Suspension system and estimation method

Abstract

A suspension system that is installed in a vehicle and estimates the state of a suspension installed in the vehicle, said suspension system comprising: a storage unit in which a plurality of neural networks are stored; an acquisition unit that acquires weight-related data, which is data pertaining to the weight of the vehicle cargo; and a vehicle state estimation unit that estimates the state of the suspension by inputting data pertaining to the travel of the vehicle to any neural network stored in the storage unit that corresponds to the weight-related data.

Description

Suspension system, estimation method

[0001] The present invention relates to a suspension system and an estimation method.

[0002] Suspension systems are widely used to absorb vibrations and shocks from the road surface. Patent Document 1 discloses a configuration for adjusting the noise of a vehicle's suspension and exhaust system in accordance with the characteristics of the occupant monitored via a camera and a microphone.

[0003] US Patent Application Publication No. 2021 / 0114603

[0004] The invention described in Patent Document 1 leaves room for improvement in estimating the state of the suspension.

[0005] A suspension system according to a first aspect of the present invention is a suspension system mounted on a vehicle and estimating a state of a suspension mounted on the vehicle, comprising: a memory unit in which a plurality of neural networks are stored; an acquisition unit that acquires weight-related data that is data related to the weight of a load on the vehicle; and a vehicle state estimation unit that estimates the state of the suspension by inputting data related to the running of the vehicle to any of the neural networks stored in the memory that corresponds to the weight-related data.A second aspect of the present invention is an estimation method used by a computer mounted on a vehicle and having a memory unit that stores a plurality of neural networks to estimate the state of a suspension mounted on the vehicle, the estimation method comprising: acquiring weight-related data that is data related to the weight of a load on the vehicle based on outputs of sensors mounted on the vehicle; and estimating the state of the suspension by inputting data related to the running of the vehicle to any of the neural networks stored in the memory that corresponds to the weight-related data.

[0006] According to the present invention, the state of the suspension can be estimated with high accuracy.

[0007] 1. Diagram of a vehicle configuration according to a first embodiment. 2. Diagram showing an example of a correspondence table. 3. Schematic diagram of a neural network. 4. Flowchart showing the processing of the reading unit. 5. Diagram showing an example of the operation of the suspension ECU. 6. Diagram showing a display on an instrument panel. 7. Hardware configuration diagram of a suspension ECU. 8. Diagram of a vehicle configuration according to a first modification. 9. Diagram showing a correspondence table in a second modification. 10. Diagram showing an example of a case where the driver inputs the position of a load using an instrument panel. 11. Diagram showing a display screen of a mobile terminal. 12. Schematic diagram showing inputs and outputs of a sensor processing unit. 13. Diagram of a neural network update system.

[0008] -First Embodiment- A first embodiment of a suspension system will be described below with reference to Figs.

[0009] 1 is a configuration diagram of a vehicle 9 including a suspension ECU 107, which is a suspension control device. The vehicle 9 includes a weight measurement unit 103, an instrument panel 105, the suspension ECU 107, a steering ECU 108, an in-vehicle network 109, and a semi-active suspension 201. The in-vehicle network 109 is a known communication network, such as IEEE 802.3 or a Controller Area Network. The suspension ECU 107 and the steering ECU 108 are ECUs, i.e., electronic control units.

[0010] The suspension ECU 107 includes a vehicle state estimation unit 202, a control value calculation unit 203, a neural network storage unit 204, a readout unit 205, an acquisition unit 206, and an output unit 208. A correspondence table 2051 is stored in the readout unit 205. The weight measurement unit 103 is a weight sensor, such as a strain gauge and an amplifier, that measures the weight of the load on the vehicle 9. The load on the vehicle 9 includes the occupants of the vehicle 9, including the driver, and luggage carried on the vehicle 9. Data related to the weight of the load on the vehicle 9 measured by the weight measurement unit 103 (hereinafter referred to as "weight-related data") is read by the acquisition unit 206. The acquisition unit 206 outputs the weight-related data measured by the weight measurement unit 103 to the readout unit 205 and the output unit 208. Note that although the weight-related data in this embodiment is a total value of the weight of the load, the concept of weight-related data can also include the distribution of weight.

