the output from the state estimation unit.

2. A vehicle control device as claimed in claim 1, wherein the state estimation unit predicts a future input signal from an input signal from the vehicle state calculation unit, and performs feedforward control by outputting a sprung velocity or a relative velocity in response to the future input signal.

3. A vehicle control device as described in claim 1, wherein the vehicle state calculation unit has: a first detection unit that uses a CAN signal mounted on the vehicle; and a second detection unit that is different from the first detection unit.

4. A vehicle control device as described in claim 3, wherein the state estimation unit learns the relationship between the input signal from the first detection unit and the sprung velocity or relative velocity calculated based on the vertical acceleration or vehicle height value from the second detection unit in learning the neural network, and calculates the future sprung velocity or relative velocity from the vertical acceleration or vehicle height value from the vehicle state calculation unit.

5. A vehicle control device as described in claim 4, wherein the vehicle has: a position information acquisition unit that acquires position information of the vehicle; and a recording unit that records the position information from the position information acquisition unit and vehicle information from the vehicle state calculation unit that has a correlation with the position information, and the state estimation unit acquires the vehicle information from the position information recorded in the recording unit based on the current position information from the position information acquisition unit.

Images