Sensory evaluation prediction system, suspension device, suspension control system

Abstract

The sensory evaluation prediction system of the present invention comprises: an input unit that reads the output of a behavior sensor that measures two or more time series information related to a moving body; a selection unit that selects two or more physical quantities from the output of the behavior sensor read by the input unit; a correlation creation unit that creates information indicating the correlation in the time series for the two or more physical quantities selected by the selection unit; and an evaluation circuit that calculates an evaluation value of a sensory index based on the information indicating the correlation in the time series.

Description

Technical Field

[0001] The present invention relates to a sensory evaluation prediction system, a suspension device, and a suspension control system. Background Art

[0002] In a car, vibration stimuli input from the road surface are transmitted to the occupants via the tires, suspension, chassis, seat rails, seat legs, and seat materials. In the sensory evaluation of ride quality, the main factor that matters is how the driver and passengers feel about the vibration stimuli. In addition, in the sensory evaluation of handling stability, the reaction when operating the steering wheel, the comfort of the response, and the presence or absence of a sense of disharmony are mainly important. Automobile manufacturers each have their own goals for ride quality, handling stability, and their balance. Trained expert drivers convey improvement points to those responsible for designing and adjusting parameters of vehicle parts, thereby improving ride quality and handling stability. Patent document 1 discloses a motion evaluation method characterized by detecting the acceleration obtained by differentiating at least the acceleration of a moving object, inputting the detected jerk into the input layer of a hierarchical neural network, and outputting the motion evaluation result from the output layer via the intermediate layer.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Laid-Open No. 7-244065 Summary of the Invention

[0006] Problems to be solved by the invention

[0007] The invention described in Patent Document 1 has room for improvement in terms of coping with variations in driver's operations.

[0008] Technical means to solve the problem

[0009] A sensory evaluation prediction system according to a first aspect of the present invention comprises: an input unit that reads the output of a behavior sensor that measures two or more time series information related to a moving body; a selection unit that selects two or more physical quantities from the output of the behavior sensor read by the input unit; a correlation creation unit that creates information indicating the correlation in the time series for the two or more physical quantities selected by the selection unit; and an evaluation circuit that calculates an evaluation value of a sensory index based on the information indicating the correlation in the time series.

[0010] The suspension device according to the second aspect of the present invention is manufactured based on the evaluation value output from the aforementioned sensory evaluation prediction system.

[0011] A third aspect of the present invention provides a suspension control system comprising: the sensory evaluation prediction system described above; and a suspension damping force variable mechanism for adjusting the damping force of a suspension device mounted on the mobile body based on the evaluation value output from the sensory evaluation prediction system.

[0012] Effects of the Invention

[0013] According to the present invention, since the correlation of a plurality of physical quantities is evaluated, the evaluation is less susceptible to the influence of variations in the driver's operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a block diagram of the sensory evaluation prediction system in the first embodiment.

[0015] Figure 2 This is a diagram showing an example of sensory index setting.

[0016] Figure 3 This figure shows an example of data specification setting.

[0017] Figure 4 FIG. 1 is a diagram showing an example of data acquired by a sensor group mounted on a vehicle.

[0018] Figure 5 Create a conceptual diagram showing the correlation between two physical quantities for demonstration purposes.

[0019] Figure 6 (a) is passed Figure 5 A graph showing the correlation between steering torque and steering angle is produced using the method shown in FIG. Figure 6 (b) means Figure 6 The information shown in (a) is divided with a resolution of 6 bits in both the vertical and horizontal directions.

[0020] Figure 7 A conceptual diagram showing the operation of an evaluation circuit using a hierarchical neural network.

[0021] Figure 8 To express Figure 7 The graph shows the relationship between the output of the hierarchical neural network and the sensory index value.

[0022] Figure 9 This figure shows an example of creating related information using three physical quantities.

[0023] Figure 10 This is a diagram showing an example of an evaluation circuit for each sensory index. Figure 10 (a) is a diagram showing the relationship between the input layer, hidden layer, and output layer of a hierarchical neural network. Figure 10(b) is a diagram showing an example of the relationship between the sensory index, two physical quantities to be evaluated, the number of hidden layer elements, and the number of output layer elements.

[0024] Figure 11 This is a flowchart showing the flow of processing of the sensory evaluation prediction system according to the first embodiment of the invention.

[0025] Figure 12 This is a diagram showing an example of a steering wheel operation and an output waveform of the evaluation circuit constituting the sensory evaluation prediction system in the first embodiment.

[0026] Figure 13 This is a diagram showing an example of visualizing index values ​​output by the aggregation unit constituting the sensory evaluation prediction system in the first embodiment.

[0027] Figure 14 This is a timing chart showing an example of the relationship between the operation of the evaluation circuit and the road surface on which the vehicle is traveling.

[0028] Figure 15 This is a block diagram of a sensory evaluation prediction system in the second embodiment.

[0029] Figure 16 This is a flowchart showing the flow of processing in the sensory evaluation prediction system according to the second embodiment.

[0030] Figure 17 This is a block diagram of a sensory evaluation prediction system in the third embodiment.

[0031] Figure 18 This is a flowchart showing the processing of the learning function of the sensory evaluation prediction system in the third embodiment.

[0032] Figure 19 This is a block diagram showing the functional configuration of a sensory evaluation prediction system according to a fourth embodiment.

[0033] Figure 20 This is a block diagram showing the functional configuration of a suspension control system according to the fifth embodiment. DETAILED DESCRIPTION

[0034] -First embodiment-

[0035] Below, reference Figures 1 to 14 A first embodiment of the sensory evaluation prediction system will be described. The sensory evaluation prediction system described below may be installed in a vehicle or may be installed outside the vehicle.

[0036] Figure 1This is a block diagram of a sensory evaluation prediction system 101 in the first embodiment. The sensory evaluation prediction system 101 includes a test result storage unit 102, a control unit 103, a register 104, a selection unit 105, an evaluation index determination unit 106, an evaluation unit 107, a totalization unit 108, a weight parameter storage unit 109, a totalization result storage unit 110, a display unit 111, and an input unit 115. The register 104 includes a sensory index setting 112, a data format setting 113, and a totalization mode setting 114. The evaluation unit 107 includes a first correlation generation unit 121, a second correlation generation unit 122, a third correlation generation unit 123, a fourth correlation generation unit 124, a fifth correlation generation unit 125, a first evaluation circuit 131, a second evaluation circuit 132, a third evaluation circuit 133, a fourth evaluation circuit 134, and a fifth evaluation circuit 135.

[0037] The first correlation generating unit 121 generates first related information and inputs it to the first evaluation circuit 131. The second correlation generating unit 122 generates second related information and inputs it to the second evaluation circuit 132. The third correlation generating unit 123 generates third related information and inputs it to the third evaluation circuit 133. The fourth correlation generating unit 124 generates fourth related information and inputs it to the fourth evaluation circuit 134. The fifth correlation generating unit 125 generates fifth related information and inputs it to the fifth evaluation circuit 135.

[0038] Hereinafter, the first correlation creation unit 121, the second correlation creation unit 122, the third correlation creation unit 123, the fourth correlation creation unit 124, and the fifth correlation creation unit 125 are collectively referred to as the correlation creation unit 120. Hereinafter, the first evaluation circuit 131, the second evaluation circuit 132, the third evaluation circuit 133, the fourth evaluation circuit 134, and the fifth evaluation circuit 135 are collectively referred to as the evaluation circuit 130. Hereinafter, the first relevant information, the second relevant information, the third relevant information, the fourth relevant information, and the fifth relevant information are collectively referred to as relevant information.

[0039] Furthermore, the five sensory indices evaluated by the evaluation circuit 130 are: feel near N (neutral), yaw response, grip, roll, and straightness. However, this is merely an example; the evaluation circuit 130 can also evaluate sensory indices other than those listed above. Furthermore, the number of sensory indices evaluated by the evaluation circuit 130 is unlimited, as long as it is two or more. Figure 1 1 shows an example in which the evaluation unit 107 evaluates five sensory indices. The number of related production units 120 and evaluation circuits 130 increases or decreases according to the number of sensory indices to be evaluated.

