Suspension control device

Abstract

To provide a suspension control device that can properly control the damping forces of shock absorbers as compared with conventional ones.SOLUTION: When shock absorbers 7A to 7D interposed between wheels 3A to 3D of a vehicle 2 and a vehicle body 4 are controlled, of a first physical quantity and a second physical quantity indicating a behavior of the vehicle 2, the first physical quantity is input to a learned learning model 35. This causes the model to estimate and output the second physical amount with high frequency components removed from the first physical amount. Based on the output second physical quantity, damping forces of the shock absorbers 7A to 7D are controlled.SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present invention relates to a suspension control device that controls the damping force of a shock absorber of a suspension. [Background technology]

[0002] A suspension is installed between the wheels and the body of a vehicle to improve the ride comfort and driving stability of the vehicle. The suspension mainly comprises suspension arms that position the axle, coil springs that support the weight of the vehicle and absorb shocks, and shock absorbers that dampen the vibrations absorbed by the coil springs, and serves the purpose of absorbing shocks from the road surface and improving the tire's ability to follow the road surface (ground contact).

[0003] Here, it has been proposed to further improve ride comfort and driving stability by controlling the damping force of the shock absorber in particular. For example, International Publication No. 2022 / 264720 discloses a technique in which the wheel speed, the longitudinal acceleration of the vehicle, the lateral acceleration of the vehicle, and the yaw rate of the vehicle are detected by various sensors provided on the vehicle as parameters indicating the behavior of the vehicle, and the parameters are input into a learned learning model to estimate the sprung velocity, which is the vertical velocity of the vehicle body, and the piston velocity, which is the stroke velocity of the piston, and the damping force of the shock absorber is controlled using the estimated sprung velocity and piston velocity. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] WO 2022 / 264720 (paragraphs 0018 and 0020) Summary of the Invention [Problem to be solved by the invention]

[0005] However, when a first physical quantity indicating the behavior of a vehicle is input to a learning model to estimate a second physical quantity different from the first physical quantity as in Patent Document 1, the second physical quantity output from the learning model contains high-frequency components and is a value that is significantly different from the true value, which is a problem. If the damping force of the shock absorber is controlled based on such a low-precision second physical quantity, appropriate control cannot be performed.

[0006] The present invention has been made to solve the above-mentioned problems in the conventional technology, and aims to provide a suspension control device that makes it possible to appropriately control the damping force of a shock absorber compared to the conventional technology, by inputting a first physical quantity that indicates the behavior of a vehicle into a learning model and estimating and outputting a second physical quantity from which high-frequency components have been removed. [Means for solving the problem]

[0007] In order to achieve the above-mentioned object, the suspension control device of the present invention is a suspension control device that controls a shock absorber interposed between a wheel and a vehicle body of a vehicle, and has a physical quantity estimation means that inputs a first physical quantity, of a first physical quantity and a second physical quantity that indicate the behavior of the vehicle, into a pre-trained learning model, thereby estimating and outputting a second physical quantity from which high-frequency components have been removed, from the first physical quantity, and a damping force control means that controls the damping force of the shock absorber based on the second physical quantity output by the physical quantity estimation means. Effect of the Invention

[0008] According to the suspension control device of the present invention having the above configuration, the first physical quantity indicating the behavior of the vehicle is input to the learning model, and the second physical quantity from which high frequency components have been removed is estimated and output, thereby making it possible to estimate the second physical quantity closer to its true value. By controlling the damping force of the shock absorber using the second physical quantity closer to its true value, it becomes possible to more appropriately control the damping force of the shock absorber than in the past. [Brief description of the drawings]

