Vehicle Evaluation System

Abstract

To provide a vehicle evaluation system which allows evaluation through discrimination of difference in a degree of a vehicle based on sound data.SOLUTION: A vehicle evaluation system according to the present invention has a processor circuit 110, and a storage device 120. The storage device 120 stores data on a leaned model, and data on centers of mass of respective k clusters obtained by changing magnitude of an error of each of variables in generated data generated by the learned model by using training data into an explanatory variable and clustering by the k-means method. The processor circuit 110 executes error calculation processing for outputting the generated data by inputting evaluation sound data to the learned model and for calculating magnitude of the error of each variable in operated data with respect to each of the variables of the generated data, and evaluation processing for calculating a distance between coordinates of the generated data and the center of mass closest to the coordinates to evaluate an object vehicle 10 according to magnitude of the distance.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a vehicle evaluation system. [Background technology]

[0002] Patent Document 1 discloses an abnormality determination device that determines an abnormality in a power transmission device using a trained model trained by machine learning. This abnormality determination device inputs variables indicating the vehicle operation status by the driver into the trained model as input variables, and causes the trained model to output an estimated value of oil temperature. Then, using the estimated value output by the trained model, this abnormality determination device determines whether the cause of the oil temperature exceeding a determination value is an abnormality in the power transmission device or vehicle operation by the driver. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-017843 Summary of the Invention [Problem to be solved by the invention]

[0004] An evaluation system is being considered that evaluates vehicles by analyzing sound data recorded from the vehicle using a trained model trained by machine learning. However, in order to evaluate a vehicle, it is necessary not only to distinguish from the sound data whether the vehicle is clearly malfunctioning and making abnormal noises, but also to distinguish and evaluate the level of the vehicle from the sound data. Therefore, a vehicle evaluation system suitable for evaluating vehicles is required. [Means for solving the problem]

[0005] The means for solving the above problems and their effects will be described below. A vehicle evaluation system for solving the above problem is a vehicle evaluation system that evaluates a vehicle using sound data recorded from the vehicle to be evaluated. The vehicle evaluation system includes a processing circuit and a storage device. The storage device stores data of a trained model trained by supervised learning to generate operation data from the training sound data, including training sound data recorded while a reference vehicle in a state serving as a reference for evaluation was operated in a measurement driving pattern for a predetermined period of time, and operation data consisting of multiple variables indicating the operating status of the vehicle collected simultaneously with the training sound data. The storage device also stores data on the centers of gravity of k clusters obtained by clustering using the k-means method with the magnitude of error of each variable in the generated data, which is the operation data generated using the trained model, as an explanatory variable. The processing circuit inputs evaluation sound data recorded while a target vehicle, which is a vehicle to be evaluated, is operated according to the measurement driving pattern into the trained model, outputs the generated data, and executes an error calculation process to calculate the magnitude of an error of each of the variables in the generated data from each of the variables in the operation data of the target vehicle collected simultaneously with the evaluation sound data. The processing circuit also executes an evaluation process to calculate the distance between a coordinate of the generated data defined by the magnitude of the error of each of the variables calculated through the error calculation process and the center of gravity closest to the coordinate, and evaluates the target vehicle according to the magnitude of the distance. [Effects of the Invention]

[0006] The magnitude of the error in the generated data from the operational data reflects the difference in the condition of the target vehicle and the reference vehicle. Therefore, the vehicle evaluation system can determine the difference in the condition of the vehicle from the sound data and perform the evaluation. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a schematic diagram showing an embodiment of a vehicle evaluation system. [Figure 2]FIG. 2 is a flowchart showing the flow of processing related to the evaluation of a target vehicle. [Figure 3] FIG. 3 is a graph showing generated data, operational data, and an error in the generated data. [Figure 4] FIG. 4 is a graph showing an example of clustering. [Figure 5] FIG. 5 is a flowchart showing the flow of the error calculation process. [Figure 6] FIG. 6 is a flowchart showing the process flow in the evaluation process. DETAILED DESCRIPTION OF THE INVENTION

[0008] An embodiment of a vehicle evaluation system will be described below with reference to FIGS. <Vehicle evaluation system configuration> As shown in FIG. 1, this vehicle evaluation system includes a data center 100 and a data acquisition device 300. The data center 100 is communicatively connected to the data acquisition device 300 via a communication network 200. As shown in FIG. 1, the data center 100 includes a storage device 120 in which a program is stored, and a processing circuit 110. The processing circuit 110 executes the program stored in the storage device 120 to perform various processes. The data center 100 also includes a communication device 130.