[0040] The control unit 103, the selection unit 105, the evaluation index determination unit 106, the evaluation unit 107, and the aggregation unit 108 perform calculations. These calculations are achieved, for example, by having a CPU (not shown) read a program from a ROM (not shown) and expand it into a RAM (not shown) for execution. However, these calculations can also be achieved by an FPGA (Field Programmable Gate Array), which is a rewritable logic circuit, or an ASIC (Application Specific Integrated Circuit), which is an integrated circuit for a specific application. In addition, these calculations can also be achieved by a combination of different structures, such as a combination of a CPU, ROM, RAM, and FPGA instead of a combination of a CPU, ROM, and RAM.

[0041] The test result storage unit 102, register 104, weight parameter storage unit 109, and total result storage unit 110 are non-volatile storage devices and may be referred to as "storage units." However, at least one of the test result storage unit 102, register 104, weight parameter storage unit 109, and total result storage unit 110 may also be a volatile storage device. In this case, information read from the non-volatile storage device (not shown) upon activation of the sensory evaluation prediction system 101 is stored in the volatile storage device. The display unit 111, such as a liquid crystal display, displays the video signal output from the control unit 103. The input unit 115 is a connection interface with the test result storage unit 102.

[0042] The test result storage unit 102 stores a learning object 1021 and an evaluation object 1022. The learning object 1021 is a combination of sensor data acquired by the sensor group installed in the evaluation vehicle in which the expert driver rides and the sensory index given by the expert driver at that time. Figure 4 This will be explained later. As will be described later, learning object 1021 is used to calculate the weight parameters stored in weight parameter storage unit 109. Furthermore, weight parameter storage unit 109 is referenced by evaluation unit 107. Evaluation object 1022 is sensor data acquired by the sensor group installed in the evaluation vehicle. Evaluation object 1022 is evaluated by evaluation unit 107.

[0043] The control unit 103 coordinates the operations of the various blocks that comprise the sensory evaluation prediction system 101. Specifically, the control unit 103 participates in all of the processing described below. However, for the sake of brevity, this involvement will not be specifically mentioned below. Specifically, the control unit 103 has the function of deactivating unnecessary evaluation circuits 130 based on the output of the evaluation index determination unit 106. For example, if the evaluation index determination unit 106 selects only the texture near N, the second through fifth evaluation circuits 132 through 135 are deactivated.

[0044] The sensory index setting 112 included in the register 104 is a register for setting the sensory index displayed on the display unit 111. An arbitrary value is preset in the sensory index setting 112. The sensory index setting 112 can be configured to be set from outside the sensory evaluation prediction system 101.

[0045] Figure 2 This is a diagram showing an example of the sensory index setting 112. Here, the sensory index setting 112 is composed of m bits, for example, 5 bits, and a sensory index is assigned to each bit. The evaluation index determination unit 106 sets "1" to the bit corresponding to the sensory index to be evaluated, and sets "0" to the other bits. Figure 2 This is just an example. As long as the same setting can be achieved, the data format of the sensory index setting 112 is not limited. Figure 1 Continue explaining.

[0046] The data specification setting 113 included in the register 104 is a register for setting the specifications of the data used for each sensory index, that is, the combination of sensor outputs. In this embodiment, the data specification setting 113 is not changed.

[0047] Figure 3 This figure shows an example of data specification setting 113. Data specification setting 113 is composed of the same number of tables as sensory indices, for example, m tables, and each table stores n bits of information. Sensor information, i.e., physical quantities, stored in the test result storage unit 102 are allocated to each bit in each table. In other words, bits corresponding to physical quantities calculated for the corresponding sensory indices are set to "1", and bits other than these are set to "0". Figure 3 This is just an example. As long as the same setting can be achieved, the data format of the data specification setting 113 is not limited. Figure 1 Continue explaining.

[0048] The aggregation mode setting 114 contained in the register 104 is setting information for determining whether the output of the evaluation unit 107 displayed on the display unit 111 is to be an instantaneous value or an average value. The aggregation mode setting 114 is read by the aggregation unit 108. The selection unit 105 outputs at least a portion of the evaluation target 1022 read from the test result storage unit 102 by the input unit 115 to each correlation creation unit 120. The selection unit 105 determines which data included in the evaluation target 1022 is output by referring to the output of the evaluation index determination unit 106 received via the control unit 103 and the data standard setting 113.

[0049] The evaluation index determination unit 106 selects sensory indices for sensory evaluation prediction based on the steering wheel operating conditions. Furthermore, one approach is to define the steering wheel operating conditions in accordance with ISO 13674-1 / 2 (Road vehicles - Test method for the quantification of on-centre handling Part 1 / 2), which specifies the test method for driving tests of steering stability. For example, when the evaluation index determination unit 106 determines that the vehicle is performing a continuous S-shaped Weave test, also known as slalom driving, based on the steering wheel operating conditions, it determines that the N-axis feel, yaw response, grip, and roll feel are the evaluation targets, while straightness is excluded. Alternatively, when the evaluation index determination unit 106 determines that the vehicle is driving in a straight line, it determines that the N-axis feel and straightness are the evaluation targets, while yaw response, grip, and roll feel are excluded. The steering wheel operating conditions include, for example, the steering wheel position, the steering wheel speed (the derivative of the steering wheel position), and the steering wheel acceleration (the derivative of the steering wheel speed). In addition, hereinafter, information indicating the operation status of the steering wheel may also be referred to as steering wheel information.

[0050] Furthermore, the steering wheel operation patterns assumed by the evaluation index determination unit 106 include "step steer," a mode in which a predetermined steering angle is maintained after a certain period of straight driving. Sensory indices corresponding to this operation pattern are also predefined. Alternatively, based on sensory indices, there are steering wheel operations assumed for each sensory index. Therefore, the steering wheel information referenced by the evaluation index determination unit 106 is the steering wheel operation status during driving, and the sensory evaluation prediction system 101 determines the sensory index for sensory evaluation prediction based on this information. Furthermore, the sensory index determination can be grouped and determined based on the steering wheel operation status defined above, taking into account differences in parameters such as vehicle speed. For example, pattern matching of steering wheel angle data can be used.

[0051] Evaluation unit 107 includes a correlation generator 120 and an evaluation circuit 130. Correlation generator 120 uses the two or more physical quantities transmitted from selection unit 105 to generate correlated information, which is information indicating a temporal correlation between these physical quantities. The correlation generators 120 may use the same or different methods for generating correlated information, such as the types of physical quantities used, scaling settings, and the order of the data used.

[0052] The summing unit 108 sums the sensory index values ​​output by the evaluation circuit 130. As previously mentioned, each sensory index has a steering wheel operation that is evaluated. Therefore, if the vehicle is traveling with a steering wheel operation that is not evaluated, the evaluation circuit 130 may not output an appropriate sensory index value. Therefore, based on the determination result of the evaluation index determination unit 106, the sensory index values ​​output by the evaluation circuit 130 are processed only when the vehicle is traveling with a steering wheel operation that is not evaluated. Sensory index values ​​output by the evaluation circuit 130 when the vehicle is traveling with a steering wheel operation that is not evaluated are excluded. The summing unit 108 determines whether the calculated result should be excluded for each sensory index and writes the sensory index values ​​that have not been excluded to the summed result storage unit 110 along with a timestamp.

[0053] The weight parameter storage unit 109 stores parameters used by the evaluation circuit 130. This embodiment assumes five sensory indices, so the weight parameter storage unit 109 has the capacity to store at least five sets of parameters. The parameters referred to here are, for example, coefficients of the equations used in the evaluation circuit 130 or the weights Wij of the inter-element connections when the evaluation circuit 130 is implemented using a hierarchical neural network.

