Information processing system, information processing program, and information processing method

Abstract

To appropriately evaluate estimation accuracy of first model data.SOLUTION: An information processing system comprises an execution device and a storage device. The storage device stores first model data, which is model data preliminarily generated by machine learning, and second model data, which is model data preliminarily generated by a rule base. The execution device determines whether the first estimation results, which are the estimation results of a failure cause of a power transmission device output by inputting first travel data into the first model data, match the second estimation results, which are the estimation results of a failure cause of the power transmission device output by inputting second travel data into the second model data (S63). The execution device increases the reliability of the first model data when the first estimation results match the second estimation results (S71).SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The present invention relates to an information processing system, an information processing program, and an information processing method.

Background Art

[0002] The failure analysis support device of Patent Document 1 includes an execution device and a storage device. The storage device stores model data using a decision tree. That is, this model data is data of a tree-structured model composed of a plurality of nodes and a plurality of leaf nodes. The model data leads to a terminal leaf node by repeating branches according to pre-determined branch conditions for each node. The leaf node indicates the cause of a failure in equipment such as a plant. The execution device receives input of information such as the content of the occurring failure. Then, while referring to the model data, the execution device estimates the cause of the failure in the equipment based on the above information.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When estimating the cause of a failure in equipment or the like using model data as in the technology disclosed in Patent Document 1, there may be a case where it is desired to evaluate the estimation accuracy of the cause of the failure by the model data. However, Patent Document 1 does not mention anything about how to evaluate the estimation accuracy.

Means for Solving the Problems

[0005] An information processing system for solving the above problems comprises an execution device and a storage device, the storage device storing first model data which is model data pre-generated by machine learning using first driving data which is various data acquired by sensors mounted on the vehicle and combinations of failure causes of the power transmission system mounted on the vehicle associated with the first driving data as training data, and second model data which is model data pre-generated by rule-based methods of second driving data which is various data acquired by sensors mounted on the vehicle and combinations of failure causes of the power transmission system associated with the second driving data, the first model data is input when the first driving data is input This enables the output of a first estimation result, which is an estimated result of the cause of failure of the power transmission device, and the second model data enables the output of a second estimation result, which is an estimated result of the cause of failure of the power transmission device, when the second driving data is input to it. The execution device performs the following actions: acquire the first driving data, input the acquired first driving data into the first model data to output the first estimation result, acquire the second driving data, input the acquired second driving data into the second model data to output the second estimation result, and increase the reliability of the first model data if the outputted first estimation result and second estimation result match.

[0006] The information processing program for solving the above problems is for an information processing system comprising an execution device and a storage device, wherein the storage device stores: first model data, which is model data pre-generated by machine learning using first driving data, which is various data acquired by sensors mounted on the vehicle, and combinations of failure causes of the power transmission device mounted on the vehicle associated with the first driving data as training data; and second model data, which is model data pre-generated by rule-based methods using second driving data, which is various data acquired by sensors mounted on the vehicle, and combinations of failure causes of the power transmission device associated with the second driving data, wherein the first model data is the first driving data The system can output a first estimation result, which is an estimation result of the cause of failure of the power transmission device, when the input of the second driving data, and the second model data can output a second estimation result, which is an estimation result of the cause of failure of the power transmission device, when the input of the second driving data, and the execution device is made to perform the following: acquire the first driving data, input the acquired first driving data into the first model data to output the first estimation result, acquire the second driving data, input the acquired second driving data into the second model data to output the second estimation result, and increase the reliability of the first model data if the output first estimation result and the second estimation result match.