[0011] The reading unit 205 refers to the correspondence table 2051 to identify the neural network corresponding to the weight-related data output by the acquisition unit 206, i.e., the weight of the load. The reading unit 205 then reads the identified neural network from the neural network storage unit 204 and outputs it to the vehicle state estimation unit 202. The reading unit 205 operates using various signals received via the in-vehicle network 109 as triggers. For example, the reading unit 205 uses a door closing signal from a door sensor, an engine or motor start signal, and a signal indicating that a seat belt is fastened as triggers. Furthermore, the reading unit 205 may also use an explicit instruction from an occupant of the vehicle 9, such as pressing an operation command button, as triggers.

[0012] Vehicle state estimation unit 202 performs inference using the neural network output by readout unit 205. Time-series physical quantities related to vehicle 9, such as the speed and acceleration of vehicle 9, are input to this neural network. Vehicle state estimation unit 202 outputs the inference result to control value calculation unit 203. Control value calculation unit 203 calculates a control value for controlling the damping force of semi-active suspension 201 using a known method, based on the physical quantities input from vehicle state estimation unit 202.

[0013] The semi-active suspension 201 controls a suspension (not shown) mounted on the vehicle based on the output of the suspension ECU 107. A plurality of trained neural networks are stored in the neural network storage unit 204. The output unit 208 outputs the data on the weight of the load output by the acquisition unit 206 to the instrument panel 105. The instrument panel 105 is a display device mounted on the vehicle 9, and outputs the information output by the output unit 208.

[0014] 1, the suspension ECU 107 is illustrated as a single hardware device, but the suspension ECU 107 may be realized by a plurality of devices. Therefore, the suspension ECU 107 can also be called a suspension system.

[0015] 2 is a diagram showing an example of the correspondence table 2051 stored in the reading unit 205. The correspondence table 2051 shows the correspondence between the weight of the load on the vehicle 9 and the neural network used by the vehicle state estimation unit 202. Specifically, it shows that "N1" is used when the load is less than 100 kg, "N2" is used when the load is 100 kg or more but less than 300 kg, and "N3" is used when the load is 300 kg or more. The neural networks N1 to N3 differ in at least one of the configuration and parameters of the neural network.

[0016] 3 is a schematic diagram of a neural network N stored in the neural network storage unit 204. The neural network N shown in FIG. 3 represents, for example, N1 to N3 in FIG.

[0017] The neural network N includes an input layer N-IN, a hidden layer N-H, and an output layer N-OUT. The input layer N-IN has a plurality of nodes, and time series values ​​of one or more physical quantities are input. Inputs to the input layer N-IN include, for example, the vehicle speed, vehicle acceleration, and yaw rate. The range of the time series is predetermined. However, the range of the time series may differ for each physical quantity.

[0018] Although only one hidden layer N-H is shown in Figure 3 for convenience of drawing, two or more layers are also possible. The number of nodes in each layer of the hidden layer N-H is arbitrary, and depending on the configuration of the neural network, there may be more or fewer than the input layer N-IN. The hidden layer N-H may or may not include a fully connected layer. The configuration of the neural network N including the hidden layer N-H is not limited.

[0019] The two nodes in the output layer N-OUT represent the instantaneous values ​​of two physical quantities. These two physical quantities are the vertical velocity of the upper part of the spring that constitutes the suspension (hereinafter also referred to as the "sprung part velocity"), and the relative velocity of the piston that constitutes the suspension with respect to the spring (hereinafter also referred to as the "piston relative velocity"). However, the two physical quantities in the output layer N-OUT may also be the sprung part velocity and the vertical velocity of the lower part of the spring that constitutes the suspension (hereinafter also referred to as the "unsprung part velocity").

