Suspension control device, suspension control method and suspension control system

Abstract

The suspension control device includes: a parameter storage unit that stores parameters, pre-determined through machine learning, representing the correspondence between behavioral information and sensor data, wherein the behavioral information represents the behavior of the vehicle and the sensor data pertains to the vehicle's suspension; a vehicle behavior estimation unit that inputs the behavioral information from communication data transmitted within the vehicle, and estimates the sensor data pertaining to the vehicle's suspension based on the behavioral information and the parameters stored in the parameter storage unit; and a control value calculation unit that calculates a suspension control value for controlling the suspension based on the estimated sensor data.

Description

Technical Field

[0001] This invention relates to a suspension control device, a suspension control method, and a suspension control system. Background Technology

[0002] A known suspension control device adjusts the suspension stiffness and characteristics based on road conditions and driving conditions to control the vehicle's attitude. The device controls the suspension by detecting sensor data from sensors that monitor piston speeds and spring accelerations in the four-wheel suspension. Alternatively, a sensorless suspension control device could be considered.

[0003] Patent document 1 describes a suspension control device that does not use expensive sensors such as travel sensors. When the value of wheel speed variation detected by the wheel speed sensor is above a specified value on the negative side with zero as the reference, it is regarded as a reduction in wheel ground load, and the damping force is increased. The damping force of the variable damper installed on the left and right front wheels is controlled based on the state quantity of the wheel speed variation of the left and right front wheels that is greater. Existing technical documents Patent documents

[0004] Patent Document 1: Japanese Patent Application Publication No. 2019-189228 Summary of the Invention The technical problem that the invention aims to solve

[0005] In the device described in Patent Document 1, the accuracy of the suspension control corresponding to the vehicle's behavior is insufficient. Technical solutions to solve technical problems

[0006] The suspension control device of the present invention includes: a parameter storage unit that stores parameters, determined in advance through machine learning, representing the correspondence between behavioral information and sensor data, wherein the behavioral information represents the behavior of the vehicle and the sensor data relates to the vehicle's suspension; a vehicle behavior estimation unit that inputs the behavioral information from communication data transmitted within the vehicle, and estimates the sensor data related to the vehicle's suspension based on the behavioral information and the parameters in the parameter storage unit; and a control value calculation unit that calculates a suspension control value for controlling the suspension based on the estimated sensor data. The suspension control method involved in this invention is a suspension control method in a suspension control device for controlling the suspension of a vehicle. In this suspension control method, parameters representing the correspondence between behavioral information and sensor data, which are determined in advance through machine learning, are stored. The behavioral information represents the behavior of the vehicle, and the sensor data relates to the suspension of the vehicle. The behavioral information is input into the communication data transmitted within the vehicle. Based on the behavioral information and the parameters, the sensor data related to the suspension of the vehicle is inferred. Based on the inferred sensor data, a suspension control value for controlling the suspension is calculated. The suspension control system of this invention includes a server device and a suspension control device for controlling the suspension of a vehicle. In this suspension control system, the server device acquires behavioral information representing the behavior of the vehicle and sensor data related to the vehicle's suspension, and determines parameters representing the correspondence between the behavioral information and the sensor data through machine learning. The suspension control device acquires the parameters through the server device, inputs the behavioral information from the communication data transmitted within the vehicle into the suspension control device, and infers the sensor data related to the vehicle's suspension based on the behavioral information and the parameters, and controls the suspension based on the inferred sensor data. Invention Effects

[0007] According to the present invention, the accuracy of suspension control corresponding to the behavior of the vehicle is improved. Attached Figure Description

[0008] Figure 1 (A), (B), and (C) are diagrams representing vehicles equipped with suspension control devices. Figure 2 This is a diagram illustrating an example of a vehicle behavior prediction unit. Figure 3 (A) and (B) are diagrams representing the modular structure of a vehicle for data acquisition and an example of the data. Figure 4 This is a block diagram of the learning device. Figure 5 It is a flowchart representing the process of acquiring data and learning from it. Figure 6 This is an example of a display of behavioral information and sensor data. Figure 7 This is a diagram representing a variation of the vehicle behavior prediction section, as shown in Example 1. Figure 8 (A), (B), and (C) are diagrams representing variations 2, 3, and 4 of the vehicle behavior prediction section. Figure 9(A) and (B) are diagrams representing variations of the vehicle behavior prediction section, as shown in Figure 5. Figure 10 This is a diagram representing a variation of example 6 of the vehicle behavior prediction section. Figure 11 This is a block diagram of the suspension control system. Detailed Implementation

[0009] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. The following description and drawings are examples for illustrating the present invention, and appropriate omissions and simplifications have been made for clarity. The present invention may also be implemented in various other ways. Unless otherwise specified, the structural elements may be singular or plural.

[0010] When multiple structural elements have the same or equivalent functions, different subscripts are sometimes added to the same label for description. However, when it is not necessary to distinguish between these multiple structural elements, the subscripts are sometimes omitted for description.