[0009] [Figure 1] 1 is a schematic configuration diagram of a vehicle according to an embodiment of the present invention. [Diagram 2] 1 is a schematic diagram of a suspension according to an embodiment of the present invention; [Diagram 3] 1 is a block diagram showing a configuration of a suspension control device according to an embodiment of the present invention. [Figure 4] FIG. 13 is a diagram illustrating an example of a trained learning model. [Diagram 5] 4 is a flowchart of a suspension control processing program according to the present embodiment. [Figure 6] 11 is a diagram showing the "sprung vertical acceleration" and the "sprung velocity" estimated as the second physical quantity. FIG. [Figure 7] 11 is a diagram comparing the "sprung vertical acceleration" estimated with high-frequency components included and the "sprung vertical acceleration" estimated with high-frequency components removed. [Figure 8] 11 is a diagram showing the relationship between the target value of the "sprung speed" and the "control current" estimated by a learning model. [Figure 9] 13 is a diagram illustrating rough road judgment using the "mid-frequency components of the sprung vertical acceleration." FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] A detailed description will be given below of a specific embodiment of a suspension control device according to the present invention with reference to the drawings. First, a vehicle 2 equipped with a suspension control device 1 according to this embodiment will be described below. Fig. 1 is a schematic diagram of the vehicle 2 according to this embodiment.

[0011] Here, the vehicle 2 may be, for example, an automobile (internal combustion engine automobile) using an internal combustion engine (engine, etc.) as a drive source, an automobile (electric automobile, fuel cell automobile, etc.) using an electric motor (motor, etc.) as a drive source, or an automobile using both of these as drive sources (hybrid automobile). In addition, the vehicle type is not limited, and may be a standard car, a large commercial truck, a bus, etc. In addition, although the following description will be of a four-wheeled automobile, it may also be a two- or three-wheeled vehicle.

[0012] In order to improve the ride comfort and running stability of the vehicle 2, suspensions 5A to 5D are respectively provided between the wheels 3A to 3D of the vehicle and the body 4. In the following description, the left front wheel is designated as 3A, the right front wheel as 3B, the left rear wheel as 3C, and the right rear wheel as 3D, and the suspension provided between the left front wheel 3A and the body 4 is designated as 5A, the suspension provided between the right front wheel 3B and the body 4 as 5B, the suspension provided between the left rear wheel 3C and the body 4 as 5C, and the suspension provided between the right rear wheel 3D and the body 4 as 5D. The type of the suspensions 5A to 5D is not particularly limited, and it is possible to adopt, for example, a strut type, a double wishbone type, a multi-link type, a torsion beam type, or the like. Also, an air suspension using an air spring instead of a metal spring may be used.

[0013] Here, the suspensions 5A to 5D mainly include suspension arms (not shown) for positioning the axles, coil springs 6A to 6D for supporting the vehicle weight and absorbing shocks, and shock absorbers 7A to 7D for damping the vibrations absorbed by the coil springs 6A to 6D, and serve to absorb shocks from the road surface and improve the tire's ability to follow the road surface (ground contact). In particular, in this embodiment, a suspension that can automatically optimize the damping force of the shock absorbers 7A to 7D depending on the road surface conditions is used. Details of the suspensions 5A to 5D will be described later.

[0014] The vehicle 2 also includes a suspension control device 1, which is a control device for controlling the damping forces of the shock absorbers 7A to 7D. Furthermore, the suspension control device 1 is also connected to various sensors mounted on the vehicle 2 via an in-vehicle network such as a CAN. The various sensors mounted on the vehicle 2 are sensors for detecting the behavior of the vehicle, and specifically include a wheel speed sensor 11, a torque sensor 12, a longitudinal acceleration sensor 13, a lateral acceleration sensor 14, a steering angle sensor 15, a yaw rate sensor 16, etc. The suspension control device 1 and the various sensors will be described in detail later.

[0015] 1, the vehicle 2 also includes basic components as the vehicle 2, but only the suspensions 5A-5D and the configuration related to their control, and the control related to the configuration will be described. In the following description, the upper portion of the vehicle 2 supported by the suspensions 5A-5D, i.e., the vehicle body 4, will also be referred to as the "sprung portion," and the lower portion of the suspensions 5A-5D that form the base, i.e., the wheels 3A-3D, will also be referred to as the "unsprung portion."

[0016] Next, the configuration of the suspensions 5A to 5D will be described in particular among the components of the vehicle 2. Since the suspensions 5A to 5D basically have the same configuration, the suspension 5A will be described as an example. Fig. 2 is a schematic configuration diagram of the suspension 5A according to this embodiment.