[0007] The information processing method for solving the above problems relates to an information processing system comprising an execution device and a storage device, wherein the storage device stores: first model data, which is model data pre-generated by machine learning using first driving data, which is various data acquired by sensors mounted on the vehicle, and combinations of failure causes of the power transmission device mounted on the vehicle associated with the first driving data, as training data; and second model data, which is model data pre-generated by rule-based methods using second driving data, which is various data acquired by sensors mounted on the vehicle, and combinations of failure causes of the power transmission device associated with the second driving data, wherein the first model data is the first driving data The first estimated result, which is the estimated cause of failure of the power transmission device, can be output when the second driving data is input to the second model data, which is the estimated cause of failure of the power transmission device, and the execution device performs the following: acquire the first driving data, input the acquired first driving data into the first model data to output the first estimated result, acquire the second driving data, input the acquired second driving data into the second model data to output the second estimated result, and increase the reliability of the first model data if the output first estimated result and second estimated result match. [Effects of the Invention]

[0008] According to the above configuration, the reliability of the first model data increases when the first and second estimation results match, meaning that the estimation result using the first model data can be judged as correct. In this way, by comparing the first estimation result using the first model data with the second estimation result using second model data separate from the first model data, a reliability level for evaluating the estimation accuracy of the first model data can be obtained. This allows for an appropriate evaluation of the estimation accuracy of the first model data. [Brief explanation of the drawing]

[0009] [Figure 1] Figure 1 is a schematic diagram of the information processing system and the vehicle. [Figure 2] Figure 2 is an explanatory diagram of the first model data. [Figure 3] Figure 3 is a flowchart showing the reliability output control. [Modes for carrying out the invention]

[0010] <Outline of the vehicle configuration> An embodiment of the present invention will be described below with reference to Figures 1 to 3. First, the vehicle 200 targeted by the information processing system 100 will be described.

[0011] As shown in Figure 1, the vehicle 200 is equipped with a control device 210. The control device 210 controls various devices of the vehicle 200, including the drive source and power transmission device. Here, the power transmission device is a device in the vehicle 200 that transmits power from the drive source to the drive wheels. An example of a power transmission device is a so-called automatic transmission. An example of a drive source for the vehicle 200 is an internal combustion engine. The control device 210 also stores various data acquired by sensors mounted on the vehicle 200.

[0012] <Outline configuration of the information processing system> Next, we will describe the general configuration of the information processing system 100. As shown in Figure 1, the information processing system 100 includes a first estimation device 10. The first estimation device 10 comprises a first device body 20, an input device 30, and a display 40. The input device 30 includes, for example, a keyboard and a pointing device. The display 40 is capable of displaying various types of information.

[0013] The first device body 20 includes a CPU 21 and a memory 22. The memory 22 includes a read-only ROM, a read and write volatile RAM, and a read and write non-volatile storage. The memory 22 pre-stores various programs and various data. Specifically, the memory 22 pre-stores a control program 22A as one of the various programs. Furthermore, the memory 22 pre-stores first model data DM1 as one of the various data. The first model data DM1 describes the relationship between predetermined input variables and output variables that show the estimated cause of failure of the power transmission system of the vehicle 200 in a format that the CPU 21 can execute. The first model data DM1 is pre-generated by machine learning. A detailed explanation of the first model data DM1 will be given later. In addition, the memory 22 pre-stores first accuracy Z1 and test data DT as data associated with the first model data DM1. Here, first accuracy Z1 is the pre-calculated estimation accuracy of the first model data DM1. Furthermore, the test data DT is the test data used when calculating the first accuracy Z1. In addition, the memory 22 pre-stores the reference data DR. The reference data DR is used to identify the estimation result from the second model data DM2, which will be described later. A detailed explanation of the reference data DR will be given later. The CPU 21 performs various processes, which will be described later, by executing the control program 22A stored in the memory 22. An example of the first device body 20 is a so-called personal computer.

[0014] The first device body 20 acquires various data from the input device 30. The first device body 20 also outputs control signals to the display 40, thereby displaying various information on the display 40.