[0011] Furthermore, in the following description, processing is sometimes described as being performed by executing a program. However, since the program is executed by a processor (e.g., CPU, GPU), the processor can be the subject of the processing, performing the specified processing while appropriately utilizing storage resources (e.g., memory) and / or interface devices (e.g., communication ports). Similarly, the subject of processing by executing a program can also be a controller, device, system, computer, or node that has a processor. The subject of processing by executing a program can be any arithmetic unit, but may also include dedicated circuitry (e.g., FPGA, ASIC) for performing specific processing.

[0012] The program can be installed from a program source onto a device such as a computer. The program source can be, for example, a program distribution server or a computer-readable storage medium. If the program source is a program distribution server, the program distribution server contains storage resources for a storage processor and the program to be distributed; the processor of the program distribution server can also distribute the program to other computers. Furthermore, in the following description, two or more programs can be implemented as one program, or one program can be implemented as two or more programs.

[0013] [Implementation Method 1] Figure 1 (A), (B), and (C) are figures representing a vehicle 101 equipped with the suspension control device 111 according to Embodiment 1 of the present invention. Figure 1 (A) is an exterior view of vehicle 101. Figure 1 (B) is a module structure diagram of the suspension control device 111. Figure 1 (C) is a diagram representing the vehicle behavior prediction unit 107.

[0014] like Figure 1 As shown in (A), vehicle 101 includes a front suspension 106a and a rear suspension 106b. By controlling the suspensions 106a and 106b, vibrations transmitted from the road surface S to vehicle 101 via the wheels during driving are suppressed. The damping force characteristics of suspensions 106a and 106b can be controlled, and suspension control device 111 controls the damping force of suspensions 106a and 106b.

[0015] like Figure 1 As shown in (B), the suspension control device 111 includes a control unit 102, a weight parameter storage unit 103, and a data reading unit 104. Furthermore, the control unit 102 includes a vehicle behavior prediction unit 107 and a control value calculation unit 108.

[0016] The suspension control device 111 obtains data (hereinafter referred to as behavior information) related to the vehicle's behavior during driving from the communication data transmitted in the vehicle via CAN (Controller Area Network) 105. Based on the behavior information and weight parameters, it calculates the suspension control value using the weight parameters in the storage unit 103 and controls the suspension 106 (106a, 106b).

[0017] In this embodiment, CAN is used as an example of an in-vehicle network, but other in-vehicle networks can also be used. For example, CAN FD (CAN with Flexible Datarate), FlexRay (in-vehicle network), and in-vehicle Ethernet can also be implemented in the same way.

[0018] Next, the damping force control of the suspension 106 implemented by the suspension control device 111 will be described. The vehicle behavior estimation unit 107 within the control unit 102 reads the weight parameters stored in the weight parameter storage unit 103 via the data reading unit 104. Furthermore, based on the behavior information and weight parameters received from the CAN 105, the vehicle behavior estimation unit 107 estimates the data required by the control value calculation unit 108 for suspension control; specifically, it estimates the instantaneous values ​​involved in suspension control. The control value calculation unit 108 calculates the suspension control value for controlling the damping force of the suspension 106 based on the instantaneous values ​​input from the vehicle behavior estimation unit 107.

[0019] Here, a comparative example compared to this embodiment will be described. In this comparative example, an acceleration sensor is installed on the suspension 106 mounted on each of the four wheels of the vehicle to measure the vertical speed of the spring (hereinafter referred to as spring speed) and the stroke speed of the piston (hereinafter referred to as piston speed), and the suspension is controlled based on the measurement results. Therefore, an acceleration sensor needs to be installed on each of the four wheels of the vehicle, which increases costs due to the increase in components such as brackets and wiring harnesses, as well as the increase in assembly time. In addition, suspension control that reflects the overall rigidity characteristics of the vehicle, which is composed of multiple components, cannot be performed.

[0020] In contrast, in this embodiment, no acceleration sensors are installed on the suspension 106 mounted on each of the four wheels of vehicle 101. In this embodiment, based on behavior information such as wheel speed, front-rear acceleration, lateral acceleration, and yaw rate, the vehicle behavior estimation unit 107 estimates the data (spring speed, piston speed) related to suspension control for each of the four wheels from the data constantly transmitted to CAN 105. Then, the estimation result is transmitted to the subsequent control value calculation unit 108. As a result, there is no need to install acceleration sensors, which reduces costs. Furthermore, the vehicle behavior estimation unit 107 estimates the sensor data related to suspension control for each of the four wheels, but the vehicle behavior estimation unit 107 is, for example, composed of a neural network. Since the weight parameters used in the neural network are parameters that are pre-determined for vehicle 101 through machine learning, they also reflect the overall rigidity characteristics of the vehicle, thus improving the accuracy of suspension control corresponding to the behavior of vehicle 101.

[0021] like Figure 1 As shown in (C), weighting parameters and behavioral information are input to the vehicle behavior estimation unit 107. Then, the vehicle behavior estimation unit 107 estimates instantaneous values, such as sensor data like spring speed and piston speed, based on the weighting parameters and behavioral information, and outputs them.