[0009] The data acquisition device 300 is, for example, a personal computer. The data acquisition device 300 includes a storage device 320 in which a program is stored, and a processing circuit 310. The processing circuit 310 executes the program stored in the storage device 320 to perform various processes. The data acquisition device 300 also includes a communication device 330. In this embodiment, the data acquisition device 300 is connected to the data center 100 by wireless communication via the communication network 200. The data acquisition device 300 also includes a display device 340 that displays information. The data acquisition device 300 also includes a microphone 350.

[0010] When using this vehicle evaluation system to evaluate a target vehicle 10, which is a vehicle to be evaluated, a microphone 350 is installed at a predetermined position relative to the target vehicle 10. In addition, a data acquisition device 300 is connected to the vehicle control unit 20 of the target vehicle 10. An operator then operates the target vehicle 10 to operate the target vehicle 10 in a measurement driving pattern. The measurement driving pattern is a driving pattern suitable for acquiring data for evaluation, and is a predetermined driving pattern spanning a predetermined period of time. While the target vehicle 10 is operating in this measurement driving pattern, the data acquisition device 300 records sound with the microphone 350. In addition, the data acquisition device 300 acquires operational data indicating the operating status of the target vehicle 10 simultaneously with recording the sound data.

[0011] The vehicle control unit 20 controls each part of the target vehicle 10. Various sensors that detect the state of the target vehicle 10 are connected to the vehicle control unit 20. When the data acquisition device 300 is connected to the vehicle control unit 20, the data acquisition device 300 can acquire information about the target vehicle 10 through the vehicle control unit 20.

[0012] <Outline of evaluation by the vehicle evaluation system> As described above, in this vehicle evaluation system, when evaluating the target vehicle 10, the data acquisition device 300 is connected to the vehicle control unit 20 of the target vehicle 10. Then, while the target vehicle 10 is operating, the data acquisition device 300 records sound with the microphone 350. The data acquisition device 300 transmits data including data on the recorded sound to the data center 100. Then, the data center 100 evaluates the target vehicle 10 using the received data.

[0013] The data acquisition device 300 records sound data collected by the microphone 350 while the target vehicle 10 is operating in a measurement driving pattern for a predetermined time in the storage device 320 as evaluation sound data. The data acquisition device 300 also stores operational data collected simultaneously with the sound data in the storage device 320. When evaluating the transmission of the target vehicle 10, the operational data may include, for example, the engine rotation speed NE, the input rotation speed Nin, the output rotation speed Nout, and the gear ratio. The data acquisition device 300 then stores the collected operational data, including the evaluation sound data, in the storage device 320 as one data set for a predetermined time.

[0014] The data acquisition device 300 extracts data from a data set for a predetermined time stored in the storage device 320 while changing the extraction start time for each piece of data within a window Tw, which has a time width shorter than the predetermined time, and shapes the extracted data into evaluation data. That is, the data acquisition device 300 extracts data for a period shorter than the predetermined time and shapes the extracted data into evaluation data. Note that, in the data shaping process for shaping the evaluation data, the data acquisition device 300 converts the evaluation sound data into a mel spectrogram, which is frequency-analyzed, and treats the resulting data as image data. In a mel spectrogram, the vertical axis represents frequency, and the frequency on the vertical axis is expressed in mel scale. The horizontal axis represents time. In a mel spectrogram, intensity is represented by color. Lower intensity portions are represented by darker blue colors, and higher intensity portions are represented by brighter red colors. The sound data included in one data set for the predetermined time becomes one mel spectrogram for the predetermined time. The data acquisition device 300 transmits the shaped evaluation data to the data center 100. Upon receiving the evaluation data, the data center 100 stores the evaluation data in the storage device 120. Then, the data center 100 executes the routine shown in FIG. 2 to evaluate the target vehicle 10. This routine is executed by the processing circuit 110 of the data center 100. Note that, hereinafter, the step number of each process is represented by a number preceded by "S."