[0015] As shown in Figure 1, the information processing system 100 includes a second estimation device 60. The second estimation device 60 includes a second device body 70 and a connecting cable 80. The second device body 70 of the second estimation device 60 can communicate wirelessly with the first device body 20 of the first estimation device 10 via a communication network (not shown). Furthermore, when the second device body 70 of the second estimation device 60 is connected to the control device 210 of the vehicle 200 via the connecting cable 80, the second device body 70 can communicate with the control device 210 of the vehicle 200. Therefore, the second device body 70 of the second estimation device 60 can acquire various data obtained by sensors mounted on the vehicle 200. In this embodiment, the first device body 20 of the first estimation device 10 can acquire various data obtained by sensors mounted on the vehicle 200 via the second estimation device 60. In this embodiment, the second estimation device 60 is an inspection and diagnostic device installed in a repair shop such as an automobile dealer.

[0016] The second device body 70 includes a CPU 71 and a memory 72. The memory 72 includes a read-only ROM, a read-and-write volatile RAM, and a read-and-write non-volatile storage. The memory 72 pre-stores various programs and various data. Specifically, the memory 72 pre-stores a control program 72A as one of the various programs. Furthermore, the memory 72 pre-stores second model data DM2 as one of the various data. The second model data DM2 describes the relationship between predetermined input variables and output variables that indicate the estimated cause of failure of the vehicle 200's power transmission system in a format that the CPU 71 can execute. In addition, the second model data DM2 is pre-generated using a rule-based method. Therefore, the second model data DM2 is pre-generated to output results corresponding to inputs according to rules previously discovered by humans. A detailed explanation of the second model data DM2 will be given later. The CPU 71 executes the control program 72A stored in the memory 72 to realize the various processes described later.

[0017] In this embodiment, CPU 21 and CPU 71 are examples of execution devices. Memory 22 and memory 72 are examples of storage devices. Furthermore, control program 22A and control program 72A are examples of information processing programs. Therefore, the execution devices, CPU 21 and CPU 71, realize various processes of the information processing method by executing control program 22A and control program 72A, which are information processing programs.

[0018] <First Estimation Control> Next, with reference to Figure 2, the first estimation control performed by the first device body 20 will be described. This first estimation control is a control for outputting the estimated result of the cause of failure of the power transmission system. For example, when an operator performs an operation to perform the first estimation control via the input device 30, the first device body 20 performs the first estimation control based on the first driving data DD1 acquired from the vehicle 200 that is the target of the first estimation control. Here, the first driving data DD1 is various data acquired by sensors mounted on the vehicle 200 while the vehicle 200 is driving. In this embodiment, the first driving data DD1 includes data for predetermined timings for the driving distance TD, accelerator operation amount ACC, vehicle speed SP, engine rotation speed NE, oil temperature TA, water temperature TB, and set gear GS. Here, an example of a predetermined timing is the timing when the warning lamp indicating that an abnormality has been detected in the vehicle 200 lights up. The first driving data DD1 also includes time-series data including predetermined timings for the hydraulic pressure PA. In this embodiment, the time-series data for hydraulic pressure PA is data for a specified period, from a certain period before a specified timing to a certain period after a specified timing. An example of a specified period is a few tenths of a second to a few seconds. Here, mileage TD is the distance traveled by vehicle 200 from the time of manufacture to the present. Accelerator operation amount ACC is the amount the accelerator pedal is operated by the driver of vehicle 200. Vehicle speed SP is the speed of vehicle 200. Engine rotation speed NE is the number of rotations per unit time of the crankshaft of the internal combustion engine in vehicle 200. Oil temperature TA is the temperature of the oil circulating in the automatic transmission of vehicle 200. Water temperature TB is the temperature of the coolant circulating in the internal combustion engine, which is the power source of vehicle 200. Set gear GS is the gear set in the automatic transmission of vehicle 200. Hydraulic pressure PA is the pressure of the oil circulating in the automatic transmission of vehicle 200. As described above, in this embodiment, the first device body 20 of the first estimation device 10 can acquire the first driving data DD1 of the vehicle 200 via the second estimation device 60.