[0022] Figure 2 This is a diagram illustrating an example of the vehicle behavior prediction unit 107. like Figure 2As shown, in this example, the vehicle behavior prediction unit 107 is composed of a neural network. The neural network is a hierarchical neural network with a three-layer structure, consisting of elements of the input layer (number of elements i) 201, the hidden layer (number of elements j) 202, and the output layer (number of elements K) 203. Each element of the input layer 201 and each element of the hidden layer 202 are combined by weights W1ij (i = 1 to I, j = 1 to J), and each element of the hidden layer 202 and each element of the output layer 203 are combined by weights W2jk (j = 1 to J, k = 1). The information of these weights (hereinafter referred to as weight parameters) is represented by the determinant of weights W1ij and weights W2jk. The weight parameters will be described in detail later, but weight parameters that are pre-obtained through machine learning and stored in the weight parameter storage unit 103 are used. In this example, the simplest hidden layer 202 shows a single-layer fully connected neural network, but it is not limited to this.

[0023] like Figure 2 As shown, the input layer 201 of the neural network consists of input layer 201a, which receives time-series data of the wheel speed of the front right wheel, and input layer 201b, which receives time-series data of the front-rear acceleration of vehicle 101. It is assumed that the instantaneous value of the velocity on the spring of the suspension mounted on the front right wheel of vehicle 101 is output to output layer 203. The number of components in hidden layer 202 is generally determined by the number of components in input layer 201 and output layer 203, but is set to maximize the accuracy of vehicle behavior prediction based on the neural network. The number of components in output layer 203 is determined by the output specifications of the vehicle behavior prediction.

[0024] As an output specification, the case where the number of elements in output layer 203 is 256 will be explained. For example, the velocity on a spring, which is inferred to represent vehicle behavior, is a simulated instantaneous value, but the output of the neural network is either 1 or 0, so it needs to be represented numerically. Therefore, the range of simulated instantaneous values ​​is divided into multiple levels, and the divided values ​​are associated with the output of the neural network in a one-to-one relationship. For example, assuming the range of instantaneous values ​​is -1.00 to 1.00, and this range is represented by 256 levels, then each level is 0.007843 (={1.00-(-1.00)}÷255). Here, the output Y1 of the neural network is assigned to 1.000000, Y2 corresponds to 0.992157, Y3 corresponds to 0.984312, ..., Y256 corresponds to -1.000000. Furthermore, only the output element corresponding to the instantaneous value is designated as "1" (high), and all other elements are designated as "0" (low). Thus, instantaneous values ​​can be represented in the output layer 203 of the neural network. In this example, the case where the number of elements in the output layer 203 is 256 is illustrated, but it is not limited to this number.

[0025] When the vehicle behavior prediction unit 107 is composed of a neural network, behavior information arriving from the CAN 105 is input into the neural network. Specifically, for behavior information arriving from the CAN 105 in a time sequence, data is acquired at a certain sampling rate within a specified time window 310 (see below). Figure 3 Define a dataset in (B). For example, if the sampling interval is set to 20 m seconds and the window width is set to 1 second, then a dataset for a certain physical quantity consists of 50 points (=1 second ÷ 20 m seconds). These 50 points are assigned to 50 neural network input elements. Then, the neural network input is updated by shifting the window width in the time direction. Figure 2 In this process, since the wheel speed of the right front wheel and the front-rear acceleration of vehicle 101 are taken as inputs, the input elements of the neural network are i = 100 points (i = 50 points × 2 kinds of physical quantities).

[0026] For ease of explanation, Figure 2 A neural network for controlling the suspension 106 mounted on the front right wheel of vehicle 101 is shown, but the vehicle behavior prediction unit 107 corresponds to a neural network with the same structure for all four wheels of vehicle 101.

[0027] Figure 3 (A) and (B) are diagrams representing the module structure of vehicle 301 for data acquisition and an example of the data. The data acquisition vehicle 301 is used to pre-acquire the correspondence between behavioral information transmitted to the CAN 105 and sensor data. Therefore, the data acquisition vehicle 301 has a suspension acceleration sensor 302 installed on the suspension 106 to acquire the spring speed and piston speed.

[0028] like Figure 3 As shown in (A), the data acquisition vehicle 301 includes a suspension acceleration sensor 302, a control unit 303, a data collection unit 304, a dataset storage unit 305, a data reading unit 306, and a display unit 307. The control unit 303 has a control value calculation unit 108. CAN 105, suspension 106, and reference... Figure 1 The same applies to (B).

[0029] The control value calculation unit 108 of the control unit 303 calculates the suspension control value of the damping force of the suspension 106 based on the instantaneous value output by the suspension acceleration sensor 302. Additionally, the data collection unit 304 acquires the behavior information transmitted to the CAN 105 and the sensor data output by the suspension acceleration sensor 302, and stores each piece of information as a dataset in the dataset storage unit 305. The display unit 307 displays the dataset stored in the dataset storage unit 305 at any time via the data reading unit 306. The display unit 307 shows an example installed on the data acquisition vehicle 301, but it can also be installed outside the vehicle via a network, such as on a server device.