[0015] 2, the data center 100 performs an error calculation process (S100) and an evaluation process (S200) to evaluate the target vehicle 10. The error calculation process (S100) is a process in which evaluation sound data included in the evaluation data is input into a learned model trained by supervised learning, generated data is output, and the magnitude of the error in the generated data is calculated.

[0016] The trained model is a model trained by supervised learning using a large amount of training data, including training sound data and operational data collected simultaneously with the training sound data, so that operational data can be generated from the training sound data. For example, the data center 100 uses a model that partially uses the image classification model ResNet-18 to handle evaluation sound data as image data. ResNet-18 is a pre-trained image classification model trained on the ImageNet dataset. ResNet-18 has been trained on over one million image data and can classify input images into 1,000 categories. The trained model stored in the storage device 120 of the data center 100 is a model obtained by transfer learning of the pre-trained ResNet-18. This trained model is a model obtained by replacing the classification output layer of ResNet-18 with a neural network MLP and training this neural network MLP by supervised learning. Training involves adjusting the weights and biases of each layer of this neural network MLP. The training sound data is sound data recorded while a reference vehicle in a state that serves as a reference for evaluation is operated in a measurement driving pattern. In this example, the reference vehicle is a vehicle that has been subjected to a certain amount of break-in after manufacture, has undergone sufficient maintenance and inspection, and has been confirmed to have no abnormalities. In other words, the reference vehicle is a vehicle in extremely good condition with almost no deterioration.

[0017] The trained model is optimized to generate operational data from sounds emitted from a reference vehicle. Therefore, if evaluation sound data emitted from a target vehicle 10 in a state different from that of the reference vehicle is input, it is not possible to correctly generate operational status data. In other words, if the state of the target vehicle 10 deviates from the state of the reference vehicle, a discrepancy occurs between the operational data stored in the dataset as correct data and the generated data. A larger discrepancy indicates that the state of the target vehicle 10 deviates that much from the state of the reference vehicle. As described above, the reference vehicle is in extremely good condition with almost no deterioration. Therefore, in this vehicle evaluation system, the smaller the discrepancy, i.e., the smaller the evaluation index value calculated based on the error in the generated data, the closer the state of the target vehicle 10 is to the state of the reference vehicle, and the higher the evaluation is.

[0018] FIG. 3 is a graph showing operational data and generated data generated using a trained model. The graph in FIG. 3 shows only one of the multiple variables included in the operational data. In other words, it shows only one of the objective variables, which are the output of the trained model. In FIG. 3, the operational data, which is the correct data, is shown with a solid line. In FIG. 3, the generated data is represented by a square symbol. Note that time ts in FIG. 3 is the start time of the measurement driving pattern, and time tf is the end time of the measurement driving pattern. In other words, FIG. 3 shows the operational data and generated data in one data set. The trained model outputs one value of the generated data in window Tw for each variable based on the data within that window Tw, which has been shaped into a list. The error ΔPred indicates the deviation of the generated data from the operational data, which is the correct data, i.e., the error. The greater the deviation of the state of the target vehicle 10 from the state of the reference vehicle, the larger the error ΔPred. Therefore, this vehicle evaluation system evaluates the target vehicle 10 using the error ΔPred calculated through the error calculation process.

[0019] As described above, the trained model is optimized to generate operation data from sounds emitted from a reference vehicle. However, depending on the operating state, the error ΔPred that occurs in the generated data generated by the trained model may become large or small. In other words, the trained model may have operating states that are good at generating operation data with high accuracy and operating states that are not good at generating operation data and result in large errors.

[0020] Figure 4 shows the distribution of the error ΔPred in the generated data generated by a trained model using training data. The vertical axis in Figure 4 represents the first error ΔPred1, which is the error ΔPred for the first variable included in the operational data. The horizontal axis in Figure 4 represents the second error ΔPred2, which is the error ΔPred for the second variable included in the operational data. Figure 4 is a diagram showing the distribution of the error ΔPred for all generated data generated from one dataset in a two-dimensional space with the first error ΔPred1 and the second error ΔPred2 as explanatory variables. In Figure 4, the open symbols correspond to each of the multiple generated data generated from one dataset.