[0017] As shown in FIG. 2, the suspension 5A has a coil spring 6A and a shock absorber 7A. For the sake of explanation, the coil spring 6A and the shock absorber 7A are shown separately in FIG. 2, but in reality, the coil spring 6A is present so as to surround the outside of the shock absorber 7A, and are arranged so as to overlap so as to be coaxial with each other. The shock absorber 7A also has a mechanism capable of electrically controlling (adjusting) the damping force (damping characteristics). Specifically, the shock absorber 7A has an actuator 8 therein that operates based on an input current. The actuator 8 can change the opening degree of an orifice provided in the piston of the shock absorber 7A or the opening degree between the valve body and the valve seat under the control of the suspension control device 1. This controls the flow rate of lubricating oil flowing between two oil chambers partitioned by the piston in the shock absorber 7A, thereby adjusting the damping force of the shock absorber 7A. For example, when the current flowing through the actuator 8 is increased, the valve closes, the lubricant oil flows less easily, and the shock absorber 7A becomes harder (large damping force). Conversely, when the current is decreased, the shock absorber becomes softer (small damping force). The current value can be adjusted within a range of, for example, 0.4A to 1.6A.

[0018] Furthermore, the suspensions 5A-5D are provided for the four wheels 3A-3D, respectively, and the suspension control device 1 can individually control the damping force of the shock absorbers 7A-7D provided for each of the suspensions 5A-5D. Therefore, it is also possible to control the shock absorbers 7A-7D so that their damping forces are different from each other. In particular, in order to control the damping force (=damping coefficient x stroke speed) of the shock absorbers 7A-7D, it is necessary to control the damping coefficient, and the damping coefficient can be controlled by controlling the current flowing through the actuator 8 of the above-mentioned shock absorbers 7A-7D.

[0019] Next, the suspension control device 1 for controlling the suspensions 5A to 5D will be described. The suspension control device 1 is a device for controlling the damping forces of the shock absorbers 7A to 7D based on physical quantities that indicate the behavior of the vehicle 2. In addition, as described later, the suspension control device 1 also estimates the physical quantities that indicate the behavior of the vehicle 2 using a learning model. Figure 3 is a block diagram showing the configuration of the suspension control device 1 according to this embodiment.

[0020] The suspension control device 1 according to this embodiment is an electronic control unit (ECU: Electronic Control Unit) mounted on a vehicle 2, which controls the suspensions 5A to 5D, and has various means as processing algorithms. For example, the physical quantity estimation means inputs the first physical quantity, of a first physical quantity and a second physical quantity indicating the behavior of the vehicle, into a pre-trained learning model, thereby estimating and outputting the second physical quantity from which high-frequency components have been removed, from the first physical quantity. The damping force control means controls the damping force of the shock absorber based on the second physical quantity output by the physical quantity estimation means.

[0021] Specifically, as shown in Fig. 3, the system includes internal storage devices such as a CPU 31 as an arithmetic unit and control unit, a RAM 32 used as a working memory when the CPU 31 performs various arithmetic processing, a ROM 33 in which a suspension control processing program (see Fig. 5) described below and the like are recorded in addition to control programs, and a flash memory 34 for storing programs read out from the ROM 33. The suspension control unit 1 is also connected to various sensors such as a wheel speed sensor 11, a torque sensor 12, a longitudinal acceleration sensor 13, a lateral acceleration sensor 14, a steering angle sensor 15, and a yaw rate sensor 16 that are installed on the vehicle 2 via an in-vehicle network such as CAN.