[0054] The display unit 111 presents the sensory index values ​​stored in the summed result storage unit 110 to the vehicle occupants. Furthermore, the sensory index output by the display unit 111 is selectable, and this can be externally selected using the sensory index setting 112 contained in the register 104. Furthermore, the sensory index values ​​output by the display unit 111 can be selected as either an instantaneous value or an average value during driving, and this can be externally set using the summed mode setting 114 contained in the register 104.

[0055] Figure 4 2 is a diagram showing an example of data acquired by a sensor group mounted on the vehicle 201 . Figure 4The sensors shown here can also be referred to as "behavior sensors" because they measure the behavior of the vehicle itself. The pitch rate, roll rate, and yaw rate shown at 202 are examples of vehicle behavior data acquired for sensory evaluation prediction. The vertical acceleration, longitudinal acceleration, and lateral acceleration shown at 203 are examples of chassis data acquired for sensory evaluation prediction. The vehicle speed, steering wheel, GPS, camera, and radar data shown at 204 are examples of other acquired data used for sensory evaluation prediction.

[0056] In particular, the chassis part selects all or part of the acceleration data related to the path from the stimulus from the road surface to the tires and to the occupants. For example, the lower spring part and upper spring part of the suspension, the area around the seat where the occupants sit, the tie rod of the steering wheel, and the steering wheel. Figure 4 Although not shown in the figure, it is also possible to obtain information about the driver's operating object, such as the accelerator pedal, brake pedal, etc. Furthermore, as long as the sensory evaluation prediction of the steering stability can be achieved, measurement points other than the ones listed here can also be used. Figure 4 The information acquired by the illustrated sensor group is stored in the test result storage unit 102 , but the sensory evaluation prediction system 101 does not necessarily need to be stored in the vehicle 201 .

[0057] refer to Figures 5 to 8 , the outline of the process of inferring sensory indices using two physical quantities is described. Figure 9 , the outline of the process of inferring sensory indices using three physical quantities is described. As mentioned above, the correlation information represents the correlation in the time series of two or more physical quantities, and sometimes also represents the correlation in the time series of three or four or more physical quantities, but Figures 5 to 8 The following describes an example of the simplest case, in which a sensory index is inferred using the correlation between two physical quantities, namely, a first physical quantity P1 and a second physical quantity P2.

[0058] Figure 5 Create a conceptual diagram showing the correlation between two physical quantities for demonstration purposes. Figure 5 The top portion of the graph shows a time series diagram with the solid line representing physical quantity P1 and the dashed line representing physical quantity P2. Time passes as you move to the right. At time t0, changes in physical quantities P1 and P2 begin, and time passes at time t1, time t2, time t3, and time t4. Figure 5 The lower part of shows the correlation between the physical quantity P1 and the physical quantity P2 during the time from time t0 to each time t1 to t4. Figure 5 In the lower part, the horizontal axis is set to the value of the physical quantity P1, and the vertical axis is set to the value of the physical quantity P2.

[0059] For example, at time t1, the line graph is plotted in the first quadrant, at time t2, the line graph is plotted in the second quadrant, at time t3, the line graph is plotted in the third quadrant, and at time t4, the line graph is plotted in the fourth quadrant. By continuously performing these plots, a scatter plot is created that visualizes the correlation between physical quantities P1 and P2.

[0060] Figure 6 (a) is passed Figure 5 A graph showing the correlation between the steering torque and the steering angle is created using the method shown. Figure 6 (b) means Figure 6 The information shown in (a) is divided with a resolution of 6 bits in both the vertical and horizontal directions.

[0061] right Figure 6 Detailed explanation of (b) in the figure is provided. The steering torque, shown on the horizontal axis, can take both positive and negative values, centered around 0 [N·m]. Furthermore, the steering angle, shown on the vertical axis, assumes a neutral steering wheel state (for straight driving) of 0 [degrees]. For example, steering to the right is represented by a positive value, while steering to the left is represented by a negative value. By assigning the values ​​of 0 to 31 (decimal) in the 6-bit digital space, a scatter plot is created with a balanced positive and negative distribution.

[0062] Furthermore, the related production unit 120 has a function of deriving the maximum and minimum values ​​assumed in the steering wheel action under the specified driving conditions for the physical quantities stored in the test result storage unit 102, such as the steering angle and the steering torque. The specified driving conditions mentioned here refer to conditions such as "driving around the poles at a speed of 100 kmh, 0.2 Hz and a maximum lateral acceleration of 0.4G". In addition, the maximum and minimum values ​​are derived by comparative calculation by obtaining the physical quantity data under the conditions from the test result storage unit 102. Normalization is performed using the value with the larger absolute value between the maximum and minimum values. Furthermore, if digitization is implemented in which "1" is set when there is drawing and "0" is set when there is no drawing, it is possible to produce an image that can set the value of 0 to 31 (decimal) on the one hand and can realize confirmation of the overall tendency on the other hand.

[0063] More specifically, the relevant information is a raster image of 64 pixels in length and 64 pixels in width, where each pixel represents whether or not it is drawn, using 1 or 0. This relevant information is represented as a 4096-dimensional column vector, for example.

[0064] Figure 7 1 is a conceptual diagram showing the operation of the evaluation circuit 130 using a hierarchical neural network. Figure 7 In, equivalent to Figure 1The evaluation circuit 401 of each evaluation circuit 130 shown is composed of a three-layer hierarchical neural network in which the elements of the input layer (number of elements I+1), the hidden layer (number of elements J+1), and the output layer (number of elements K) are coupled in a hierarchical manner. Figure 7 As shown, one element representing a bias term is set in each of the input layer and hidden layer. Each element in the input layer is coupled to each element in the hidden layer using weights W1ij (i = 1 to I+1, j = 1 to J+1), and each element in the hidden layer is coupled to each element in the output layer using weights W2jk (j = 1 to J+1, k = 1 to K). As previously mentioned, information about these weights is stored in the weight parameter storage unit 109.

[0065] picture Figure 6 As described in (b), when each physical quantity is represented by 6 bits, the scatter plot can be considered as a 64-pixel × 64-pixel digital image that represents the presence or absence of the drawing with pixel values ​​of 1 and 0. The information of the digital image is input to the evaluation circuit 401. If a data conversion is implemented in which the 0 value of each physical quantity is set as the center of the digital space, the position of the pixel also makes sense, so the pixel data itself is set to the input of the hierarchical neural network. For example, the pixel values ​​from the upper left of the digital image are set to the inputs a11 to a1I of the hierarchical neural network in point order toward the lower right of the digital image. If a1I is specified as the I-th input element, then I = 4096 (= 64 × 64). In this neural network, only any one element of the output layer outputs "1", and the output layer elements other than this output "0".

[0066] Figure 8 To express Figure 7 The graph shows the relationship between the output of the hierarchical neural network and the sensory index value. The output layer element specifications are set to a maximum score of 8.00, a minimum score of 4.00, and a score interval of 0.25. The number of output layer elements K in this case is 17.

[0067] The determination of the weight parameters stored in the weight parameter storage unit 109, that is, the learning of the so-called evaluation circuit 401 is performed as follows. The learning object 1021 stored in the test result storage unit 102 includes a large number of combinations of sensor outputs when the expert driver is riding in the vehicle and sensory index values ​​answered by the expert driver. In a certain driving test, when the expert driver answered the sensory index value of the hand feeling near N as 7.75 points, the input value and the output value are set to the following combination for learning. That is, the input value is obtained by plotting the correlation between the time series of the steering torque and the steering angle on a two-dimensional plane. Figure 6 The raster image information shown in (b) is shown in FIG. Among the output values, only the output layer element a32 corresponding to 7.75 points is "1", and the other output layer elements are "0".

[0068] Using many such input and output value pairs, a hierarchical neural network learns the correlation between a large amount of time series data and sensory index values ​​answered by expert drivers. The commonly known back-propagation method can be used for neural network learning.