[0021] The measurement driving pattern includes various operating states in order to collect sufficient sound data to evaluate the target vehicle 10. Therefore, as shown in Fig. 4, even in the generated data generated using a learned model trained using the training data, an error ΔPred occurs.

[0022] Figure 4 shows the results of clustering using the k-means method with the first error ΔPred1 and the second error ΔPred2 as explanatory variables. The k-means method is a clustering algorithm that classifies data into a pre-specified number of clusters. Figure 4 shows an example where the number of clusters to be classified, k, is set to three. In Figure 4, the generated data classified into the first cluster is indicated by a hollow triangle. In Figure 4, the generated data classified into the second cluster is indicated by a hollow circle. In Figure 4, the generated data classified into the third cluster is indicated by a hollow square. Furthermore, in Figure 4, the center of gravity of each cluster is indicated by a cross. Center of gravity C1 is the center of gravity of the first cluster. Center of gravity C2 is the center of gravity of the second cluster. Center of gravity C3 is the center of gravity of the third cluster.

[0023] In this way, the storage device 120 of the data center 100 stores data on the center of gravity of each of k clusters clustered by the k-means method using the magnitude of the error ΔPred of each variable in the generated data as an explanatory variable. Note that while FIG. 4 shows an example with two explanatory variables, the number of explanatory variables may be equal to the number of variables included in the operational data. Furthermore, the number of clusters k to be classified is not limited to three; it may be two, four, or more.

[0024] This vehicle evaluation system performs evaluation processing using data on the center of gravity of each cluster stored in the storage device 120. The error calculation process and the evaluation process will be described below with reference to flowcharts.

[0025] <About error calculation processing> As described above, upon receiving the evaluation data, the data center 100 stores the evaluation data in the storage device 120. Then, the processing circuit 110 of the data center 100 executes the routine shown in FIG.

[0026] As shown in FIG. 2, when the processing circuit 110 starts this routine, it first executes an error calculation process (S100). 5 is a flowchart showing the flow of processing in the error calculation routine. In the error calculation routine, the processing circuit 110 repeatedly executes this routine until processing is completed for all lists included in the received evaluation data.

[0027] When this routine starts, the processing circuit 110 reads one list of data from the evaluation data (S110). Then, the processing circuit 110 inputs the data of the read list into the learned model to calculate data on the operating status (S120). That is, the processing circuit 110 generates generated data. Next, the processing circuit 110 calculates the magnitude of the error ΔPred of each variable in the generated data from each variable in the operating data collected simultaneously with the evaluation sound data (S130). Then, the processing circuit 110 stores and records the calculated error ΔPred in the storage device 120 (S130). The data of the error ΔPred recorded in the storage device 120 is a list of data for each variable in the operating data. Once the error ΔPred has been recorded, the processing circuit 110 temporarily terminates this routine.

[0028] The processing circuit 110 repeats the processes of S110 to S130 until the calculation and recording of the error ΔPred for all lists included in the evaluation data is completed. Then, when the calculation and recording of the error ΔPred for all lists included in the evaluation data is completed, the processing circuit 110 ends the error calculation process.

[0029] As shown in FIG. 2, after the error calculation process (S120) is completed, the processing circuit 110 next executes the evaluation process (S200). <About evaluation processing> 6 is a flowchart showing the flow of processing in the evaluation processing routine. When this routine starts, the processing circuit 110 reads one piece of data on the error ΔPred recorded in the storage device 120 (S210). Then, the processing circuit 110 calculates the distance Dist and records the distance Dist in the storage device 120.

[0030] As shown in FIG. 4, the distance Dist is the distance between the coordinates of the generated data defined by the magnitude of the error ΔPred of each variable read through the processing of S210 and the centroid of each of the k clusters that is closest to the coordinates.