[0022] In addition to various data related to the vehicle 2, the flash memory 34 also stores a trained learning model 35. As the learning model 35, for example, a neural network is used, and more preferably, an LSTM (Long Short Term Memory) is used. The LSTM is a learning model that can estimate the present based on past information, and can take into account a longer time series than a similar RNN (Recurrent Neural Network). In particular, the learning model 35 of this embodiment is a learning model that is pre-trained to estimate and output a second physical quantity that indicates the behavior of the vehicle in the same manner as the first physical quantity, such as "medium frequency components of the vertical acceleration occurring in the vehicle body 4", "low frequency components (including resonant frequencies of heave, roll, and pitch) of the acceleration of the center of gravity of the vehicle (heave acceleration, roll angular velocity, and pitch angular velocity)", and "sprung velocity which is the vertical velocity of the vehicle body 4", by inputting a physical quantity that indicates the behavior of the vehicle other than the vertical acceleration, specifically, a first physical quantity that indicates the behavior of the vehicle, such as "wheel speed", "front-rear and lateral acceleration", "yaw rate", "steering angle and steering angular velocity", "wheel rotation torque", and "actual current (current flowing through the shock absorbers 7A to 7D)". Here, in the learning model 35 using a neural network, data obtained by multiplying each neuron by a weight (weighting coefficient), which is output data after processing the first physical quantity input to the input layer in the input layer, as shown in FIG. 4, is input to the next intermediate layer. Similarly, in the intermediate layer, data obtained by multiplying each neuron, which is output data after processing in the intermediate layer, by a weight (weighting coefficient) is input to the next output layer. Then, the second physical quantity is finally output from the output layer. In the learning of the learning model 35, the weight (weighting coefficient) is appropriately changed and set to a more appropriate value as the learning progresses. Furthermore, in the LSTM, an input gate and an output gate (which control how much new data is reflected or output at each time) are provided for each of the input and output data of the intermediate layer, and the weights constituting the gates are also subject to learning.

[0023] In this embodiment, the suspension control device 1 has a learning model that has been sufficiently learned in advance using teacher data (the true value of the measured second physical quantity) and the like, but the learning model may also be learned in parallel with the output of the second physical quantity. Learning without a teacher is also possible. During learning, a specific frequency component may be learned. Furthermore, the learning model 35 does not necessarily have to be composed of only one neural network, and may include multiple neural networks.

[0024] On the other hand, the wheel speed sensor 11 is attached to the wheels 3A to 3D of the vehicle 2, detects the rotation speed of the wheels, and outputs a speed signal. Note that the suspension control device 1 can also identify the speed of the vehicle 2 in the traveling direction (vehicle speed) from the detection signal of the wheel speed sensor 11.

[0025] Similarly, the torque sensors 12 are attached to the wheels 3A to 3D of the vehicle 2, and detect and output the rotational torque input to the wheels 3A to 3D.

[0026] Furthermore, the longitudinal acceleration sensor 13 detects and outputs acceleration occurring in the longitudinal direction (parallel to the vehicle's traveling direction) with respect to the body 4 of the vehicle 2. Similarly, the lateral acceleration sensor 14 detects and outputs acceleration occurring in the left-right direction (direction intersecting the vehicle's traveling direction) with respect to the body 4 of the vehicle 2. Note that the longitudinal acceleration sensor 13 and the lateral acceleration sensor 14 may be sensors capable of detecting acceleration in multiple axial directions.

[0027] In addition, the steering angle sensor 15 is attached inside the steering device, detects the steering angle when the steering wheel is turned, and outputs a steering angle signal. Note that the suspension control device 1 can also identify the speed at which the steering angle changes from the detection signal of the steering angle sensor 15.

[0028] Further, the yaw rate sensor 16 detects the yaw rate of the vehicle 2 and outputs the detected yaw rate.

[0029] Based on the output signals from the various sensors, the suspension control device 1 can acquire the wheel speed, the acceleration in the longitudinal and lateral directions, the yaw rate, the steering angle and the steering angle velocity, and the rotational torque of the wheels, which are physical quantities indicating the current behavior of the vehicle 2. These physical quantities become the first physical quantities to be input to the input layer of the learning model 35 described above.

[0030] Next, a suspension control processing program executed in the suspension control device 1 having the above configuration will be described with reference to Fig. 5. Fig. 5 is a flowchart of the suspension control processing program according to this embodiment. Here, the suspension control processing program is executed after the ACC power supply (accessory power supply) of the vehicle is turned on, and is a program that estimates physical quantities indicating the behavior of the vehicle using a learning model, and controls the damping forces of the shock absorbers 7A to 7D using the estimated physical quantities. The program shown in the flowchart in Fig. 5 below is stored in the RAM 32, ROM 33, etc. provided in the suspension control device 1, and is executed by the CPU 31.