[0069] Figure 9 This figure shows an example of creating related information by selecting three physical quantities. Figure 9 The example shown shows an example in which the physical quantities P1 to P3 are combined, with the physical quantity P1 being set as the X axis, the physical quantity P3 being set as the Y axis, and the physical quantity P2 being set as the Z axis. Figure 6 The example shown is the correlation of two physical quantities, so it is plotted on a two-dimensional plane. Figure 9 It is the correlation of three physical quantities, so it is plotted in three-dimensional space.

[0070] Then, the image is divided into voxels of a predetermined size, and a value of "1" or "0" is set according to whether or not the image is drawn in the voxel. Then, the value of the voxel is output in a predetermined order as the related information. The generated related information is input to the evaluation circuit 130 for processing and Figure 8 The same as shown, so the explanation is omitted. Furthermore, since it is difficult to visualize more than four physical quantities, the explanation here is omitted, but the same method can be used to deal with it, and there is no upper limit to the number. For example, the correlation of 10 physical quantities in time series can be used as relevant information.

[0071] Figure 10 This is a diagram showing an example of an evaluation circuit for each sensory index. Figure 10 (a) is a diagram showing the relationship between the input layer, hidden layer, and output layer of a hierarchical neural network. Figure 10 (b) is a diagram showing an example of the relationship between the sensory index, two physical quantities to be evaluated, the number of hidden layer elements, and the number of output layer elements. Figure 10 The evaluation circuit shown in (a) is Figure 7 The evaluation circuit 401 is composed of a hierarchical neural network in which an input layer 501, a hidden layer 502, and an output layer 503 are coupled in a hierarchical manner. Figure 1 The evaluation circuit is set for each sensory index, such as the first evaluation circuit 131 to the fifth evaluation circuit 135. As mentioned above, the specifications of these evaluation circuits may be the same or different.

[0072] Figure 10 (b) is a diagram showing an example of the specifications of each evaluation circuit. For example, the evaluation circuit corresponding to the touch feeling near N is Figure 1 In the first evaluation circuit 131, the physical quantities used in the evaluation are the steering torque and the steering angle, the number of hidden layer elements J = 100, and the evaluation circuit corresponding to the yaw response is Figure 1 In the second evaluation circuit 132, the physical quantities used in the evaluation are the steering angle and the yaw rate, and the number of hidden layer elements J = 200. In addition, the evaluation circuit corresponding to the grip is Figure 1 In the third evaluation circuit 133, the physical quantities used in the evaluation are the yaw rate and the lateral acceleration, the number of hidden layer elements J = 250, and the evaluation circuit corresponding to the roll feeling is Figure 1 In the fourth evaluation circuit 134, the physical quantity used for evaluation is lateral acceleration, and the number of hidden layer elements is J = 500. Furthermore, the number of output layer elements is K = 17 in all evaluation circuits. The selection of these physical quantities and parameter values ​​is merely an example; other parameter values ​​may also be used.

[0073] Figure 11 This is a flowchart showing the flow of processing of the sensory evaluation prediction system according to the first embodiment of the present invention: First, in step S701, the control unit 103 sets the flag value indicating the implementation status of the sensory evaluation to "0" indicating that the sensory evaluation has not been performed.

[0074] In step S702, the control unit 103 determines whether sensory evaluation is enabled, that is, whether an instruction to conduct a sensory evaluation has been issued, based on an operation by a vehicle occupant. If sensory evaluation is enabled, the control unit 103 determines that an instruction to conduct a sensory evaluation has been issued, and the process proceeds to step S703. If sensory evaluation is disabled, the control unit 103 determines that an instruction to conduct a sensory evaluation has not been issued, and the process proceeds to step S715.

[0075] In step S703, the control unit 103 sets the previously mentioned flag value indicating the status of sensory evaluation to "1," indicating that the evaluation has begun. Next, the control unit 103 acquires steering wheel operation information (step S704) and analyzes the steering wheel operation over time (step S705). Subsequently, in step S706, the control unit 103 uses the evaluation index determination unit 106 to determine the sensory index corresponding to the steering wheel operation.

[0076] In step S707, the control unit 103 selects the evaluation circuit corresponding to the sensory indicator determined as the evaluation indicator in step S706 from among the evaluation circuits 130 provided for each sensory indicator. In step S708, the control unit 103 selects the storage area of ​​the total result storage unit 110 corresponding to the evaluation circuit selected in step S707 as the storage block for the evaluation value.

[0077] In step S709, the selection unit 105 extracts the evaluation targets 1022 from the test result storage unit 102 over a predetermined time range, generating time-series data to be used as evaluation data. In step S710, the selection unit 105 uses information such as nearby road surface information and vehicle speed information to adjust the timing for starting operation of the evaluation circuit 130. The time-series data generated in step S709 is then deployed as evaluation data to the evaluation circuit selected in step S707 based on the adjusted timing.

[0078] In step S711 , the evaluation circuit selected in step S707 in the evaluation circuit 130 calculates an evaluation value for the evaluation index determined in step S706 based on the evaluation data input from the selection unit 105 in step S710 .

[0079] In step S712, the totaling unit 108 determines whether the set totaling mode is the instantaneous value totaling mode or the average value totaling mode based on the value of the totaling mode setting 114. For example, if the value of the totaling mode setting 114 is "0," the totaling unit 108 determines that the instantaneous value totaling mode is set and proceeds to step S713. If the value of the totaling mode setting 114 is "1," the totaling unit 108 determines that the average value totaling mode is set and proceeds to step S714.

[0080] In step S713, the totaling unit 108 transmits and displays the evaluation value calculated in step S711 on the display unit 111. Thus, the instantaneous value of the evaluation value for the evaluation indicator determined in step S706 is output externally using the display unit 111. Furthermore, depending on the evaluation value calculation cycle, the instantaneous value may change too rapidly to be easily observed. In such cases, a predetermined time average value may be calculated and displayed instead of the instantaneous value.

[0081] In step S714, the totaling unit 108 writes the evaluation value calculated in step S711 to the storage block selected in step S708. After completing steps S713 or S714, the process returns to step S702 and repeats the aforementioned process. Thus, the series of steps S703 to S714 continues until the sensory evaluation is determined to be off in step S702.

[0082] If it is determined in step S702 that the sensory evaluation is off, the control unit 103 determines in step S715 whether the flag value indicating the implementation status of the sensory evaluation is set to "1." If the flag value is set to "1," the sensory evaluation is determined to have been implemented through the series of processes from steps S703 to S714, and the process proceeds to step S716. If the flag value is set to "0," the sensory evaluation is determined not to have been implemented, and the process returns to step S701.

[0083] In step S716, similar to step S712 described above, the totaling unit 108 determines whether the set totaling mode is the instantaneous value totaling mode or the average value totaling mode. If the average value totaling mode is set, the process proceeds to step S717. If the instantaneous value totaling mode is set, the process returns to step S701. In step S717, the totaling unit 108 reads the evaluation value stored in the totaling result storage unit 110.

[0084] In step S718, the totaling unit 108 calculates the total value from the start of the evaluation based on the evaluation value read out in step S714. Figure 11 The average value of the evaluation values ​​for each evaluation indicator from the processing onwards is calculated. In the following step S719, the aggregation unit 108 transmits and displays the average value calculated in step S718 on the display unit 111. In this way, the average values ​​of the evaluation values ​​when the vehicle is traveling on the evaluation target road surface are aggregated and output externally using the display unit 111. When the processing of step S719 is completed, the process returns to step S701.

[0085] Figure 12 This diagram shows an example of steering wheel operation and output waveforms of the evaluation circuit, which constitutes the sensory evaluation prediction system in the first embodiment. Reference numeral 801 represents the time-series change in steering wheel operation, while reference numerals 802 through 804 each represent a sensory evaluation prediction waveform for the feel near N within intervals 1 through 3. Reference numerals 805 through 807 each represent a sensory evaluation prediction waveform for straightness within intervals 1 through 3.