[0031] FIG. 4 shows an example of the distance Dist. In FIG. 4, the coordinates of the generated data are indicated by black square symbols. In this case, the centroid closest to the coordinates is centroid C3. Note that the distance Dist may be calculated using any distance calculation method, such as Euclidean distance, Mahalanobis distance, or Manhattan distance. A calculation method suitable for evaluation may be adopted.

[0032] 6, the processing circuit 110 determines whether all the data for the error ΔPred for the evaluation data has been processed (S230). If it determines that the processing of all the data for the error ΔPred has not been completed (S230: NO), the processing circuit 110 returns the processing to S210. Then, the processing circuit 110 reads another piece of data for the unprocessed error ΔPred (S210) and executes the processing from S220 onwards again.

[0033] If it is determined that all data for the error ΔPred has been processed (S230: YES), the processing circuit 110 proceeds to S240. Then, the processing circuit 110 calculates an evaluation index value (S240). Here, the processing circuit 110 calculates the sum of all distances Dist for the error ΔPred of the generated data for a predetermined time period stored in the storage device 120, and sets this as the evaluation index value. After calculating the evaluation index value in this way, the processing circuit 110 proceeds to S250.

[0034] The processing circuit 110 determines an evaluation rank based on the evaluation index value and outputs the evaluation rank (S250). The processing circuit 110 determines the evaluation rank by selecting an evaluation rank according to the magnitude of the evaluation index value from four evaluation ranks: S rank, A rank, B rank, and C rank. S rank is the highest evaluation rank among the four evaluation ranks. C rank is the lowest evaluation rank among the four evaluation ranks. The evaluations decrease in order from S rank to A rank, B rank, and C rank. The processing circuit 110 transmits the evaluation rank to the data acquisition device 300 and outputs the evaluation rank (S250). After outputting the evaluation rank in this way, the processing circuit 110 ends this routine. That is, the processing circuit 110 ends the evaluation process and ends the series of routines shown in FIG. 2.

[0035] In this way, the evaluation process is a process in which the sum of the distance Dist calculated for each data window Tw over a predetermined time period is calculated as an evaluation index value, and the target vehicle 10 is evaluated according to the magnitude of the distance Dist.

[0036] The data acquisition device 300 receives the evaluation rank and displays the received evaluation rank on the display device 340 as the evaluation rank of the target vehicle 10 . <Operation of this embodiment> The data center 100 executes a generation process using the trained model to generate generated data by restoring operation data from the evaluation data. The trained model is a neural network that uses the feature quantities of data extracted from a predetermined time's worth of data as explanatory variables and the operating status at the time corresponding to the extracted data as a target variable. The data center 100 then executes an error calculation process (S100) to calculate an error ΔPred in the generated data.

[0037] In the evaluation process (S200), the data center 100 calculates the distance Dist between the coordinates of the generated data defined by the magnitude of the error ΔPred and the closest center of gravity among the centers of gravity of each of the k clusters.The data center 100 then calculates the sum of the distances Dist over a predetermined time period as an evaluation index value.The data center 100 determines the evaluation rank of the target vehicle 10 based on the evaluation index value.

[0038] <Effects of this embodiment> (1) The difference between the generated data and the operational data, i.e., the magnitude of the error ΔPred, reflects the difference in the condition of the target vehicle 10 and the reference vehicle. Therefore, the vehicle evaluation system can determine the difference in the condition of the target vehicle 10 from the sound data and perform an evaluation.

[0039] (2) Depending on the operating state, the error ΔPred in the generated data generated by the trained model may become large or small. That is, even a trained model may have operating states that are good at generating operating data with high accuracy and operating states that are not good at generating operating data, resulting in a large error ΔPred. If the magnitude of the error ΔPred is uniformly reflected in the evaluation without considering such effects, the accuracy of the evaluation may decrease. In contrast, this vehicle evaluation system evaluates the target vehicle 10 using the distance Dist from the center of gravity of the cluster closest to the generated data. That is, this vehicle evaluation system evaluates the target vehicle 10 according to the distance Dist from the center of gravity of the cluster that has a similar tendency in the magnitude of the error ΔPred of each variable. Therefore, this vehicle evaluation system can improve the accuracy of the evaluation compared to reflecting the magnitude of the error ΔPred in the evaluation on a uniform scale.