[0031] First, in step (hereinafter abbreviated as S) 1, the CPU 31 acquires detection results of various sensors such as the wheel speed sensor 11, the torque sensor 12, the longitudinal acceleration sensor 13, the lateral acceleration sensor 14, the steering angle sensor 15, and the yaw rate sensor 16 via the CAN or the like as physical quantities indicating the behavior of the vehicle. Specifically, the CPU 31 acquires the "wheel speed", the "longitudinal and lateral acceleration", the "yaw rate", the "steering angle and steering angle speed", and the "wheel rotation torque". In addition, the CPU 31 also acquires the "actual current (current flowing through the shock absorbers 7A to 7D)". Each physical quantity acquired in the above S1 is hereinafter referred to as a first physical quantity. The "wheel speed", the "wheel rotation torque", and the "actual current" are acquired for each of the four wheels (shock absorbers). Note that the first physical quantity may be a physical quantity other than those listed above. However, the vertical acceleration occurring in the vertical direction relative to the vehicle 2 is excluded from the first physical quantity because it is the physical quantity to be estimated from the first physical quantity by the learning model 35 in S2 described below.

[0032] Next, in S2, the CPU 31 inputs the first physical quantity acquired in S1 to the input layer of the trained learning model 35, thereby estimating and outputting a second physical quantity that indicates the vehicle behavior in the same way as the first physical quantity but is different from the first physical quantity. As described above, the learning model 35 is a learning model that has been trained in advance to estimate and output the second physical quantity by inputting the first physical quantity (FIG. 4). However, learning of the learning model 35 may also be performed in parallel with the estimation of the second physical quantity by the learning model 35.

[0033] Here, in this embodiment, the second physical quantities are the "mid-frequency component of the sprung vertical acceleration", which is the vertical acceleration occurring in the vehicle body 4, the "low-frequency component (including the resonant frequencies of heave, roll, and pitch) of the acceleration of the vehicle's center of gravity (heave acceleration, roll angular acceleration, and pitch angular acceleration)", and the "sprung velocity", which is the vertical velocity of the vehicle body 4. As shown in FIG. 6, for the "sprung vertical acceleration α" and the "sprung velocity v", the vertical acceleration and vertical velocity at a point on the vehicle body directly above that supported by the suspensions 5A-5D are output for each of the four suspensions 5A-5D.

[0034] It is also possible to output only one of the "mid-frequency component of the sprung vertical acceleration" and the "low-frequency component of the acceleration at the center of gravity of the vehicle." In addition, the second physical quantity may be an output target other than the above-mentioned physical quantities. For example, the output target may be the "low-frequency component of the vertical acceleration occurring in the vehicle body 4" or the "mid-frequency component of the acceleration at the center of gravity of the vehicle." In addition, the output target may be the "stroke speed (the vertical speed of the vehicle body 4 relative to the wheels 3A to 3D) of each suspension 5A to 5D" and the "stroke displacement (how the vehicle body 4 is displaced in the vertical direction relative to the wheels 3A to 3D) of each suspension 5A to 5D."

[0035] Furthermore, in addition to estimating the second physical quantity using the learning model 35, the "unsprung high-frequency vibration components, which are the vibrations of the wheels 3A to 3D," may also be calculated by performing a band-pass filter calculation on the wheel speed of each wheel detected by the wheel speed sensor 11.

[0036] Here, in this embodiment, the learning model 35 is trained so that the second physical quantity to be output by the learning model 35 is estimated and output as a physical quantity from which high-frequency components have been particularly removed. When estimating the second physical quantity, "sprung vertical acceleration," using the learning model 35 as shown in FIG. 7, if the estimation is performed with high-frequency components included, a large gap occurs between the true value (including high-frequency components) and the estimated value as shown in the upper diagram of FIG. 7. On the other hand, if the estimation is performed with, for example, mid-frequency components excluding the high-frequency components, the true value (only mid-frequency components) and the estimated value have a high degree of agreement as shown in the lower diagram of FIG. 7. Although not shown, a similar high degree of agreement is obtained when estimating with low-frequency components.