[0086] exist Figure 12 In the example shown, the evaluation targets for the feel near sensory index N are sections 1 and 3, which can be determined as slalom driving based on steering wheel operation, while section 2, which is only for straight driving, is excluded. Furthermore, for the sensory index of straightness, sections 1 and 3 are excluded, while section 2 is the target. The evaluation circuit targets sensory indicators of steering stability, which are subject to fluctuations, so it is not expected to output a constant fixed value. However, if the steering wheel operation is outside the evaluation target, the intended learning will not be achieved, and it is expected that the waveform will saturate to the upper limit value, as shown in symbols 805 and 807, or fluctuate with a large amplitude, as shown in symbol 803.

[0087] In other words, the sensory evaluation prediction values ​​obtained when driving with a steering wheel operation other than the evaluation target are considered to be less reliable. On the other hand, if the steering wheel operation is the evaluation target, the expected learning has been achieved, so it is expected that a certain range of values ​​will be output, and waveforms such as symbols 802, 804, and 806 can be expected. Therefore, the evaluation index determination unit 106 uses the steering wheel information to determine whether each evaluation circuit 130 is an evaluation target and excludes sensory evaluation prediction values ​​obtained with steering wheel operations other than the evaluation target.

[0088] Figure 13 This is a diagram showing an example of visualizing index values ​​output by the aggregation unit constituting the sensory evaluation prediction system in the first embodiment. Figure 13 (a) is an example of visualizing sensory evaluation prediction values ​​for 5 sensory indices, and the chart type is a radar chart. Furthermore, the sensory indices to be visualized can be selected based on the unillustrated setting value stored in register 104. The unillustrated setting value is a collection of 1-bit registers corresponding to the sensory indices, for example, consisting of an N vicinity feel selection register, a yaw response selection register, a grip selection register, a roll selection register, and a straightness selection register. If "1" is set for each register value, it will be displayed, and if "0" is set for each register value, it will not be displayed. Therefore, Figure 13 Case (a) shows a case where the register values ​​of the five display selection registers are all set to "1".

[0089] Furthermore, the sensory evaluation prediction value is aggregated differently depending on the setting of aggregation mode setting 114. For example, when the register value of aggregation mode setting 114 is "1," the sensory evaluation prediction value is the average of the sensory evaluation prediction values ​​when the vehicle is operating the steering wheel to be evaluated. When the register value of aggregation mode setting 114 is "0," the sensory evaluation prediction value is the instantaneous value when the vehicle is operating the steering wheel to be evaluated. This is an example; the possible register values ​​can be expanded to form a moving average. Furthermore, a register with a width of 2 bits or more can be used to set the window width for calculating the moving average.

[0090] Figure 13 (b) is an example of visualizing one sensory index, namely, grip, and the chart type is a bar chart. This is equivalent to the case where only the register value of the grip selection register is "1" and the other values ​​are set to be non-display. The setting specifications of the total mode setting 114 are the same as those described above. Figure 13 The same as (a) of , so the description is omitted. Figure 13In the display, the benchmark is displayed as a sensory index value of 6.0 points. "6.0" is not required, but it is important to visualize the sensory evaluation prediction value when comparing it to the benchmark point.

[0091] Figure 14 This is a time chart showing an example of the relationship between the operation of the evaluation circuit constituting the sensory evaluation prediction system in the first embodiment and the travel road surface.

[0092] Symbol 1001 indicates the waveform of the timing for detecting the type of steering wheel operation, symbol 1002 indicates the operation timing of the evaluation circuit for the feel near N, symbol 1003 indicates the operation timing of the evaluation circuit for the yaw response, symbol 1004 indicates the operation timing of the evaluation circuit for the grip feel, symbol 1005 indicates the operation timing of the evaluation circuit for the roll feel, and symbol 1006 indicates the operation timing of the evaluation circuit for the straightness.

[0093] First, the steering wheel operation 801 transitions from section 1 to section 3 via section 2. The evaluation index determination unit 106 acquires the steering wheel operation information and detects the type of the steering wheel operation. Figure 14 , the timing at which the pulse waveform of symbol 1001 becomes High (1) is detected. The evaluation index determination unit 106 determines the feel, yaw response, and roll feel near the sensory index N as sensory evaluation within one of the intervals, and outputs evaluation index selection signals 1002, 1004, and 1006 that become High (1) when driving on the evaluation target road surface. On the other hand, for grip feel and straightness, which are not the evaluation targets, the evaluation index determination unit 106 outputs evaluation index selection signals 1003 and 1005 that become Low (0).

[0094] The timing of switching the evaluation index selection signal is the timing of detecting the type of steering wheel operation. Figure 14 In the example, evaluation index selection signals 1002 to 1006 are displayed assuming that the type of steering wheel operation is switched, as in the case of steering wheel operation 801. However, this is only an example. There are various concepts for sensory indices, and evaluation index selection signals are generated based on these concepts.

[0095] Furthermore, in this embodiment, for simplicity and ease of understanding, the description focuses on a two-dimensional image composed of two physical quantities, P1 and P2. However, as long as the sensory index of steering stability can be determined, the image is not limited to a two-dimensional image and can also be a three-dimensional or higher data space using three or more physical quantities. In particular, as long as a hierarchical neural network is used, the amount of information that can be grasped by humans does not necessarily need to be limited.

[0096] According to the first embodiment described above, the following effects are achieved.

[0097] (1) The sensory evaluation prediction system 101 includes an input unit 115 that reads the output of a behavior sensor that measures two or more time-series information related to a moving object; a selection unit 105 that selects two or more physical quantities from the output of the behavior sensor read by the input unit 115; a correlation generating unit 120 that generates information indicating the time-series correlation between the two or more physical quantities selected by the selection unit 105; and an evaluation circuit 130 that calculates an evaluation value of a sensory index based on the time-series correlation information. Therefore, the sensory evaluation prediction system 101 evaluates the correlation between multiple physical quantities and is therefore less susceptible to variations in the driver's operation.

[0098] (2) The evaluation circuit 130 is capable of calculating a plurality of sensory indices. The sensory evaluation prediction system 101 includes: an evaluation index determination unit 106 that determines a sensory index of an evaluation target based on the steering wheel operation of a moving object; and a register 104 that stores a data specification setting 113 that associates the sensory index determined by the evaluation index determination unit 106 with two or more physical quantities corresponding to the sensory index. The selection unit 105 refers to the data specification setting 113 and determines two or more physical quantities based on the determination of the evaluation index determination unit 106. Therefore, the steering stability of the driver's steering wheel operation can be evaluated using an appropriate evaluation index.

[0099] (3) When the selection unit 105 selects two physical quantities, the correlation production unit 120 plots the time series correlation of the two physical quantities on a two-dimensional plane and outputs the plot in the form of raster image information. Therefore, the time series correlation of the two physical quantities can be concisely expressed, and the tolerance to data deviations is good. Outputting the plot in the form of vector image information is also considered, but considering that it is used for input to the input layer of the hierarchical neural network, the robustness of vector image information is low, and it is difficult to obtain a stable output. Therefore, the method of using raster image information, in other words, the value of each pixel, for input to the input layer as in this embodiment is relatively excellent.

[0100] (4) When the selection unit 105 selects three physical quantities, the correlation generating unit 120 plots the time series correlations of the three physical quantities in a three-dimensional space and outputs the plot in the form of voxel information. Therefore, the time series correlations of the three physical quantities can be concisely expressed with high tolerance to data variations.

[0101] (5) The evaluation circuit 130 includes a plurality of evaluation sub-circuits, namely, a first evaluation circuit 131 to a fifth evaluation circuit 135, corresponding to a plurality of sensory indices. The control unit 103 deactivates any evaluation circuit 130 that does not calculate a sensory index, based on a selection made by the evaluation index determination unit 106. This reduces power consumption. This is particularly useful when the sensory evaluation prediction system 101 is installed in a vehicle.

[0102] (6) Data Specification Setting 113 The combination of physical quantities varies for each sensory index. Therefore, the optimal combination of physical quantities can be used for each sensory index.