[0030] Figure 3 (B) is an example of a dataset acquired by the data acquisition vehicle 301. The upper segment of the figure represents, as an example, wheel speed from the behavioral information transmitted to the CAN 105, and the lower segment represents, as an example, piston speed from the instantaneous values ​​output by the suspension acceleration sensor 302. The horizontal axis represents time.

[0031] like Figure 3 As shown in (B), the wheel speed transmitted to CAN 105 corresponds to the piston speed output by the suspension acceleration sensor 302. For example, the wheel speed shown in window 310 corresponds to the instantaneous value 311 output by the suspension acceleration sensor 302.

[0032] Not limited to the data acquisition vehicle 301, a simulation device that mimics the data acquisition vehicle 301 can also be used to acquire behavior information representing the vehicle's behavior and sensor data related to the vehicle's suspension. In this case, the acquisition device can also display the behavior information and the acquired sensor data on a display unit.

[0033] Figure 4 This is a block diagram of the learning device 401. The learning device 401 includes a control unit 402, a learning unit 403, a dataset storage unit 404, and a weight parameter storage unit 405. Data is stored in the data storage unit 404. Figure 3 The dataset acquired by the data acquisition vehicle 301 is shown. Specifically, the dataset stored in the dataset storage unit 305 of the data acquisition vehicle 301 is transmitted to the dataset storage unit 404 of the learning device 401 via a network or the like.

[0034] According to the instruction of the control unit 402 to start learning, the learning unit 403 reads the dataset from the dataset storage unit 404 and learns the correlation between the behavior information transmitted to the CAN 105 and the sensor data from the suspension acceleration sensor 302.

[0035] Learning Department 403, for example, by reference Figure 2 The same neural network configuration is described. Behavioral information stored in the dataset storage unit 404 and sensor data from the suspension acceleration sensor 302 are used to implement neural network-based machine learning. Specifically, behavioral information, such as that sent to the CAN 105, is set on the elements of the neural network input layer (number of elements i) 201. Figure 3 The discrete values ​​of the data contained in window 310 shown in (B) are sampled, and the instantaneous values ​​311 of the sensor data from the corresponding suspension acceleration sensor 302 are set on the elements of the output layer (number of elements K) 203 of the neural network.

[0036] The relationship between window 310 and the instantaneous value 311 of sensor data is explained. Figure 3 In (B), the width of window 310 is easily determined to be 1 second, but the instantaneous value 311 of the sensor data is set to a specification inferred from the driving history 1 second earlier. Therefore, if the sampling interval is, for example, 20 m seconds, then there are 50 sampling points in window 310 (=1 second ÷ 20 m seconds). The combination of this 50 data sets and the instantaneous value 311 of the sensor data is defined as the learning dataset. The dataset is multiplied by n by sliding window 310 and the instantaneous value 311 of the sensor data every 20 m seconds or every 20 m seconds × m (m: an integer) in the time direction.

[0037] Weight parameters, representing the results of machine learning performed by the learning unit 403, which is composed of a neural network, are stored in the weight parameter storage unit 405. These weight parameters represent the correlation, or correspondence, between the behavioral information transmitted to the CAN 105 and the sensor data from the suspension acceleration sensor 302. Then, the weight parameters stored in the weight parameter storage unit 405 are appropriately stored... Figure 1 The weight parameters shown in (B) are stored in the storage unit 103. Then, as already explained, in a vehicle 101 where no acceleration sensor is installed on the suspension 106, the spring speed and piston speed are estimated to control the suspension.

[0038] Figure 5 It is a flowchart representing the process of acquiring data and learning from it. In step S501, as referred to Figure 3 As described in (A), data is initially acquired by the vehicle 301, which has a suspension acceleration sensor 302 installed on the suspension 106.

[0039] In step S502, while the data acquisition vehicle 301 is moving, it acquires behavior information transmitted to the CAN 105 and sensor data from the suspension acceleration sensor 302, and stores each piece of information as a dataset in the dataset storage unit 305. This is the data acquisition stage, in which the data acquisition vehicle 301 is tested on the road while acquiring a dataset for learning the neural network.

[0040] In step S503, it is determined whether the dataset is an anomaly, such as exceeding the expected range or being fixed at a certain value. This determination is made by the control unit 303 of the data acquisition vehicle 301. If the dataset is an anomaly, it is discarded and not stored in the dataset storage unit 305. Then, the process returns to step S502. If the dataset is not an anomaly, the process proceeds to step S504.

[0041] In step S504, the dataset stored in the dataset storage unit 305 is transferred to the dataset storage unit 404 of the learning device 401.

[0042] In the next step S505, as referenced Figure 4 The learning device 401 performs machine learning on the correlation between the behavioral information transmitted to the CAN 105 and the sensor data from the suspension acceleration sensor 302. This is the learning phase, which uses the dataset acquired in the data acquisition phase to implement the learning of the neural network.