[0019] In the first estimation control, the CPU 21 of the first device main body 20 outputs an estimation result of the cause of failure of the power transmission device by inputting the acquired first travel data DD1 into the first model data DM1. Here, as shown in FIG. 2, an example of the first model data DM1 is model data using a so-called decision tree. Therefore, for example, the first model data DM1 includes nodes N11 to N15 and leaf nodes N21 to N26. The nodes N11 to N15 define branching conditions in advance. Further, the leaf nodes N21 to N26 define information regarding whether the power transmission device is normal and information regarding the cause of failure of the power transmission device in advance. In the first estimation control, for example, when the first travel data DD1 satisfies all the branching conditions defined in the nodes N11 to N13, the CPU 21 specifies that the estimation result of the cause of failure of the power transmission device corresponds to the information defined in the leaf node N21. That is, the CPU 21 outputs, as an estimation result of the cause of failure of the power transmission device, that component a included in the power transmission device has failed.

[0020] The first model data DM1 is generated in advance as follows, for example. First, an engineer who generates the first model data DM1 prepares a plurality of power transmission devices. Here, the plurality of power transmission devices include those with failures and those without failures. Further, the engineer identifies the true cause of failure in the power transmission device by, for example, disassembling the power transmission device with a failure. Furthermore, the engineer acquires the first travel data DD1 acquired from the vehicle 200 in which the power transmission device is mounted for each of the above power transmission devices. Then, the engineer generates the first model data DM1 by machine learning using the first travel data DD1 and the combination of the true cause of failure associated with the first travel data DD1 as teacher data.

[0021] <Second Estimation Control> Next, the second estimation control executed by the second device main body 70 will be described. This second estimation control is for outputting an estimation result of the cause of a failure in the power transmission device. For example, when an operator performs an operation for executing the second estimation control via a switch or the like (not shown), the second device main body 70 executes the second estimation control based on second travel data DD2 acquired from the vehicle 200 targeted by the second estimation control. Here, the second travel data DD2 is various data acquired by sensors mounted on the vehicle 200 when the vehicle 200 is traveling. In the present embodiment, the second travel data DD2 is time-series data including travel distance TD, accelerator operation amount ACC, vehicle speed SP, engine rotational speed NE, oil temperature TA, water temperature TB, set gear stage GS, and oil pressure PA at predetermined specified timings. Here, an example of the specified timing is the timing when an alarm lamp indicating that an abnormality has been detected in the vehicle 200 lights up. In the present embodiment, the time-series data for the travel distance TD and the like is data for a specified period from a certain period before the specified timing to a certain period after the specified timing with respect to the specified timing. Note that an example of the specified period is about several comma seconds to several seconds. As described above, in the present embodiment, when the second device main body 70 of the second estimation device 60 and the control device 210 of the vehicle 200 are connected via the connection cable 80, the second device main body 70 can acquire the second travel data DD2 of the vehicle 200.

[0022] In the second estimation control, the CPU 71 of the second device body 70 outputs an estimated result of the cause of failure of the power transmission system by inputting the acquired second driving data DD2 into the second model data DM2. In this embodiment, the second model data DM2 predefines multiple conditions and multiple estimation results. For example, if the second driving data DD2 satisfies the first condition defined in the second model data DM2, the CPU 71 identifies that it corresponds to the estimation result associated with the first condition in the second model data DM2. Here, the estimation result from the second model data DM2, like the first model data DM1, includes information on whether the power transmission system is normal or not, and information on the cause of failure of the power transmission system. As described above, the second model data DM2 is pre-generated by rule-based methods. On the other hand, the first model data DM1 is pre-generated by machine learning. Therefore, the first estimation result, which is the estimation result from the first model data DM1, and the second estimation result, which is the estimation result from the second model data DM2, may or may not match. Furthermore, if the CPU 71 of the second device body 70 outputs a second estimation result, the CPU 71 of the second device body 70 outputs a fault code CB indicating the second estimation result. In this embodiment, the fault code CB is used by workers at repair shops such as automobile dealerships when performing maintenance on the vehicle 200. For example, a worker can determine the cause of the power transmission failure from the fault code CB by referring to a repair manual that has a pre-associated fault code CB with the cause of the power transmission failure.