[0103] (7) The sensory evaluation prediction system 101 is mounted on a mobile object. The input unit 115 reads the output of the behavior sensor mounted on the mobile object. The sensory evaluation prediction system 101 includes a summing unit 108 for summing the calculation results of the evaluation circuit 130. The summing unit 108 can switch between an instantaneous evaluation mode, in which the instantaneous value or the moving average of the calculation results of the evaluation circuit 130 is output, and a comprehensive evaluation mode, in which the average value of the calculation results of the evaluation circuit 130 over a predetermined period is output.

[0104] (Variation 1)

[0105] In the first embodiment described above, the evaluation index determination unit 106 uses a pattern matching method to determine the evaluation index based on the steering wheel information. However, the relationship between the steering wheel information and the evaluation index can also be implemented through inference based on learning using a hierarchical neural network. In this hierarchical neural network, for example, time-series steering wheel information divided into fixed periods is input, and elements corresponding to each sensory index are output layer elements. During the learning phase, weight parameters are learned so that elements corresponding to the sensory index answered by the expert driver are "1" and all others are "0."

[0106] According to this modification, in addition to the effects of the first embodiment described above, the following effects are obtained.

[0107] (8) The evaluation index determination unit 106 determines the relationship between the steering wheel operation information and the sensory index based on learning. In the pattern matching described above, it is necessary to determine in advance whether the steering wheel is to be evaluated or not. In contrast, the use of a hierarchical neural network has the following advantages. Specifically, training data can be obtained from sensory evaluation tests conducted by expert drivers, enabling the selection of evaluation indexes consistent with actual concepts.

[0108] (Variation 2)

[0109] When the sensory evaluation prediction system 101 is mounted on a vehicle, it is not necessary to include the test result storage unit 102 . In this case, the output of the sensor group mounted on the vehicle is input to the input unit 115 .

[0110] (Variation 3)

[0111] The sensory evaluation prediction system 101 may evaluate only one sensory index. In this case, the sensory evaluation prediction system 101 may not include the evaluation index determination unit 106 .

[0112] -Second embodiment-

[0113] refer to Figures 15 and 16 A second embodiment of the sensory evaluation prediction system will be described. In the following description, components identical to those in the first embodiment are denoted by the same reference numerals, with the main focus on differences. Aspects not specifically described are identical to those in the first embodiment. This embodiment differs from the first embodiment primarily in that a single evaluation circuit is used for multiple sensory indices.

[0114] Figure 15 1 is a block diagram of a sensory evaluation prediction system 101A in the second embodiment. An evaluation unit 107A is different from the first embodiment in that it includes only a correlation creation unit 126 and an evaluation circuit 136 .

[0115] The relevant production unit 126 is to Figure 1 The first correlation generating unit 121 to the fifth correlation generating unit 125 described in the previous section are combined. The correlation generating unit 126 obtains the information of the evaluation index for evaluation by the evaluation circuit 136 through the control unit 103, and generates the related information in the time series based on the multiple physical quantities corresponding to the evaluation index. The correlation generating unit 126 outputs the generated related information to the evaluation circuit 136. Figure 15 In the example shown, only one correlation generating unit 126 is shown. However, the evaluation unit 107A may include multiple correlation generating units 126. As long as the evaluation circuit 136 and the correlation generating unit 126 are shared for at least two or more sensory indices, any number of correlation generating units 126 may be provided in the evaluation unit 107A.

[0116] The evaluation circuit 136 is a circuit that evaluates the sensory indexes in the first embodiment. Figure 1 The first evaluation circuit 131 to the fifth evaluation circuit 135 for each sensory index described in the previous section are shared. That is, the correlation production unit 126 and the evaluation circuit 136 are shared for each sensory index described above, such as the five sensory indices of N vicinity feel, yaw response, grip, roll feel, and straightness. Figure 15While only one evaluation circuit 136 is shown in the example, the evaluation unit 107A may include multiple evaluation circuits 136. As long as a single evaluation circuit 136 is used for at least two sensory indices, any number of evaluation circuits 136 may be provided within the evaluation unit 107A. Furthermore, the number of associated production units 126 and evaluation circuits 136 may differ.

[0117] Figure 16 This is a flowchart showing the processing flow of the sensory evaluation prediction system according to the second embodiment of the present invention. Figure 11 Compared with the flowchart, Figure 16 The difference of the flowchart of the embodiment is that step S1201 is provided instead of step S707. In addition, except for these processing steps that are different from the first embodiment, the description thereof will be omitted unless otherwise required.

[0118] In step S1201, the control unit 103 reads the weight parameters for the evaluation circuit corresponding to the sensory index selected as the evaluation index in step S706 from the weight parameter storage unit 109. The read weight parameters are then set in the evaluation circuit 136. Thus, the evaluation unit 107A adjusts the evaluation circuit 136 according to the evaluation index.

[0119] In step S711 , the evaluation circuit 136 adjusted according to the evaluation index in step S1201 calculates an evaluation value for the evaluation index based on the evaluation data input from the selection unit 105 in step S710 .

[0120] According to the second embodiment of the present invention described above, in addition to the same effects as those of the first embodiment, the following effects are also achieved.

[0121] (9) Evaluation circuit 130 includes evaluation circuit 136, a small evaluation circuit that commonly corresponds to multiple sensory indices. Evaluation circuit 136 is adjusted based on the evaluation index selected by evaluation index determination unit 106, and the evaluation value is calculated using the adjusted evaluation circuit 136. Specifically, evaluation circuit 136 is constructed using a neural network composed of multiple elements coupled in a hierarchical manner, and the weight parameter of each element is adjusted based on the evaluation index. This allows for a reduction in circuit size.

[0122] -Third embodiment-

[0123] refer to Figures 17 and 18The third embodiment of the sensory evaluation prediction system will be described. In the following description, the same reference numerals are assigned to the same components as in the first embodiment, and the differences are mainly described. Aspects not specifically described are the same as in the second embodiment. This embodiment differs from the second embodiment primarily in that the physical quantities used for estimating sensory indices are determined through learning, and the data specification settings are created.

[0124] Figure 17 This is a block diagram of a sensory evaluation prediction system 101B in the third embodiment. In this embodiment, the register 104 also stores an exploration pattern 1041 and a learning determination threshold 1042. However, at the start of the processing described below, the data specification setting 113 may not be stored in the register 104; the data specification setting 113 is created through the processing described below. Furthermore, in this embodiment, at the start of the processing described below, the weight parameter storage unit 109 may not store data; data is stored in the weight parameter storage unit 109 through the processing described below.

[0125] When "1" is set to the exploration mode 1041, the sensory evaluation prediction system 101B switches to the exploration mode and creates the data specification setting 113. When "0" is set to the exploration mode 1041, the sensory evaluation prediction system 101B switches to the non-exploration mode and performs the operations described in the first embodiment using the pre-created data specification setting 113 or the data specification setting 113 read from an external source.

[0126] In this embodiment, a learning unit 107B is provided in place of the evaluation unit 107A. In addition to the functions of the evaluation unit 107A in the second embodiment, the learning unit 107B also has the learning function described below. The learning unit 107B operates similarly to the second embodiment in the non-exploration mode and performs the learning function in the exploration mode.

[0127] In the exploration mode, the learning unit 107B explores combinations of physical quantities used in estimating sensory indices as follows. The learning unit 107B first selects an arbitrary combination of physical quantities to create first related information. Next, the learning unit 107B performs learning using a hierarchical neural network based on the relationship between the first related information and the sensory index values ​​of steering stability acquired from an expert driver. If the output error of the hierarchical neural network is smaller than the learning determination threshold 1042, that is, if the difference from the training data has decreased to a certain degree or more, learning is determined to be possible, and the combination of physical quantities used in the first related information is recorded in the data specification setting 113. Furthermore, the learned parameters are stored in the weight parameter storage unit 109.