[0043] Since learning requires computing resources, the learning device 401 is a server device or a personal computer, etc., and a window 310 (see reference) is provided in the input element of the neural network constituting the learning unit 403. Figure 3 The time-series data within (B) is used to set the instantaneous values ​​from the suspension acceleration sensor 302 in the output element of the neural network. In this state, the weight parameters of the neural network that minimize the output error are determined, for example, by learning through error backpropagation.

[0044] In the next step S506, the learning device 401 compares the learning result with, for example, the expected value, and determines whether the error between the learning result and the expected value is small enough. Machine learning is performed until the error is sufficiently small. If the error is determined to be small enough, the process proceeds to step S507, where the weight parameters of the learning result are reflected in... Figure 1 In vehicle 101 shown in (A), specifically, the weight parameters of the learning results are transmitted to the weight parameter storage unit 103.

[0045] Then, in vehicle 101, suspension control can be performed. That is, in vehicle 101, the behavior information from CAN 105 is input to the input element of a neural network with learned weight parameters, and the value of the output element is derived by calculating the value of the input element and the weight parameters, thereby obtaining the instantaneous value equivalent to that of the acceleration sensor.

[0046] Figure 6 This is an example of a display screen 601 showing behavioral information and sensor data. The display screen 601 is in Figure 3 The data displayed in the display unit 307 shown is in Figure 5 The steps shown in step S502 are illustrated.

[0047] The data acquisition vehicle 301 travels on various road surfaces within a limited time to collect diverse data. Therefore, it is desirable to be able to confirm in real time, appropriately, whether valid data has been acquired, via the display screen 601.

[0048] exist Figure 6 The text describes an example of displaying behavioral information sent to the CAN 105, specifically wheel speed 602 and front / rear acceleration 603, in the upper section of display screen 601. Then, it shows an example of displaying time-series data of the piston speed 604 from the suspension acceleration sensor 302 in the lower section of display screen 601. (The text repeats itself here.) Figure 5 As described in step S503, the control unit 303 of the data acquisition vehicle 301 determines whether the dataset contains an abnormal value, such as exceeding the expected range or being fixed at a certain value. If it is an abnormal value, the control unit 303 displays identification information indicating the abnormal value on the display screen 601, or notifies the user by sound or other means through a notification device (not shown in the illustration).

[0049] Additionally, the dataset is output as text to be fed into the neural network. Button 605 is a button that indicates the start of text output. When button 605 is indicated, the control unit 303 of the data acquisition vehicle 301 outputs the dataset as a file, for example, in CSV format. Button 605 can be a physical button, or it can be selected via touch operation, or by indicating a device, etc.

[0050] Figure 7 This is a diagram representing a variation of the vehicle behavior prediction unit 107. This variation 1 represents an example of a vehicle behavior prediction unit 107 composed of a neural network. like Figure 7As shown, the input layer 701 consists of input layer 701a, input layer 701b, input layer 701c, input layer 701d, and input layer 701d, which inputs time-series data of the wheel speed of the right front wheel, the front-rear acceleration of the vehicle 101, the left-right acceleration of the vehicle 101, and the yaw rate of the vehicle 101. Hidden layer 702 and output layer 703... Figure 2 Since they are the same, their description is omitted.

[0051] In this variation 1, as time series data, the input consists of four physical quantities: the wheel speed of the front right wheel, the front-rear acceleration of vehicle 101, the left-right acceleration of vehicle 101, and the yaw rate of vehicle 101. Therefore, the input elements of the neural network are i = 200 points (i = 50 points × 4 physical quantities).

[0052] For ease of explanation, Figure 7 A neural network for controlling the suspension 106 mounted on the front right wheel of vehicle 101 is shown, but the vehicle behavior prediction unit 107 corresponds to a neural network with the same structure for all four wheels of vehicle 101.

[0053] Figure 8 (A), (B), and (C) are diagrams representing variations 2, 3, and 4 of the vehicle behavior prediction unit 107. like Figure 8 As shown in (A), the vehicle behavior prediction unit 801 inputs wheel speed, other behavior information, and suspension control values ​​output by the control value calculation unit 108 located after the vehicle behavior prediction unit 801, based on communication data constantly transmitted to the CAN 105. The control value calculation unit 108 outputs appropriate suspension control values ​​based on vehicle behavior, and the suspension control values ​​also include tendencies related to vehicle behavior. In this regard, by adding suspension control values ​​to the input of the vehicle behavior prediction unit 801, the prediction accuracy of data involved in suspension control, such as sensor data, is improved.

[0054] like Figure 8 As shown in (B), the vehicle behavior estimation unit 802 inputs wheel speed, other behavior information, steering wheel information, and accelerator or brake pedal operation information based on communication data constantly transmitted to the CAN 105. Since the front-rear acceleration of the vehicle 101 can be estimated based on the steering information or pedal operation information, the estimation accuracy of the sensor data involved in suspension control is improved by setting them as inputs to the vehicle behavior estimation unit 802.