[0020] The configuration of the neural network N is determined in advance, and the parameters, i.e., the weights between nodes, have been trained using previously collected data. The previously collected data are multiple pairs of input and output values ​​of the neural network N. A known method, such as backpropagation, can be used to train the parameters of the neural network N. Note that the configuration of the neural network N may be changed during the training process.

[0021] Figure 4 is a schematic diagram showing the input and output to the neural network N. In the example shown in Figure 4, the input to the neural network N is time-series data of vehicle speed, and only the vertical velocity of the sprung part is shown among the outputs. At time t2, the vehicle speed from time t1 to time t2 is input to the neural network N. The neural network N infers and outputs the vertical velocity of the sprung part at time t2. If the period from time t1 to time t2 is 1 second and the sampling period of the vehicle speed is 10 ms, 100 vehicle speeds are input to the neural network N.

[0022] 5 is a flowchart showing the processing of the reading unit 205. First, in step S301, the reading unit 205 determines whether or not a trigger has been received. Examples of triggers include a door closing signal, an engine or motor start signal, and a signal indicating that a seat belt has been fastened. If the reading unit 205 determines that a trigger has been received, the process proceeds to step S302, and if the reading unit 205 determines that a trigger has not been received, the process remains in step S301.

[0023] In step S302, the reading unit 205 recognizes the weight of the load on the vehicle 9. For example, if there are multiple weight measurement units 103, the weights output by each weight measurement unit 103 are added together. In the following step S303, the reading unit 205 refers to the correspondence table 2051 and determines the neural network corresponding to the weight recognized in step S302 as the read target. In the following step S304, the reading unit 205 reads the target determined in step S303 from the neural network storage unit 204 and outputs it to the vehicle state estimation unit 202. Thereafter, the vehicle state estimation unit 202 performs inference using the neural network output in this step.

[0024] In the next step S305, the reading unit 205 determines whether the ignition switch is set to OFF. If the reading unit 205 determines that the ignition switch is set to OFF, the processing shown in Fig. 5 is terminated. If the reading unit 205 determines that the ignition switch is not set to OFF, the processing returns to step S301.

[0025] FIG. 6 is a diagram showing an example of the operation of the suspension ECU 107 from time t10 to time t15. In FIG. 6, time elapses from left to right. From top to bottom, FIG. 6 shows the vehicle speed, trigger, weight condition, and neural network read into the vehicle state estimation unit 202 of the vehicle 9. At time t10, the vehicle 9 is stopped, and the weight of the load on the vehicle 9 is less than 100 kg. At time t11, the trigger is turned on due to reasons such as the ignition switch being turned on (S301: YES). Therefore, the reading unit 205 reads "N1," which corresponds to a weight of less than 100 kg, from the correspondence table 2051 and outputs it to the vehicle state estimation unit 202.

[0026] Thereafter, the vehicle 9 starts traveling, and at time t12, the vehicle speed becomes zero again. At this time, for example, the vehicle 9 is carrying multiple occupants, and the weight of the cargo exceeds 300 kg. However, since a trigger has not yet occurred, there is no change in the neural network of the vehicle state estimation unit 202 at this timing. At time t13, a trigger is generated by closing the door, for example, and the reading unit 205 reads "N3" in the correspondence table 2051, which corresponds to a weight of 300 kg or more. Therefore, in FIG. 6, the neural network is set to "N3" from time t13 onwards. After this, no trigger is generated, and the vehicle 9 continues traveling until time t15.

[0027] 7 is a diagram showing a display on the instrument panel 105 based on the output of the output unit 208. In this figure, the positions where the weight is detected by the weight measuring unit 103 are indicated by hatched circles. Note that in this figure, all positions where the weight is detected are displayed uniformly, but the display may be different depending on the magnitude of the weight. For example, the color, shape, and density of the icon may be changed depending on the magnitude of the weight.