[0128] On the other hand, if the hierarchical neural network's output error exceeds the learning determination threshold of 1042 during a trial learning of the hierarchical neural network, that is, if the difference from the training data has not decreased by a certain amount, learning is determined to be unsuccessful. In this case, different combinations of physical quantities are selected to create second related information, and hierarchical neural network learning is trialed based on the relationship between the second related information and the sensory index value of steering stability obtained from the expert driver. In this manner, physical quantity combinations are explored until learning is determined to be possible.

[0129] Furthermore, various methods can be used to explore combinations of physical quantities used to generate relevant information. For example, two or more physical quantities can be randomly selected from a plurality of physical quantities. Alternatively, physical quantities can be prioritized based on the perception of an expert driver and combinations can be explored in descending order of priority. Furthermore, exploration can be conducted through trial and error while conducting evaluation, as in reinforcement learning, a form of AI (artificial intelligence).

[0130] Figure 18 This is a flowchart illustrating the processing of the learning function of the sensory evaluation prediction system 101C in the third embodiment. First, in step S1402, the control unit 103 determines whether the exploration mode 1041 is set to active, that is, whether the learning mode is set to active, based on an operation by a vehicle occupant. If the exploration mode 1041 is set to "1," the process determines that a learning instruction has been issued and proceeds to step S1403. If the exploration mode 1041 is set to "0," the process returns to step S1402.

[0131] The control unit 103 then acquires steering wheel operation information (step S1403) and analyzes the steering wheel operation in time series (step S1404). The control unit 103 then uses the evaluation index determination unit 106 to determine a sensory index corresponding to the steering wheel operation in step S1405.

[0132] In step S1406, the control unit 103 determines a combination of physical quantities corresponding to the sensory indices determined in step S1405. As described above, this combination of physical quantities is determined randomly, for example. In the following step S1407, the control unit 103 reads the information on the physical quantities determined in step S1406 and the sensory indices provided by the expert driver from the learning object 1021 in the test result storage unit 102. In the following step S1408, the learning unit 107B performs learning using the physical quantities read in step S1407 and the sensory indices provided by the expert driver.

[0133] In step S1409, the control unit 103 determines whether the output error is less than the learning determination threshold 1042. If the control unit 103 determines that the output error is greater than the learning determination threshold 1042, the control unit 103 returns to step S1406 and proceeds to the processing after step S1407 using a different combination of physical quantities. If the control unit 103 determines that the output error is less than the learning determination threshold 1042, the control unit 103 records the combination of physical quantities determined in step S1406 in the data specification setting 113 and stores the parameters learned in step S1408 in the weight parameter storage unit 109.

[0134] According to the third embodiment described above, the following effects are obtained.

[0135] (10) The test result storage unit 102 stores a combination of the output of the behavior sensor and the evaluation value of the sensory index, namely the learning object 1021. The learning unit 107B uses the learning object 1021 to learn the combination of two or more physical quantities contained in the output of the behavior sensor used in the calculation of the evaluation value. The learning unit 107B uses a combination of arbitrarily selected physical quantities to conduct trial learning. When the output error obtained by the learning is less than the learning judgment threshold 1042, the sensory index is associated with the arbitrarily selected multiple physical quantities and recorded in the data specification setting 113. Therefore, it is possible to explore the appropriate combination of physical quantities while performing learning of the hierarchical neural network in the evaluation circuit 130. Furthermore, this exploration helps to clarify the relationship between the steering wheel operation and the vehicle behavior at that time and the sensory evaluation of the occupants.

[0136] -Fourth embodiment-

[0137] Next, use Figure 19 A fourth embodiment of the present invention will be described. In this embodiment, an example of manufacturing a suspension device using a sensory evaluation prediction system will be described.

[0138] Figure 19 This is a block diagram showing the functional configuration of the sensory evaluation prediction system according to the fourth embodiment of the present invention. Figure 1 Compared with the sensory evaluation prediction system 101, Figure 19 The sensory evaluation prediction system 101C shown differs in that it is mounted on a mobile device, further includes a transceiver 901, and is connected to a computer center 150 and an evaluation value collection center 1502 via a network. Furthermore, the sensory evaluation prediction system 101C includes a sensor group 900 in place of the test result storage unit 102.

[0139] The transceiver unit 901 is connected to the computer center 150 via a network such as the Internet, receives learned data such as weight parameters sent from the computer center 150, and outputs the data to the control unit 103. The learned data includes, for example, evaluation index determination data used by the evaluation index determination unit 106 to select a sensory index (evaluation index) to be evaluated for each road surface type from a plurality of sensory indicators, weight parameters stored in the weight parameter storage unit 109, and data specification settings 113.

[0140] The sensor group 900 includes, for example, an acceleration sensor, a gyro sensor, a vehicle speed sensor, a camera, a laser rangefinder, etc. Outputs of the sensor group 900 are input to the input unit 115 .

[0141] The evaluation value collection center 1502 collects and aggregates the evaluation values ​​for each sensory index calculated by the sensory evaluation prediction system 101C as the vehicle travels on various roads, and provides them to the designer 1503. Furthermore, the evaluation value collection center 1502 can be connected to multiple sensory evaluation prediction systems 101C, each mounted on a different vehicle, to collect evaluation values ​​from each sensory evaluation prediction system 101C. Designers 1503, receiving evaluation values ​​from the evaluation value collection center 1502, use these evaluation values ​​as a reference to design the suspension device 1505 and provide the design information to the manufacturing process 1504. In the manufacturing process 1504, which has received the design information, the suspension device 1505 is manufactured using this design information. Thus, the suspension device 1505 can be manufactured based on the evaluation values ​​output from the sensory evaluation prediction system 101C.

[0142] Furthermore, the above describes an example of manufacturing the suspension device 1505 using the evaluation values ​​output from the sensory evaluation prediction system 101C, which is the same as the sensory evaluation prediction system 101 described in the first embodiment. However, the suspension device 1505 can also be manufactured by configuring the sensory evaluation prediction system 101C in the same manner as the sensory evaluation prediction systems 101A and 101B described in the second and third embodiments, respectively, and using the evaluation values ​​output from the sensory evaluation prediction system 101C.

[0143] According to the fourth embodiment of the present invention described above, the following effects are obtained.

[0144] (11) The suspension device 1505 is manufactured based on the evaluation values ​​output from the sensory evaluation prediction system 101C. Therefore, the suspension device 1505 can be easily manufactured based on the evaluation values ​​of each sensory index acquired for various roads, thereby providing a high-performance suspension device that improves the ride quality.

[0145] (Variation 1 of the Fourth Embodiment)

[0146] The sensor group 900 may be installed in a vehicle equipped with the sensory evaluation prediction system 101C instead of the sensor group 900 being provided by the sensory evaluation prediction system 101C. The display unit 111 may be installed in a vehicle equipped with the sensory evaluation prediction system 101C instead of the sensory evaluation prediction system 101C being provided.

[0147] -Fifth embodiment-

[0148] Next, use Figure 20 In this embodiment, an example of a control system that can adjust the damping force of a suspension device using a sensory evaluation prediction system will be described.

[0149] Figure 20 This is a block diagram showing the functional configuration of a suspension control system according to a fifth embodiment of the present invention. Figure 20 The suspension control system shown in the figure is composed of a sensory evaluation prediction system 101D and a suspension damping force variable mechanism 1702. The structure and operation of the sensory evaluation prediction system 101D are the same as those described in the first embodiment. Figure 1 The sensory evaluation prediction system 101 is the same.

[0150] The suspension damping force variable mechanism 1702 adjusts the damping force of a suspension system (not shown) mounted on the vehicle, based on the evaluation values ​​of each sensory index output from the sensory evaluation prediction system 101D. For example, a control command value or control parameter corresponding to the evaluation value is set for the suspension system, which can adjust the damping force based on an externally input control command value or control parameter. This allows adjustments to the suspension system to be made based on the sensory evaluation results obtained by the sensory evaluation prediction system 101D.