[0055] like Figure 8As shown in (C), the vehicle behavior prediction unit 803 inputs wheel speed, other behavior information, and the spring speed or piston speed previously output by the vehicle behavior prediction unit 803, based on communication data constantly transmitted to the CAN 105. The spring speed or piston speed is data related to suspension control, but it represents continuous instantaneous value changes and is influenced by previous instantaneous values. Therefore, in addition to the communication data input from the CAN 105, by referring to the instantaneous values ​​immediately following the previous output of the vehicle behavior prediction unit 803, the prediction accuracy of the sensor data related to suspension control is improved.

[0056] Figure 9 (A) and (B) are diagrams representing variations of the vehicle behavior prediction unit 107, example 5. like Figure 9 As shown in (A), the behavior information from CAN 105 is not directly input to the vehicle behavior prediction unit 901, but is instead input to the preprocessing unit 902. The preprocessing unit 902 aggregates multiple behavior information into a single behavior information. For example, it aggregates wheel speed with other behavior information and inputs the aggregated wheel speed to the vehicle behavior prediction unit 901. As described later, the number of input terminals of the vehicle behavior prediction unit 901 can be reduced through the preprocessing unit 902. When the vehicle behavior prediction unit 901 is implemented using a neural network, the computational resources required by the onboard control unit 102 can be reduced due to the large amount of matrix calculations performed.

[0057] Figure 9 (B) means Figure 9 A diagram showing an example of the preprocessing unit 902, as shown in (A). The preprocessing unit 902 includes a differentiator 903 and an adder / subtractor 904. The wheel speed from CAN 105 is converted into wheel acceleration by the differentiator 903. Then, the adder / subtractor 904 performs addition and subtraction on the wheel acceleration and the front-to-back acceleration between the acceleration values. This allows the generation of a sum of the wheel acceleration and the front-to-back acceleration, eliminating the need to input the front-to-back acceleration itself into the neural network. This eliminates the need for an input layer for the front-to-back acceleration, resulting in a reduction in the size of the neural network. Figure 9 In the example shown in (B), the input layer 905 of the neural network consists of an input layer 905a for inputting wheel acceleration, an input layer 905b for inputting yaw rate, and an input layer 905c for inputting the suspension control value of the previous generation.

[0058] exist Figure 9 In (B), an example of using forward and backward acceleration to reduce the circuit size of a neural network is shown, but in addition, left and right acceleration, yaw rate or other behavioral information can also be used for preprocessing.

[0059] Figure 10 This is a diagram representing a variation of the vehicle behavior prediction unit 107, specifically example 6. like Figure 10 As shown, the vehicle behavior prediction unit 1001 inputs not only wheel speed and other behavior information from the communication data constantly transmitted to the CAN 105, but also behavior information based on image data from stereo cameras and other devices increasingly prevalent in autonomous driving or autonomous driving assistance systems. The image data-based behavior information includes, for example, information on the vertical, forward, backward, and horizontal movements of the vehicle 101 detected through image comparison in each frame. This improves the accuracy of vehicle behavior prediction and allows for the prediction of sensor data related to suspension control.

[0060] According to Embodiment 1, by utilizing the behavior information transmitted to the CAN 105, suspension control can be performed in a vehicle where no acceleration sensor is installed on the suspension 106. Furthermore, the dataset related to the behavior information and suspension control is a dataset acquired while driving the vehicle 101. Based on a dataset containing the overall rigidity characteristics of the vehicle, which consists of multiple components, weight parameters are determined through machine learning. Therefore, the accuracy of suspension control corresponding to the behavior of the vehicle 101 is improved.

[0061] [Implementation Method 2] Figure 11 This is a block diagram of the suspension control system. The suspension control system consists of a server device 1101, a data acquisition vehicle 1102, and a driving vehicle 1103.

[0062] Server device 1101 includes a transceiver interface 1104, a control unit 1105, a learning unit 1107, a dataset storage unit 1106, and a weight parameter storage unit 1108. The structure of server device 1101, except for the transceiver interface 1104, is the same as that described in Embodiment 1. Figure 4 The learning device 401 described is the same.

[0063] The data acquisition vehicle 1102 includes a suspension 106, a suspension acceleration sensor 302, a control unit 303, a data collection unit 304, a dataset storage unit 305, a data reading unit 306, a display unit 307, and a transceiver interface 1109. Furthermore, the control unit 303 includes a control value calculation unit 108. The structure of the data acquisition vehicle 1102, except for the transceiver interface 1109, is the same as that described in Embodiment 1. Figure 3 The data acquisition method described is the same as that used for vehicle 301.

[0064] The vehicle 1103 includes a control unit 102, a weight parameter storage unit 103, a data reading unit 104, a CAN bus 105, a suspension 106, a transceiver interface 1110, and a data writing unit 1111. Furthermore, the control unit 102 includes a vehicle behavior prediction unit 107 and a control value calculation unit 108. The vehicle 1103 is equivalent to the one described in Embodiment 1. Figure 1 Vehicle 101 is described.