[0028] 8 is a hardware configuration diagram of the suspension ECU 107. The suspension ECU 107 includes a CPU 41, which is a central processing unit, a ROM 42, which is a read-only storage device, a RAM 43, which is a readable and writable storage device, a storage device 44, which is a non-volatile recording medium, and a communication device 45. The CPU 41 loads a program stored in the ROM 42 into the RAM 43 and executes it, thereby realizing a vehicle state estimation unit 202, a control value calculation unit 203, a reading unit 205, an acquisition unit 206, and an output unit 208.

[0029] The suspension ECU 107 may be realized by a field programmable gate array (FPGA), which is a rewritable logic circuit, or an application specific integrated circuit (ASIC), which is an integrated circuit for a specific application, instead of the combination of the CPU 41, ROM 42, and RAM 43. Furthermore, the suspension ECU 107 may be realized by a combination of different configurations, for example, a combination of the CPU 41, ROM 42, RAM 43, and an FPGA, instead of the combination of the CPU 41, ROM 42, and RAM 43.

[0030] The storage device 44 is, for example, a hard disk drive, and realizes the neural network storage unit 204. The communication device 45 is a communication interface with the in-vehicle network 109. The communication device 45 is used by the vehicle state estimation unit 202, the control value calculation unit 203, the reading unit 205, the acquisition unit 206, and the output unit 208, which require input and output of data from and to the outside of the suspension ECU 107.

[0031] According to the first embodiment described above, the following advantageous effects can be obtained. (1) The suspension ECU 107 is mounted on the vehicle 9 and estimates the state of the semi-active suspension 201 mounted on the vehicle 9. The suspension ECU 107 includes a neural network storage unit 204 in which a plurality of neural networks are stored, an acquisition unit 206 that acquires weight-related data that is data related to the weight of a load on the vehicle 9, and a vehicle state estimation unit 202 that estimates the state of the suspension by inputting data related to the running of the vehicle into one of the neural networks stored in the neural network storage unit 204 that corresponds to the weight-related data. Therefore, the suspension ECU 107 uses different neural networks depending on the weight of the vehicle 9, and can estimate the state of the suspension with high accuracy.

[0032] (2) The suspension ECU 107 includes a readout unit 205 that reads out a neural network corresponding to the weight-related data from the neural network storage unit 204 and outputs the neural network to the vehicle state estimation unit 202. The vehicle state estimation unit 202 estimates the state of the suspension using the neural network output by the readout unit 205. Therefore, only data necessary for calculation is output to the vehicle state estimation unit 202, so that the vehicle state estimation unit 202 does not need to include an excessive amount of memory.

[0033] In a conventional suspension system, an acceleration sensor is installed on each of the four semi-active suspensions mounted on the vehicle body to measure the vertical velocity of the sprung mass and the relative velocity of the piston stroke, and the suspension control value is calculated based on the measurement results. This configuration increases the installation and assembly steps of the acceleration sensor, and requires additional brackets and harnesses, increasing the cost of the sensor itself and other components. In contrast, in this embodiment, the vehicle state estimation unit 202 estimates physical values ​​related to suspension control based on specific information, such as wheel speed and vehicle behavior indicators such as longitudinal acceleration, lateral acceleration, and yaw rate, from the data constantly transmitted to the CAN 109. The estimation results are then transferred to the downstream control value calculation unit 203, eliminating the need for an acceleration sensor and achieving lower costs.

[0034] (3) Weight-related data is the weight of the cargo loaded on the vehicle.

[0035] (4) As shown in Fig. 6, the reading unit 205 reads out one of the neural networks from the neural network storage unit 204 based on the weight-related data each time it detects the closing of a door of the vehicle 9, the fastening of a seat belt in the vehicle 9, or an instruction from an occupant of the vehicle 9. This allows the suspension ECU 107 to reread the neural network at an appropriate timing.