[0151] Typically, the damping force characteristics of a suspension system can change due to oil leakage and aging changes in mechanical properties, affecting the vehicle's ride quality. Therefore, in the suspension control system of this embodiment, when a change in evaluation values ​​under similar driving conditions is detected in a vehicle equipped with a sensory evaluation prediction system 101D, the suspension damping force variable mechanism 1702 adjusts the suspension's damping force to eliminate the change. This allows the suspension characteristics to be corrected even in the event of suspension failure or degradation, extending the life of the suspension. Furthermore, the suspension characteristics can be adjusted based on the type of road surface on which the vehicle is traveling. This ensures an optimal ride quality regardless of the road surface.

[0152] Furthermore, the above description describes an example in which the damping force of the suspension device is adjusted via the suspension damping force variable mechanism 1702 using the evaluation value output from the sensory evaluation prediction system 101, which is similar to the sensory evaluation prediction system 101 described in the first embodiment. However, the sensory evaluation prediction system 101D may be configured similarly to the sensory evaluation prediction systems 101A and 101B described in the second and third embodiments, respectively, and the damping force of the suspension device may be adjusted using the evaluation value output from the sensory evaluation prediction system 101D.

[0153] According to the fifth embodiment of the present invention described above, the following effects are obtained.

[0154] (12) The suspension control system includes a sensory evaluation prediction system 101D and a suspension damping force variable mechanism 1702. The suspension damping force variable mechanism 1702 adjusts the damping force of the suspension device mounted on the vehicle based on the evaluation value output from the sensory evaluation prediction system 101D. Therefore, a suspension device can be provided that can extend the service life of the suspension device and provide an optimal ride quality regardless of the type of road surface.

[0155] In the above-mentioned embodiments and variations, the functional block configuration is merely an example. Several functional configurations presented as different functional blocks may be integrated, or a configuration presented as a single functional block diagram may be divided into two or more functions. Furthermore, a configuration may be provided in which a portion of the functions of each functional block is performed by another functional block.

[0156] The above embodiments and modifications may also be combined. Various embodiments and modifications have been described above, but the present invention is not limited to these contents. Other forms that can be considered within the scope of the technical concept of the present invention are also included in the scope of the present invention.

[0157] The disclosure of the following priority basic application is incorporated into this specification by reference:

[0158] Japanese patent application 2020-11277 (filed on January 28, 2020).

[0159] Explanation of symbols

[0160] 101, 101A, 101B, 101C, 101D...Sensory Evaluation Prediction System

[0161] 102…Test result storage unit

[0162] 103…Control Department

[0163] 104…Register

[0164] 105…Selection Department

[0165] 106…Evaluation Index Judgment Department

[0166] 107, 107A…Evaluation Department

[0167] 107B…Study Department

[0168] 108…Total Department

[0169] 109…weight parameter storage unit

[0170] 110…Total result storage unit

[0171] 111…Display unit

[0172] 112…Sensory index setting

[0173] 113…Data specification setting

[0174] 114…Total mode setting

[0175] 115…Input

[0176] 120…Related production department

[0177] 130…Evaluation circuit

[0178] 201…Vehicle

[0179] 401…Evaluation circuit

[0180] 801…steering wheel operation

[0181] 1021…Learning Object

[0182] 1022…Evaluation object

[0183] 1041…Exploration Mode

[0184] 1042…Learning judgment threshold

[0185] 1504…Manufacturing process

[0186] 1505…Suspension

[0187] 1702…Suspension damping force variable mechanism.

Claims

1. A sensory evaluation prediction system, characterized in that: have: an input unit that reads an output of a behavior sensor that measures two or more types of time-series information related to a moving object; a selection unit that selects two or more physical quantities from the output of the behavior sensor read by the input unit; a correlation generating unit that generates information indicating a time-series correlation with respect to the two or more physical quantities selected by the selecting unit; as well as an evaluation circuit that calculates an evaluation value of a sensory index based on the information indicating the correlation in time series, The evaluation circuit can calculate multiple sensory indicators. The sensory evaluation prediction system further has: an evaluation index determination unit that determines a sensory index of an evaluation object based on a steering wheel operation of the mobile object; and a storage unit storing a data specification setting that associates the sensory index determined by the evaluation index determination unit with two or more physical quantities corresponding to the sensory index; The selection unit refers to the data specification setting and determines the two or more physical quantities based on the determination of the evaluation index determination unit. The evaluation circuit includes small evaluation circuits commonly corresponding to the plurality of sensory indices, and the small evaluation circuits are adjusted according to the sensory indices, and the evaluation value is calculated using the adjusted small evaluation circuits.

2. The sensory evaluation prediction system according to claim 1, wherein When the selection unit selects two physical quantities, the correlation creation unit plots the time series correlation of the two physical quantities on a two-dimensional plane and outputs the plot in the form of raster image information.

3. The sensory evaluation prediction system according to claim 1, wherein When the selection unit selects three physical quantities, the correlation generating unit plots the time series correlations of the three physical quantities in a three-dimensional space and outputs the plot in the form of voxel information.

4. The sensory evaluation prediction system according to claim 1, wherein The evaluation index determination unit determines the relationship between the information on the steering wheel operation and the sensory index based on learning.

5. The sensory evaluation prediction system according to claim 1, wherein The data specification setting is different according to the combination of physical quantities for each of the sensory indicators.

6. The sensory evaluation prediction system according to claim 1, wherein: The storage unit also stores a learning object, which is a combination of the output of the behavior sensor and the evaluation value of the sensory index. The sensory evaluation prediction system further includes a learning unit that uses the learning object to learn a combination of two or more physical quantities included in the output of the behavior sensor used for calculating the evaluation value. The learning unit performs trial learning using a combination of arbitrarily selected multiple physical quantities, and when an output error obtained from the learning is smaller than a predetermined learning determination threshold, associates the sensory index with the arbitrarily selected multiple physical quantities and records the results in the data specification setting.

7. The sensory evaluation prediction system according to claim 1, wherein: The sensory evaluation prediction system is mounted on the mobile object. The input unit reads the output of the behavior sensor mounted on the mobile object, The sensory evaluation prediction system further includes a summing unit for summing the calculation results of the evaluation circuit. The totaling unit can switch between an instantaneous evaluation mode that outputs an instantaneous value or a moving average value of the calculation result of the evaluation circuit and a comprehensive evaluation mode that outputs an average value of the calculation result of the evaluation circuit within a predetermined period.

8. A sensory evaluation prediction system, characterized in that: have: an input unit that reads an output of a behavior sensor that measures two or more types of time-series information related to a moving object; a selection unit that selects two or more physical quantities from the output of the behavior sensor read by the input unit; a correlation generating unit that generates information indicating a time-series correlation with respect to the two or more physical quantities selected by the selecting unit; as well as an evaluation circuit that calculates an evaluation value of a sensory index based on the information indicating the correlation in time series, The evaluation circuit can calculate multiple sensory indicators. The sensory evaluation prediction system further has: an evaluation index determination unit that determines a sensory index of an evaluation object based on a steering wheel operation of the mobile object; and a storage unit storing a data specification setting that associates the sensory index determined by the evaluation index determination unit with two or more physical quantities corresponding to the sensory index; The selection unit refers to the data specification setting and determines the two or more physical quantities based on the determination of the evaluation index determination unit. The evaluation circuit has a plurality of evaluation circuits corresponding to the plurality of sensory indices. The sensory evaluation prediction system further includes a control unit configured to stop the evaluation subcircuit that does not calculate the sensory index based on the selection of the evaluation index determination unit.

9. A suspension device, characterized in that: This suspension device is manufactured based on the evaluation value output from the sensory evaluation prediction system according to any one of claims 1 to 6 and 8.

10. A suspension control system, characterized in that: have: The sensory evaluation prediction system according to any one of claims 1 to 6 and 8; and A suspension damping force variable mechanism adjusts the damping force of a suspension device mounted on the mobile body based on the evaluation value output from the sensory evaluation prediction system.

Images