[0065] While driving, the data acquisition vehicle 1102 acquires behavior information transmitted to the CAN 105 and sensor data from the suspension acceleration sensor 302, and stores the corresponding information as a dataset in the dataset storage unit 305. Then, the dataset stored in the dataset storage unit 305 is transmitted to the server device 1101 via the transceiver interface 1109.

[0066] The dataset stored in the dataset storage unit 305 via the data reading unit 306 can also be displayed visually at any time on the display unit 307 of the data acquisition vehicle 1102. Then, as referred to Figure 6 As explained, the dataset can be output as a file in CSV format by pressing button 605.

[0067] Server device 1101 receives a dataset from data acquisition vehicle 1102 via transceiver interface 1104 from a wide area network. Then, control unit 1105 stores the received dataset in dataset storage unit 1106. Learning unit 1107 reads the dataset from dataset storage unit 1106 according to the learning start instruction from control unit 1105, and uses the dataset to represent the behavior information and sensor data of data acquisition vehicle 1102, performing machine learning on the correlation between the behavior information transmitted to CAN 105 and the sensor data from suspension acceleration sensor 302. The weight parameters of the learned results are stored in weight parameter storage unit 1108.

[0068] The vehicle 1103 acquires weight parameters stored in the weight parameter storage unit 1108 of the server device 1101 via the transceiver interface 1110 through a wide area network. The data writing unit 1111 writes the acquired weight parameters into the weight parameter storage unit 103. The suspension control device, i.e., the control unit 102, includes a vehicle behavior prediction unit 107. The vehicle behavior prediction unit 107 is composed of a neural network and reads the weight parameters through the data reading unit 104, reflecting the learning results of the learning unit 1107 of the server device 1101. The control unit 102 controls the suspension 106 based on the reflected learning results.

[0069] According to this embodiment, in addition to achieving the effects described in Embodiment 1, if the vehicles are of the same model, multiple driving vehicles 1103 can reflect the weight parameters acquired by one data acquisition vehicle 1102. Furthermore, as the data acquisition vehicle 1102 continues to acquire datasets, the server device 1101 continuously performs machine learning on the acquired datasets, thereby updating the weight parameters of the driving vehicles 1103 and enabling highly accurate suspension control that reflects vehicle characteristics.

[0070] The following effects can be obtained by implementing the methods described above. (1) The suspension control device 111 includes: a parameter storage unit 103, which stores parameters pre-determined through machine learning that represent the correspondence between behavioral information and sensor data, wherein the behavioral information represents the behavior of vehicle 101 and the sensor data pertains to the suspension 106 of vehicle 101; a vehicle behavior estimation unit 107, which inputs behavioral information from communication data transmitted within vehicle 101 into the vehicle behavior estimation unit 107, which estimates the sensor data pertaining to the suspension 106 of vehicle 101 based on the behavioral information and the parameters stored in the parameter storage unit 103; and a control value calculation unit 108, which calculates suspension control values ​​for controlling the suspension 106 based on the estimated sensor data. This improves the accuracy of suspension control corresponding to vehicle behavior.

[0071] (2) The suspension control method is a suspension control device 111 that controls the suspension 106 of vehicle 101. In this method, parameters representing the correspondence between behavioral information and sensor data, pre-determined through machine learning, are stored. The behavioral information represents the behavior of vehicle 101, and the sensor data pertains to the suspension 106 of vehicle 101. Behavioral information is input into communication data transmitted within vehicle 101. Based on the behavioral information and parameters, the sensor data involved in the suspension 106 of vehicle 101 is inferred. Based on the inferred sensor data, a suspension control value for controlling the suspension 106 is calculated. This improves the accuracy of suspension control corresponding to the vehicle's behavior.

[0072] (3) The suspension control system is a suspension control system equipped with a server device 1101 and a suspension control device 102 that controls the suspension 106 of the vehicle 1103. The server device 1101 acquires behavior information representing the behavior of the vehicle 1103 and sensor data related to the suspension 106 of the vehicle 1103. It determines parameters representing the correspondence between the behavior information and the sensor data through machine learning. The suspension control device 102 acquires parameters from the server device 1101, inputs the behavior information from the communication data transmitted within the vehicle 1103, and infers the sensor data related to the suspension 106 of the vehicle 1103 based on the behavior information and parameters. Based on the inferred sensor data, it controls the suspension 106. This improves the accuracy of suspension control corresponding to the vehicle's behavior.

[0073] This invention is not limited to the embodiments described above. Other embodiments that can be considered within the scope of the technical concept of this invention, as long as they do not impair the characteristics of this invention, are also included within the scope of this invention. Additionally, it may be a structure that combines the above embodiments and multiple variations. Label Explanation

[0074] 101…Vehicle, 102, 303, 402…Control unit, 103, 401, 405, 1108…Weight parameter storage unit, 104…Data reading unit, 105…CAN, 106…Suspension, 107, 801, 802, 803, 901, 1001…Vehicle behavior prediction unit, 108…Control value calculation unit, 111…Suspension control device, 201, 905…Input layer, 202…Hidden layer, 203…Output layer, 301…Data acquisition vehicle, 302…Acceleration sensor for suspension, 30… 4… Data collection unit, 305, 404… Data set storage unit, 306… Data reading unit, 307… Display unit, 401… Learning device, 403, 1107… Learning unit, 605… Button, 902… Preprocessing unit, 903… Differentiator, 904… Adder / subtractor, 1101… Server device, 1102… Data acquisition vehicle, 1103… Driving vehicle, 1104, 1109, 1110… Transceiver interface, 1105… Control unit, 1106… Data set storage unit, 1111… Data writing unit.