[0037] Thereafter, in S3, the CPU 31 calculates the target damping coefficients of the shock absorbers 7A-7D using the second physical quantities estimated in S2, particularly the "sprung velocity" which is the vertical velocity of the vehicle body 4. As described above, the shock absorbers 7A-7D have built-in actuators 8 which operate based on an input current, and are capable of electrically controlling (adjusting) the damping force (damping characteristics) (FIG. 2). Specifically, the damping coefficients can be controlled by controlling the current flowing through the actuators 8 of the shock absorbers 7A-7D, and the damping force (=damping coefficient x stroke speed) can be controlled by controlling the damping coefficient.

[0038] Therefore, in S3, the CPU 31 calculates the target value of the "control current" to be passed through the shock absorbers 7A to 7D by multiplying the "sprung speed" estimated in S2 by a proportional gain. As shown in FIG. 8, the estimated "sprung speed" and the target value of the "control current" are in a proportional relationship, and the target value of the "control current" increases as the "sprung speed" increases in the positive direction (upward is positive, downward is negative), and the target value of the "control current" decreases as the sprung speed increases in the negative direction. As shown in FIG. 6, the "sprung speed" is estimated for each of the suspensions 5A to 5D, so the target value of the "control current" is also calculated for each of the shock absorbers 7A to 7D using the "sprung speed" corresponding to each of the shock absorbers 7A to 7D. That is, a different value is calculated as the target value of the "control current" for each of the shock absorbers 7A to 7D. Since the damping coefficient is also determined by the current flowing through the actuator 8 of the shock absorbers 7A to 7D, the target damping coefficients of the shock absorbers 7A to 7D are also determined by the target value of the "control current" calculated in S3 (i.e., the damping coefficients when the target value of the "control current" is flowed through the shock absorbers 7A to 7D become the target damping coefficients).

[0039] Next, in S4, the CPU 31 performs a bad road judgment as to whether or not the vehicle is traveling on a bad road by using, among the second physical quantities estimated in S2, in particular, the "mid-frequency component of the sprung vertical acceleration" which is the vertical acceleration occurring in the vehicle body 4, or the "low-frequency component (including the resonance frequency of heave, roll, and pitch) of the acceleration of the center of gravity of the vehicle (heave acceleration, roll angular acceleration, and pitch angular acceleration)." Note that the bad road judgment may be performed by using only one of the "mid-frequency component of the sprung vertical acceleration" and the "low-frequency component of the acceleration of the center of gravity of the vehicle," or may be performed by using both. In addition, when the "low-frequency component of the vertical acceleration occurring in the vehicle body 4," the "mid-frequency component of the acceleration of the center of gravity of the vehicle," or the "stroke displacement and stroke speed of each of the suspensions 5A to 5D" are estimated as the second physical quantities, it is also possible to perform the bad road judgment by using any of them.

[0040] Specifically, if either or both of the "mid-frequency component of the sprung vertical acceleration" and the "low-frequency component of the acceleration of the vehicle's center of gravity" exceed a threshold, it is determined that the vehicle is traveling on a rough road. The threshold can be set appropriately, but for example, the "mid-frequency component of the sprung vertical acceleration" is set to ±0.2 m / s as shown in FIG. 9. 2 (upper is positive, lower is negative) is the threshold, 0.2m / s 2 Greater than or -0.2m / s 2 If the "mid-frequency component of the sprung vertical acceleration" is used to determine whether the vehicle is traveling on a rough road, the "mid-frequency component of the sprung vertical acceleration" is estimated for each of the suspensions 5A to 5D as shown in Fig. 6, and it is determined whether any of the "mid-frequency components of the sprung vertical acceleration" exceeds a threshold value.

[0041] In addition, when the "unsprung high frequency vibration component, which is the vibration of the wheels 3A to 3D" is calculated from the wheel speed of each wheel, it is also possible to perform a rough road judgment using the "unsprung high frequency vibration component." In this case, it is judged that the vehicle is on a rough road when the "unsprung high frequency vibration component" exceeds a threshold value.