[0040] (3) The above vehicle evaluation system divides data for a predetermined time into multiple sections and analyzes them. The results are then integrated to calculate evaluation index values. Therefore, the above vehicle evaluation system requires a smaller trained model than when analyzing data for a predetermined time all at once.

[0041] (4) The vehicle evaluation system outputs the evaluation results by applying them to a preset evaluation rank. Therefore, this evaluation system makes it easier to understand the relative level of the condition of the target vehicle 10 in the used car market.

[0042] <Example of change> This embodiment can be modified as follows: This embodiment and the following modifications can be combined and implemented within the scope of technical compatibility.

[0043] The data used as evaluation data does not have to be a mel spectrogram. For example, a spectrogram obtained by wavelet transforming sound data may be used. A spectrogram obtained by short-time Fourier transforming sound data may also be used. It is not necessary to convert sound data into image data. For example, features may be extracted from the sound data itself and used as evaluation data. In this case, there is no need to use ResNet-18, which handles image data, as a trained model. Also, although a model using transfer learning of ResNet-18 has been shown as an example, the model structure is not limited to this. The trained model only needs to be able to output generated data based on the evaluation data.

[0044] The number of evaluation ranks does not have to be four. For example, the number of evaluation ranks may be increased. Conversely, the number of evaluation ranks may be decreased. Although an example of determining the evaluation rank based on the evaluation index value has been shown as the evaluation process, the evaluation process is not limited to this mode. For example, the evaluation index value may be output as is, with a larger value indicating a lower evaluation, and displayed on the display device 340.

[0045] In the above vehicle evaluation system, an example is shown in which a vehicle in extremely good condition is used as the reference vehicle to evaluate the target vehicle 10. The reference vehicle does not necessarily have to be a vehicle in good condition. For example, a vehicle in extremely poor condition with a low rating can also be used as the reference vehicle. The evaluation index value in the above evaluation process is a value that indicates the degree of deviation between the condition of the target vehicle 10 and that of the reference vehicle. Therefore, if an aging vehicle with an extremely low rating is used as the reference vehicle, the smaller the evaluation index value, the lower the rating. The target vehicle 10 can also be evaluated using such evaluation index values.

[0046] A vehicle evaluation system can also be configured using the data acquisition device 300 by storing data on the trained model and data on the center of gravity of each cluster in the storage device 320 of the data acquisition device 300. In this case, the processing circuit 310 of the data acquisition device 300 executes the error calculation process and the evaluation process.

[0047] The data acquisition device 300 itself does not need to be equipped with a microphone 350. Sound data can be acquired from an external device and subjected to data shaping processing, generation processing, and evaluation processing. The generation processing and evaluation processing may be performed using multiple sound data recorded using multiple microphones 350.

[0048] In the above embodiment, an example was shown in which the operation data included data on the rotational speed of a rotating shaft in the powertrain as an example of evaluating the transmission of the target vehicle 10. The vehicle evaluation system may also evaluate the target vehicle 10 by evaluating other units in the target vehicle 10. For example, to evaluate an engine, data on the engine rotational speed NE, ignition timing, and engine load factor may be used as operation data. For example, to evaluate a drive motor, data on the rotational speed of the motor's output shaft may be used as operation data. To evaluate a four-wheel drive vehicle, data on the rotational speed of each drive wheel may be used as operation data. The vehicle evaluation system may also evaluate the target vehicle 10 by referring to the evaluation index values ​​of various units.

[0049] In the above embodiment, an example was shown in which the error ΔPred of each variable included in the operational data was used as an explanatory variable in clustering. However, in addition to information about the error ΔPred, the actual operational data itself at the time the sound data was collected may be included as an explanatory variable in clustering. In this case, the vehicle evaluation system will perform evaluation using the clustering results that reflect the operating conditions under which the data was collected. This allows the vehicle evaluation system to perform evaluations with higher accuracy.