[0036] (5) The output unit 208 outputs the weight-related data acquired by the acquisition unit 206 to the instrument panel 105, which is a display device provided in the vehicle 9. Therefore, the occupants of the vehicle 9 can confirm the weight of the interior of the vehicle 9 recognized by the suspension ECU 107.

[0037] (Variation 1) FIG. 9 is a configuration diagram of vehicle 9 in Variation 2. This variation is significantly different from the first embodiment in that correspondence table 2051 is stored in vehicle state estimation unit 202. In this variation, reading unit 205 reads all of the neural networks stored in neural network storage unit 204 to vehicle state estimation unit 202. Also, in this variation, information on the weight of the loaded object acquired by acquisition unit 206 is output to vehicle state estimation unit 202 rather than reading unit 205. In addition to the processing in the first embodiment, vehicle state estimation unit 202 further selects a neural network, which was performed by reading unit 205 in the first embodiment.

[0038] According to this modification 1, the following advantageous effects can be obtained: (6) A plurality of neural networks are loaded into the vehicle state estimation unit 202. The vehicle state estimation unit 202 selects a neural network corresponding to the weight-related data from the plurality of loaded neural networks, and estimates the state of the suspension.

[0039] (7) The vehicle state estimation unit 202 reselects the neural network corresponding to the weight-related data each time it detects any of the following: the door of the vehicle 9 being closed, the seat belt being fastened in the vehicle 9, and an instruction from an occupant of the vehicle 9.

[0040] (Modification 2) Fig. 10 is a diagram showing a correspondence table 2051 in Modification 2. In this correspondence table 2051, a neural network is set in advance for each wheel of the vehicle 9 corresponding to weight-related data. For example, when the weight is "less than 100 kg," it is shown that "N1-LF" is used to estimate the state of the suspension for the left front tire, "N1-RF" is used to estimate the state of the suspension for the right front tire, "N1-LB" is used to estimate the state of the suspension for the left rear tire, and "N1-RB" is used to estimate the state of the suspension for the right rear tire.

[0041] The reading unit 205 reads out four neural networks corresponding to the weight of the load on the vehicle 9 from the neural network storage unit 204 and outputs them to the vehicle state estimation unit 202. The vehicle state estimation unit 202 uses a different neural network for each wheel to estimate the state of the suspension corresponding to that wheel. The weight applied to each tire of the vehicle 9 is not necessarily uniform and there is a bias, so this weight bias can be reflected in the estimation of the suspension state.

[0042] According to this modification 2, the following advantageous effects can be obtained. (8) The vehicle state estimation unit 202 estimates the suspension state for each wheel of the vehicle 9. Therefore, the state of each suspension can be estimated with high accuracy. Note that instead of providing a neural network for each wheel, a neural network that outputs eight physical quantities may be used. The eight physical quantities are the sprung section velocity and the piston relative velocity for each of the four wheels.

[0043] (Variation 3) FIG. 11 is a diagram showing a correspondence table 2051 in Variation 3. The weight-related data in this variation is a combination of weight and position, or a combination of weight and distribution. In this correspondence table 2051, a neural network corresponding to each combination of weight and position is determined in advance. The reading unit 205 reads out the neural network corresponding to the combination of weight and position from the neural network storage unit 204 and outputs it to the vehicle state estimation unit 202.

[0044] According to Modification 3, the following advantageous effects can be obtained: (9) The weight-related data is the weight and position of the load carried on the vehicle 9. Therefore, the suspension state can be estimated with high accuracy using an optimal neural network that takes into account not only the weight but also the position of the load.

[0045] (Modification 4) Fig. 12 is a diagram showing a correspondence table 2051 in Modification 4. The weight-related data in this modification is a combination of weight and position, or a combination of weight and distribution. In this correspondence table 2051, a neural network is set in advance for each wheel of the vehicle 9 for each combination of weight and position. This modification is, so to speak, a combination of Modifications 2 and 3.