Claims

1. A suspension control device, characterized in that, include: The parameter storage unit stores parameters that are pre-determined through machine learning and represent the correspondence between behavioral information and sensor data. The behavioral information represents the behavior of the vehicle and includes at least the wheel speed and front-rear acceleration of the vehicle. The sensor data corresponds to at least one of the piston speed of each suspension involving multiple wheels of the vehicle and the acceleration on the spring. The vehicle behavior estimation unit inputs the behavior information from the communication data transmitted within the vehicle into the vehicle behavior estimation unit. The vehicle behavior estimation unit includes a neural network that, based on the behavior information and the parameters stored in the parameter storage unit, infers the sensor data related to the suspension of the vehicle for each suspension. as well as The control value calculation unit calculates the suspension control value for controlling the suspension based on the inferred sensor data.

2. The suspension control device as described in claim 1, characterized in that, The vehicle behavior prediction unit includes a plurality of neural networks configured for each wheel, each of the neural networks using the behavior information and the parameters to predict the sensor data of the suspension installed on the wheel corresponding to the neural network.

3. The suspension control device as described in claim 1 or 2, characterized in that, The communication data is data transmitted to the vehicle's onboard network.

4. The suspension control device as described in claim 1 or 2, characterized in that, The behavioral information, in addition to the vehicle's wheel speed and longitudinal acceleration, also includes at least one of the vehicle's lateral acceleration and yaw rate.

5. The suspension control device as described in claim 4, characterized in that, The behavioral information also includes sensor data that the vehicle behavior estimation unit has previously inferred.

6. The suspension control device as described in claim 4, characterized in that, It includes a preprocessing unit that aggregates multiple behavioral information input to the vehicle behavior prediction unit into a single behavioral information.

7. The suspension control device as described in claim 4, characterized in that, The behavioral information also includes information representing the behavior of the vehicle detected based on image data from a camera mounted on the vehicle.

8. A suspension control system, characterized in that, Include: The suspension control device as described in claim 1 or 2; and The acquisition device acquires behavioral information representing the behavior of the vehicle, as well as sensor data related to the vehicle's suspension. The acquisition device displays the behavioral information and the acquired sensor data.

9. A suspension control method, wherein the suspension control method is a suspension control device for controlling the suspension of a vehicle, characterized in that, The system stores parameters pre-determined through machine learning, representing the correspondence between behavioral information and sensor data. The behavioral information, indicating the vehicle's behavior, includes at least the vehicle's wheel speeds and front-to-rear accelerations. The sensor data corresponds to at least one of the piston speeds and spring accelerations of the suspensions involving the vehicle's multiple wheels. The behavioral information from the communication data transmitted within the vehicle is input into a neural network. Based on the behavioral information and the parameters, the neural network is used to infer the sensor data related to each suspension of the vehicle. Based on the inferred sensor data, the suspension control value for controlling the suspension is calculated.

10. The suspension control method as described in claim 9, characterized in that, Multiple neural networks are configured for each of the aforementioned wheels. Using each of the neural networks, the behavioral information and the parameters are used to infer the sensor data of the suspension installed on the wheel corresponding to the neural network.

11. A suspension control system, comprising a server device and a suspension control device for controlling the suspension of a vehicle, characterized in that... The server device acquires behavioral information and sensor data, and uses machine learning to determine parameters representing the correspondence between the behavioral information and the sensor data. The behavioral information represents the behavior of the vehicle and includes at least the wheel speed and front-rear acceleration of the vehicle. The sensor data corresponds to at least one of the piston speed and spring acceleration of each suspension involving multiple wheels of the vehicle. The suspension control device obtains the parameters through the server device, inputs the behavioral information from the communication data transmitted in the vehicle into the neural network, and, based on the behavioral information and the parameters, uses the neural network to infer the sensor data involved in the suspension of the vehicle for each suspension, and controls the suspension based on the inferred sensor data.

12. The suspension control system as described in claim 11, characterized in that, The server device includes a plurality of neural networks configured for each of the wheels, each of the neural networks performing machine learning on parameters representing the correspondence between the behavioral information and the sensor data. The suspension control device includes a plurality of neural networks configured for each of the wheels, each of the neural networks using parameters corresponding to the behavioral information to infer the sensor data of the suspension installed on the wheel corresponding to the neural network.

13. The suspension control system as described in claim 11 or 12, characterized in that, It includes an acquisition device that acquires behavioral information representing the behavior of the vehicle and sensor data related to the vehicle's suspension. The acquisition device displays the behavioral information and the acquired sensor data.

Images