[0023] The second model data DM2 is pre-generated, for example, as follows: First, the engineer generating the second model data DM2 prepares multiple power transmission devices. These multiple power transmission devices include those that are malfunctioning and those that are not. The engineer then identifies the true cause of failure in the malfunctioning power transmission device, for example, by disassembling the power transmission device. Furthermore, for each of the above power transmission devices, the engineer obtains the second driving data DD2 acquired from the vehicle 200 on which the power transmission device is installed. The engineer then generates the second model data DM2 using a rule-based system of combinations of the second driving data DD2 and the true cause of failure associated with that second driving data DD2.

[0024] <Reliability Output Control> Next, with reference to Figure 3, the reliability output control performed by the first device body 20 will be described. This reliability output control is a control for outputting the reliability DC of the first model data DM1. In this embodiment, the initial value of reliability DC is the same as the value of the first precision Z1. Therefore, the initial value of reliability DC is a value from "0%" to "100%". Furthermore, a larger reliability DC value indicates that the first estimation result is more likely to match the true cause of failure. In addition, a larger reliability DC value indicates that the first estimation result is more likely to match the second estimation result. Therefore, reliability DC can be used as an indicator for evaluating the first precision Z1, which is the estimation accuracy of the first model data DM1. In this embodiment, the first device body 20 performs reliability output control each time it acquires the first travel data DD1 and the fault code CB based on the second travel data DD2 obtained at the same timing as the first travel data DD1.

[0025] As shown in Figure 3, when the CPU 21 of the first device body 20 starts reliability output control, it executes the process in step S61. In step S61, the CPU 21 inputs the acquired first running data DD1 to the first model data DM1 and outputs the estimated result of the failure cause of the power transmission device. That is, the CPU 21 outputs the first estimation result, which is the estimation result based on the first model data DM1. The process in step S61 is the same as the first estimation control described above. After step S61, the CPU 21 proceeds to step S62.

[0026] In step S62, the CPU 21 identifies a second estimated result, which is the estimation result from the second model data DM2, based on the fault code CB. Specifically, the CPU 21 identifies the second estimated result, which is the estimation result from the second model data DM2, by applying the fault code CB to the reference data DR. Here, the reference data DR shows the correspondence between the fault code CB and the cause of the power transmission failure, similar to the repair manual described above. After step S62, the CPU 21 proceeds to step S63.

[0027] In step S63, the CPU 21 determines whether the first estimation result from step S61 and the second estimation result from step S62 match. If the CPU 21 determines in step S63 that the first estimation result from step S61 and the second estimation result from step S62 match (S63: YES), the CPU 21 proceeds to step S71.

[0028] In step S71, the CPU 21 calculates the corrected reliability DC by correcting the reliability DC at the time of processing in step S71 to a higher value. For example, the CPU 21 calculates the corrected reliability DC by adding a predetermined value to the reliability DC at the time of processing in step S71. If the value obtained by adding the predetermined value to the reliability DC at the time of processing in step S71 is greater than the upper limit of "100%", the CPU 21 sets the corrected reliability DC to "100%". After step S71, the CPU 21 terminates the reliability output control for this step.

[0029] On the other hand, if the CPU 21 determines in step S63 that the first estimation result in step S61 and the second estimation result in step S62 do not match (S63: NO), the CPU 21 proceeds to step S81.

[0030] In step S81, the CPU 21 generates new first model data DM1. Specifically, the CPU 21 corrects the predetermined branching conditions for the nodes from the root node to the leaf node showing the first estimation result among the multiple nodes included in the first model data DM1, so that the likelihood of leading to a different leaf node than the aforementioned leaf node increases.