[0046] The reading unit 205 reads out four neural networks corresponding to combinations of weight and position from the neural network storage unit 204 and outputs them to the vehicle state estimation unit 202. The vehicle state estimation unit 202 uses a different neural network for each wheel to estimate the state of the suspension corresponding to that wheel. The weight applied to each tire of the vehicle 9 is not necessarily uniform and there is a bias, so this weight bias can be reflected in the estimation of the suspension state.

[0047] - Second embodiment - A second embodiment of a suspension system will be described with reference to Figures 13 to 16. In the following description, the same components as those in the first embodiment are designated by the same reference numerals, and differences will be mainly described. Points that are not particularly described are the same as those in the first embodiment. This embodiment differs from the first embodiment mainly in that the state of the vehicle interior is estimated instead of directly measuring weight. However, at least a portion of the instrument panel 105 in this embodiment is a touch panel and also has an input function.

[0048] 13 is a configuration diagram of a vehicle 9A according to the second embodiment. The second embodiment differs from the first embodiment in that it does not include a weight measurement unit 103, but includes a seat belt state detection unit 104, a camera 106, and a communication unit 111. The suspension ECU 107A according to the second embodiment further includes a sensor processing unit 207. The instrument panel 105 according to the second embodiment receives input from the user and outputs the input to the sensor processing unit 207.

[0049] The sensor processing unit 207 receives outputs from the seat belt state detection unit 104, the camera 106, and the communication unit 111. The sensor processing unit 207 estimates the weight of the cargo in the vehicle 9 and outputs it to the acquisition unit 206. The seat belt state detection unit 104 detects the fastening state of the seat belt in each seat in the vehicle 9 and outputs the detected seat data to the sensor processing unit 207. It can be assumed that an occupant is sitting in a seat where the seat belt is fastened.

[0050] The camera 106 captures an image of the interior of the vehicle 9 and outputs the captured image to the sensor processing unit 207. The angle of view of the camera 106 is adjusted to include each seat in the vehicle 9. The communication unit 111 communicates with a mobile terminal 1206 carried by a user, receives data related to occupants input by the user, and outputs the data to the sensor processing unit 207.

[0051] 14 is a diagram showing an example of a case where the driver inputs the location of cargo using the instrument panel 105. The instrument panel 105 is, for example, a touch panel, and the driver inputs the presence or absence of passengers and luggage at each position by touch operation. In this example, passengers are indicated by circles as shown by reference numeral 1204, and luggage is indicated by squares as shown by reference numeral 1205. In this example, it is shown that there are passengers in the driver's seat and passenger seat, and that luggage is present in the rear right position.

[0052] 15 is a diagram showing the display screen of the mobile terminal 1206 that is carried by the user and into which data related to the occupants is input. An occupant setting app is running on the mobile terminal 1206. This app includes a number of people input field 1208 and a weight input field 1209. When the user inputs numerical values ​​into the number of people input field 1208 and the weight input field 1209, the mobile terminal 1206 transmits the numerical data to the communication unit 111. The communication unit 111 outputs the data received from the mobile terminal 1206 to the sensor processing unit 207. Note that this app may be configured to allow the location and weight of luggage to be input.

[0053] FIG. 16 is a schematic diagram showing input and output of the sensor processing unit 207. The sensor processing unit 207 receives data on the position of the loaded object input by the driver, whether the seat belt is fastened, and the image captured by the camera 106. The driver's input and whether the seat belt is fastened can be used to estimate the number of occupants with high accuracy. If the number of occupants can be estimated, the sensor processing unit 207 can calculate the weight by multiplying the estimated number by, for example, the standard weight of an adult. Furthermore, the input of the loading position by the driver may be configured to set not only the number of people but also the weight. The sensor processing unit 207 may estimate the number of occupants in the vehicle 9 by performing known face recognition processing or person detection processing on the image captured by the camera 106. The sensor processing unit 207 may further recognize the physique, i.e., body size, of the occupants from the captured image and estimate the weight based on the physique.