[0042] Then, in S5, the CPU 31 judges whether or not the vehicle is determined to be traveling on a bad road in the bad road judgment in S4. If it is determined that the vehicle is not traveling on a bad road (S5: NO), the process proceeds to S6. On the other hand, if it is determined that the vehicle is traveling on a bad road (S5: YES), the process proceeds to S7.

[0043] In S6, the CPU 31 controls the current flowing through the actuators 8 of the shock absorbers 7A-7D so that the "control current" reaches the target value (target damping coefficient) calculated in S3. Since the target value of the "control current" differs for each shock absorber 7A-7D, the current is also controlled for each shock absorber 7A-7D. As a result, it becomes possible to optimally control the damping force of the shock absorbers 7A-7D according to the current vehicle situation, improving ride comfort and driving stability.

[0044] On the other hand, in S7, the CPU 31 sets upper and lower limits for the damping amount of the shock absorbers 7A to 7D. Specifically, the upper and lower limits are set for the amount of current flowing through the shock absorbers 7A to 7D, thereby setting the upper and lower limits for the damping amount. For example, the upper limit is set to 1.3 A and the lower limit is set to 0.7 A. Note that, although both the upper and lower limits are set in this embodiment, only the upper limit may be set, or only the lower limit may be set.

[0045] After that, in S8, the CPU 31 controls the current flowing through the actuators 8 of the shock absorbers 7A to 7D so that the target value of the "control current" calculated in S3 (target damping coefficient) is reached. However, the control is performed within a range in which the damping force of the shock absorbers 7A to 7D does not exceed the upper and lower limits set in S7 (a range in which the amount of current does not exceed the upper and lower limits). Specifically, as shown in FIG. 9, if the target value of the "control current" exceeds the upper or lower limit set in S7, the upper or lower limit is set as the target value. As a result, the damping force of the shock absorbers 7A to 7D is controlled within the range between the upper and lower limits that have been set.

[0046] Since the target value of the "control current" differs for each shock absorber 7A-7D, the current is also controlled for each shock absorber 7A-7D. As a result, it becomes possible to optimally control the damping force of shock absorbers 7A-7D according to the current vehicle conditions while preventing the damping force from becoming larger than necessary when driving on rough roads, thereby improving ride comfort and driving stability.

[0047] As described above in detail, in the suspension control device 1 and the computer program executed by the suspension control device 1 according to this embodiment, when controlling the shock absorbers 7A-7D interposed between the wheels 3A-3D of the vehicle 2 and the vehicle body 4, the first physical quantity of the first and second physical quantities indicating the behavior of the vehicle 2 is input to the pre-trained learning model 35, the second physical quantity from which high-frequency components have been removed is estimated from the first physical quantity and output (S2), and the damping forces of the shock absorbers 7A-7D are controlled based on the output second physical quantity (S6-S8), so that it is possible to estimate the second physical quantity closer to the true value. Then, by controlling the damping forces of the shock absorbers 7A-7D using the second physical quantity closer to the true value, it is possible to appropriately control the damping forces of the shock absorbers 7A-7D compared to the conventional method. Furthermore, if the second physical quantity output from the learning model 35 exceeds a threshold value, it is determined that the vehicle 2 is traveling on a rough road (S5), and if it is determined that the vehicle 2 is traveling on a rough road, at least one of an upper limit and a lower limit is set for the damping force of the shock absorbers 7A-7D (S7), and the damping force of the shock absorbers 7A-7D is controlled within a range not exceeding the set upper limit or lower limit (S8). Therefore, by performing the rough road determination using the second physical quantity closer to the true value, the accuracy of the rough road determination is improved compared to the conventional method. And, by controlling the damping force of the shock absorbers 7A-7D based on the highly accurate rough road determination result, it becomes possible to appropriately control the shock absorbers 7A-7D. In addition, since the first physical quantity is a physical quantity that indicates the behavior of the vehicle other than the vertical acceleration, and the second physical quantity includes either a mid-frequency component or a low-frequency component that is obtained by removing high-frequency components in the vertical acceleration occurring on the vehicle body, or a mid-frequency component or a low-frequency component that is obtained by removing high-frequency components in the acceleration of the center of gravity of the vehicle, it is possible to estimate the second physical quantity that is closer to the true value by removing the high-frequency components. Then, by controlling the damping forces of the shock absorbers 7A-7D using the second physical quantity that is closer to the true value, it is possible to perform a rough road judgment with higher accuracy than before, and to appropriately control the damping forces of the shock absorbers 7A-7D. In addition, by inputting the first physical quantity into the learning model 35, the sprung velocity, which is the vertical velocity of the vehicle body, is estimated and output in addition to the second physical quantity (S2), target damping coefficients of shock absorbers 7A to 7D are calculated based on the sprung velocity (S3), and the damping forces of shock absorbers 7A to 7D are controlled based on the target damping coefficients (S6 to S8).By controlling the damping forces of shock absorbers 7A to 7D using sprung velocity that is closer to the true value, it is possible to more appropriately control the damping forces of shock absorbers 7A to 7D than in the conventional method.