[0031] For example, as shown in Figure 2, assume that the first estimation result in step S61 is information defined by leaf node N21. In this case, the CPU 21 identifies nodes N11 to N13 as the nodes from the root node N11 to the leaf node N21 that shows the first estimation result, out of the multiple nodes N11 to N15 included in the first model data DM1. The CPU 21 then corrects the branching conditions predetermined for nodes N11 to N13 so that the likelihood of leading to a different leaf node than the leaf node N21 that shows the first estimation result increases. In this embodiment, the CPU 21 corrects the branching conditions as follows. In the above case, the branching condition A predetermined for node N11 is satisfied when going from the root node N11 to the leaf node N21 that shows the first estimation result. Therefore, in order to increase the likelihood of leading to a leaf node different from leaf node N21, which shows the first estimation result, CPU 21 corrects the predetermined branching condition A for node N11 so that it becomes less likely to be met. For example, suppose that the predetermined branching condition A for node N11 is that the mileage TD is greater than or equal to a predetermined first reference value. In this case, CPU 21 corrects branching condition A by multiplying the first reference value by a predetermined correction coefficient to increase the corrected first reference value, that is, to make it less likely that branching condition A will be met. In the above example, CPU 21 uses a predetermined fixed value greater than "1" as the correction coefficient. Also, if it is to correct branching condition A so that it becomes more likely to be met, CPU 21 uses a predetermined fixed value less than "1" as the correction coefficient. In the same manner as above, CPU 21 corrects the predetermined branch condition B for node N12 and the predetermined branch condition C for node N13. As shown in Figure 3, after step S81, CPU 21 proceeds to step S82.

[0032] In step S82, the CPU 21 uses the test data DT to newly calculate the first accuracy Z1, which is the estimated accuracy of the first model data DM1 generated in step S81. Here, the test data DT is data that is not used as training data, and includes the first driving data DD1 and the combination of true failure causes associated with the first driving data DD1. As mentioned above, the test data DT is the same test data that was used when calculating the first accuracy Z1 which was stored in memory 22 at the start of step S82. After step S82, the CPU 21 proceeds to step S83.

[0033] In step S83, the CPU 21 updates the first model data DM1 up to the start of step S81 with the first model data DM1 generated in step S81. In other words, the CPU 21 sets the first model data DM1 generated in step S81 as the first model data DM1 to be used in the first estimation control. The CPU 21 also updates the first precision Z1 up to the start of step S82 with the first precision Z1 calculated in step S82. Then, the CPU 21 updates the initial value of the confidence level DC with the first precision Z1 calculated in step S82. After step S83, the CPU 21 terminates the confidence level output control for this step.

[0034] <Operation of this embodiment> In this embodiment, the first model data DM1 is pre-generated by machine learning. On the other hand, the second model data DM2 is pre-generated by rule-based methods. Therefore, when the first estimation result, which is the estimation result from the first model data DM1, and the second estimation result, which is the estimation result from the second model data DM2, coincide, there is a high probability that the first estimation result, which is the estimation result from the first model data DM1, is correct. Accordingly, as shown in Figure 3, in step S63 of the confidence output control, the CPU 21 proceeds to step S71 if the first estimation result from step S61 and the second estimation result from step S62 coincide. In step S71, the CPU 21 increases the corrected confidence DC by correcting the confidence DC at the time of processing in step S71.

[0035] <Effects of this embodiment> (1) According to this embodiment, the confidence level DC of the first model data DM1 is high when it can be determined that the first estimation result, which is the estimation result from the first model data DM1, is correct. By comparing the first estimation result from the first model data DM1 with the second estimation result from the second model data DM2 in this way, a confidence level DC for evaluating the estimation accuracy of the first model data DM1 can be obtained. This makes it possible to appropriately evaluate the estimation accuracy of the first model data DM1.

[0036] (2) Generally, when the first estimation result, which is the estimation result from the first model data DM1, and the second estimation result, which is the estimation result from the second model data DM2, do not match, it is highly likely that the first estimation result, which is the estimation result from the first model data DM1, is incorrect. In other words, it is highly likely that the result shown by a different leaf node in the first model data DM1 than the leaf node that shows the first estimation result in this case is correct.