[0054] According to the second embodiment described above, the following effects can be obtained.

[0055] (10) Weight-related data is data related to the weight of a person riding in the vehicle 9. The acquisition unit 206 acquires the output of the camera 106 that captures images of the interior of the vehicle 9. The suspension ECU 107A includes a sensor processing unit 207 that detects a person from an image captured by the camera 106, estimates the weight of the person, and outputs the weight to the acquisition unit 206.

[0056] (11) The acquisition unit 206 acquires weight-related data based on the output of a sensor mounted on the vehicle 9 or input by a vehicle occupant.

[0057] -Third embodiment- A third embodiment of a suspension system will be described with reference to Figure 17. In the following description, the same components as those in the first embodiment are denoted by the same reference numerals, and differences will be mainly described. Points that are not specifically described are the same as those in the first embodiment. This embodiment differs from the first embodiment mainly in that the neural network is updated.

[0058] 1 is a configuration diagram of a neural network updating system S. The neural network updating system S includes a server 1301 and a vehicle 9B. The server 1301 and the vehicle 9B are connected via an external network 1401. The external network 1401 is a network that includes wireless communication. The external network 1401 may include the Internet, or may include a base station managed by a communications carrier.

[0059] The configuration of vehicle 9B is substantially the same as that of vehicle 9 in the first embodiment. The difference is that vehicle 9B further includes an in-vehicle communication unit 112. In-vehicle communication unit 112 writes data transmitted from server 1301 to neural network storage unit 204.

[0060] The server 1301 includes a server communication unit 1304, a control unit 1305, a teacher data storage unit 1306, a learning unit 1307, and a learning result storage unit 1308. The server communication unit 1304 communicates with the vehicle 9B via an external network 1401. The server communication unit 1304 is, for example, a network interface card.

[0061] The teacher data storage unit 1306 and the learning result storage unit 1308 are non-volatile storage devices, such as hard disk drives. The teacher data storage unit 1306 stores pre-collected teacher data for learning. This teacher data is a plurality of sets of values ​​of the input layer N-IN and the output layer N-OUT of the neural network N shown in FIG. 3. The control unit 1305 issues a learning execution instruction to the learning unit 1307. When the learning process by the learning unit 1307 is completed, the control unit 1305 transmits the data stored in the learning result storage unit 1308 to the vehicle 9B via the server communication unit 1304.

[0062] The learning unit 1307 uses the training data stored in the training data storage unit 1306 to update the parameters of the neural network N stored in the learning result storage unit 1308. However, in this embodiment, the configuration of the neural network N is determined in advance. For example, when the number of epochs reaches a predetermined threshold, the learning unit 1307 ends the learning process and stores the parameters in the learning result storage unit 1308.

[0063] 17, if the vehicles are of the same model, the weight parameters learned using a data set acquired in one learning vehicle can be reflected in multiple vehicles 9B. Furthermore, if data sets are continuously acquired from learning vehicles and the server 1301 performs learning or additional learning, the weight parameters can be updated, and as a result, the accuracy of physical value estimation by the vehicle state estimation unit 202 is improved, thereby achieving better ride comfort.

[0064] According to the third embodiment described above, the neural network stored in the neural network storage unit 204 can be updated as needed.

[0065] In the above-described embodiments and modifications, the functional block configurations are merely examples. Some functional configurations shown as separate functional blocks may be integrated, or a configuration shown in a single functional block diagram may be divided into two or more functions. Furthermore, some of the functions of each functional block may be provided by other functional blocks.