[0050] In the above embodiment, an example was shown in which the error ΔPred of each variable included in the operational data is used as an explanatory variable in clustering. However, in addition to information about the error ΔPred, data indicating the measurement conditions when the sound data was collected may be included as an explanatory variable for clustering. For example, information about the outside air temperature and information about the oil temperature may be included as explanatory variables. In this case, the vehicle evaluation system will perform evaluation using clustering results that reflect the measurement conditions under which the data was collected. This allows the vehicle evaluation system to perform evaluations with higher accuracy.

[0051] In the above embodiment, an example was shown in which the sum of the distances Dist was used as the evaluation index value. However, the evaluation index value may be the sum of values ​​obtained by standardizing the distance Dist in consideration of the degree of dispersion of the data distribution in each cluster. For example, the evaluation index value may be the sum of the quotients obtained by dividing the distance Dist by the average value of the distances between the center of gravity used for the calculation and each piece of data classified into a cluster defined by that center of gravity. This allows the magnitude of the distance Dist to be reflected in the evaluation, taking into account the spread of the distribution in each cluster.

[0052] In the above embodiment, the vehicle evaluation system includes a processing circuit 110 and a storage device 120 to execute software processing. However, this is merely an example. For example, the vehicle evaluation system may include a dedicated hardware circuit (e.g., an ASIC) that processes at least part of the software processing executed in the above embodiment. That is, the vehicle evaluation system may have any of the following configurations (A) to (C). (A) The vehicle evaluation system includes an execution device that executes all processing in accordance with a program and a storage device that stores the program. That is, the vehicle evaluation system includes a software execution device. (B) The vehicle evaluation system includes an execution device that executes part of the processing in accordance with the program and a storage device. Furthermore, the vehicle evaluation system includes a dedicated hardware circuit that executes the remaining processing. (c) The vehicle evaluation system includes a dedicated hardware circuit that executes all processing. Here, there may be multiple software execution devices and / or dedicated hardware circuits. That is, the above processing may be executed by a processing circuitry that includes at least one of one or more software execution devices and one or more dedicated hardware circuits. A storage device, i.e., a computer-readable medium, that stores a program includes any available medium accessible by a general-purpose or dedicated computer. [Explanation of symbols]

[0053] 10...Target vehicle, 20...Vehicle control unit, 100...Data center, 110...Processing circuit, 120...Storage device, 130...Communication device, 200...Communication network, 300...Data acquisition device, 310...Processing circuit, 320...Storage device, 330...Communication device, 340...Display device, 350...Microphone

Claims

1. A vehicle evaluation system that evaluates a vehicle to be evaluated using sound data that records sounds emitted from the vehicle, the system comprising: a processing circuit; and a storage device; Data of a trained model trained by supervised learning to generate the operation data from the training sound data using training data including training sound data recorded while a reference vehicle in a state that serves as a reference for evaluation is operated in a measurement driving pattern for a predetermined period of time, and operation data consisting of a plurality of variables that indicate the operation status of the vehicle collected simultaneously with the training sound data; and Data of the centers of gravity of k clusters clustered by the k-means method using the magnitude of error of each variable in the generated data, which is the operation data generated using the trained model, as an explanatory variable; and is stored in the storage device, an error calculation process in which evaluation sound data recorded while a target vehicle, which is a vehicle to be evaluated, is operated according to the measurement driving pattern is input into the trained model, the generated data is output, and the magnitude of the error of each variable in the generated data from each variable in the operation data of the target vehicle collected simultaneously with the evaluation sound data is calculated; an evaluation process for calculating a distance between a coordinate of the generated data defined by the magnitude of the error of each variable calculated through the error calculation process and the center of gravity closest to the coordinate, and evaluating the target vehicle according to the magnitude of the distance; The processing circuit executes the vehicle evaluation system.

2. The evaluation process calculates, as an evaluation index value, a sum of the distances calculated for each of the data for a period shorter than the predetermined time, extracted from the data for the predetermined time while changing the extraction start time. The vehicle evaluation system according to claim 1 .

3. The explanatory variables include the operational data. The vehicle evaluation system according to claim 1 .

4. The explanatory variables include data indicating the measurement conditions when the sound data was collected. The vehicle evaluation system according to claim 1 .

5. The operation data includes data on the rotation speed of a rotating shaft in a powertrain. The vehicle evaluation system according to claim 1 .

Images