[0048] Incidentally, the present invention is not limited to the above-described embodiment, and it is needless to say that various improvements and modifications are possible without departing from the spirit and scope of the present invention. For example, in this embodiment, LSTM (Long Short Term Memory), which is one type of neural network, is used as the learning model 35, but other types of learning models may be used for the format of the learning model 35. For example, RNN (Recurrent Neural Network) may be used.

[0049] In addition, in this embodiment, the first physical quantity is an example shown in Fig. 4, but as long as it is a physical quantity that indicates the behavior of the vehicle, it is not limited to the example shown in Fig. 4. Similarly, the second physical quantity estimated from the first physical quantity is an example shown in Fig. 4, but as long as it is a physical quantity that can be used to determine a bad road or calculate a target damping coefficient, it is not limited to the example shown in Fig. 4.

[0050] In the present embodiment, the suspension control processing program shown in Fig. 5 is executed by the suspension control device 1, but may be executed by another vehicle-mounted device such as a navigation device. Also, a part of the processing may be executed by an external server device. [Explanation of symbols]

[0051] 1...Suspension control device, 2...Vehicle, 3A to 3D...Wheels, 4...Vehicle body, 5A to 5D...Suspension, 6A to 6D...Coil spring, 7A to 7D...Shock absorber, 8...Actuator, 31...CPU, 35...Learning model

Claims

1. A suspension control device that controls a shock absorber interposed between a wheel and a vehicle body of a vehicle, a physical quantity estimating means for estimating and outputting a second physical quantity from which high frequency components have been removed, based on a first physical quantity, by inputting the first physical quantity, of a first physical quantity and a second physical quantity, which indicate a behavior of a vehicle, into a previously trained learning model; and a damping force control means for controlling the damping force of the shock absorber based on the second physical quantity output by the physical quantity estimation means.

2. a rough road determination means for determining that the vehicle is traveling on a rough road when the second physical quantity output by the physical quantity estimation means exceeds a threshold value, 2. A suspension control device as described in claim 1, wherein the damping force control means sets at least one of an upper limit and a lower limit for the damping force of the shock absorber when the rough road determination means determines that the vehicle is traveling on a rough road, and controls the damping force of the shock absorber within a range not exceeding the set upper limit or lower limit.

3. The first physical quantity is a physical quantity indicating a behavior of the vehicle other than the vertical acceleration, 3. The suspension control device according to claim 1 or 2, wherein the second physical quantity includes either a mid-frequency component or a low-frequency component obtained by removing high-frequency components from the vertical acceleration acting on the vehicle body, or a mid-frequency component or a low-frequency component obtained by removing high-frequency components from the acceleration of the center of gravity of the vehicle.

4. A sprung velocity estimation means for estimating and outputting a sprung velocity, which is a vertical velocity of the vehicle body, in addition to the second physical quantity by inputting the first physical quantity into the learning model; a target damping coefficient calculation means for calculating a target damping coefficient of the shock absorber based on the sprung speed, 3. The suspension control device according to claim 1, wherein the damping force control means controls the damping force of the shock absorber based on the target damping coefficient.

Images