[0066] In the above-described embodiments and modifications, the program is stored in the ROM 42, but the program may instead be stored in the storage device 44. Furthermore, the suspension ECU 107 may be provided with an input / output interface (not shown), and the program may be loaded from another device as needed via the input / output interface and a medium available to the suspension ECU 107. Here, the medium refers to, for example, a storage medium that is detachable from the input / output interface, or a communication medium, i.e., a wired, wireless, or optical network, or a carrier wave or digital signal that propagates through the network. Furthermore, some or all of the functions realized by the program may be realized by a hardware circuit or an FPGA.

[0067] The above-described embodiments and modifications may be combined with each other. Although various embodiments and modifications have been described above, the present invention is not limited to these. Other embodiments conceivable within the scope of the technical concept of the present invention are also included within the scope of the present invention.

[0068] 9: Vehicle 103: Weight measurement unit 105: Instrument panel 106: Camera 107: Suspension ECU 201: Semi-active suspension 202: Vehicle state estimation unit 204: Neural network storage unit 205: Read unit 206: Acquisition unit 207: Sensor processing unit 208: Output unit 2051: Correspondence table

Claims

1. A suspension system that is mounted on a vehicle and estimates a state of a suspension mounted on the vehicle, comprising: a memory unit in which a plurality of neural networks are stored; an acquisition unit that acquires weight-related data which is data relating to the weight of cargo on the vehicle; and a vehicle state estimation unit that estimates the state of the suspension by inputting data relating to the vehicle's traveling to any one of the neural networks stored in the memory unit that corresponds to the weight-related data.

2. A suspension system as described in claim 1, wherein the plurality of neural networks are loaded into the vehicle state estimation unit, and the vehicle state estimation unit selects the neural network corresponding to the weight-related data from the plurality of neural networks loaded and estimates the state of the suspension.

3. A suspension system as described in claim 1, further comprising a reading unit which reads out the neural network corresponding to the weight-related data from the memory unit and outputs it to the vehicle state estimation unit, and the vehicle state estimation unit estimates the state of the suspension using the neural network output by the reading unit.

4. A suspension system according to claim 1, wherein the weight-related data is data on the weight of the load.

5. A suspension system according to claim 1, wherein the weight-related data is data on the weight of the load and the position of the load.

6. A suspension system as described in claim 1, wherein the weight-related data is data relating to the weight of a person riding in the vehicle, and the acquisition unit further includes a sensor processing unit that acquires output from a camera that photographs the interior of the vehicle, detects the person from an image captured by the camera, estimates the weight of the person, and outputs the weight to the acquisition unit.

7. A suspension system as described in claim 3, wherein the reading unit reads out one of the neural networks from the memory unit based on the weight-related data each time it detects any one of the following: a door of the vehicle is closed, a seat belt is fastened in the vehicle, or an instruction from an occupant of the vehicle.

8. A suspension system as described in claim 2, wherein the vehicle state estimation unit reselects the neural network corresponding to the weight-related data each time it detects any one of the following: a door of the vehicle is closed, a seat belt is fastened in the vehicle, or an instruction from an occupant of the vehicle.

9. A suspension system according to claim 1, wherein the vehicle state estimating unit estimates the state of the suspension for each wheel of the vehicle.

10. A suspension system according to claim 2, wherein the acquisition unit acquires the weight-related data based on an output from a sensor mounted on the vehicle or an input from an occupant of the vehicle.

11. A suspension system according to claim 1, further comprising an output unit which outputs the weight-related data acquired by the acquisition unit to a display device provided in the vehicle.

12. An estimation method for estimating a state of a suspension mounted on a vehicle by a computer having a memory unit in which a plurality of neural networks are stored, the estimation method including: acquiring weight-related data, which is data relating to the weight of cargo on the vehicle, based on the output of a sensor mounted on the vehicle; and estimating the state of the suspension by inputting data relating to the running of the vehicle to any one of the neural networks stored in the memory unit that corresponds to the weight-related data.

Images