[0037] In this regard, in step S63 of the confidence output control, if the first estimation result in step S61 and the second estimation result in step S62 do not match, the CPU 21 proceeds to step S81. In step S81, the CPU 21 generates new first model data DM1. Specifically, the CPU 21 corrects the predetermined branching conditions for the nodes from the root node to the leaf node showing the first estimation result among the multiple nodes included in the first model data DM1, so as to increase the likelihood of leading to a different leaf node than the aforementioned leaf node. Therefore, in a situation where the first estimation result, which is the estimation result from the first model data DM1, is incorrect, correcting the branching conditions of the nodes in the first model data DM1 increases the likelihood of leading to a different leaf node than the leaf node showing the current first estimation result. By correcting the first model data DM1 in this way, the possibility of the estimation result from the first model data DM1 being incorrect is suppressed. As a result, the estimation accuracy of the first model data DM1 can be improved.

[0038] (3) Generally, hydraulic pressure PA, which is the pressure of the oil circulating in the automatic transmission of vehicle 200, is more likely to change rapidly than oil temperature TA, which is the temperature of the oil circulating in the automatic transmission of vehicle 200. In this regard, the first driving data DD1 includes time-series data that includes specified timings for hydraulic pressure PA. Therefore, compared to the case where the first driving data DD1 only includes data for specified timings for hydraulic pressure PA, the changes in hydraulic pressure PA are reflected in the first driving data DD1. As a result, even in situations where hydraulic pressure PA changes rapidly, the estimation accuracy of the first model data DM1 can be improved by inputting the first driving data DD1, which takes into account the changes in hydraulic pressure PA, into the first model data DM1.

[0039] (4) The first driving data DD1 contains predetermined timing data for mileage TD, accelerator pedal operation ACC, vehicle speed SP, engine rotation speed NE, oil temperature TA, water temperature TB, and set gear GS. Therefore, the data size of the first driving data DD1 is smaller than if it were time-series data for mileage TD, etc. This can reduce the processing load on the CPU 21 when estimating using the first model data DM1.

[0040] <Example of changes> This embodiment can be implemented with the following modifications. This embodiment and the following modifications can be combined with each other to the extent that they do not contradict each other technically.

[0041] In the above embodiment, the reliability output control may be modified. For example, the first driving data DD1 may be changed. Specifically, the first driving data DD1 may be predetermined timing data for the distance traveled TD, accelerator pedal operation ACC, vehicle speed SP, engine rotation speed NE, oil temperature TA, water temperature TB, set gear position GS, and hydraulic pressure PA. Also, specifically, the first driving data DD1 may be time-series data including predetermined timing for the distance traveled TD, accelerator pedal operation ACC, vehicle speed SP, engine rotation speed NE, oil temperature TA, water temperature TB, set gear position GS, and hydraulic pressure PA. Furthermore, specifically, the first driving data DD1 may include other values ​​in addition to, or instead of, the distance traveled TD, accelerator pedal operation ACC, vehicle speed SP, engine rotation speed NE, oil temperature TA, water temperature TB, set gear position GS, and hydraulic pressure PA. Other values ​​that can be used include the operating status of solenoids in the power transmission system.

[0042] For example, the method of generating the new first model data DM1 in step S81 may be changed. Specifically, the correction coefficient used when correcting the branching conditions is not limited to a fixed value, but may also be a variable value. In this case, the correction coefficient may be changed depending on the leaf node showing the first estimation result.

[0043] For example, other processes may be used instead of the processes in steps S81 to S83. Specifically, if the CPU 21 determines that the result is negative in step S63, it may perform a process to calculate the corrected confidence level DC by correcting the confidence level DC to a lower value.

[0044] For example, the method of correcting the confidence score DC in step S71 may be changed. As a specific example, suppose the first estimation result in step S61 is information defined by leaf node N21. In this case, in step S71, CPU21 may correct only the confidence score DC corresponding to leaf node N21 that shows the first estimation result. In other words, the confidence score DC may be an index for evaluating the estimation accuracy of each leaf node included in the first model data DM1.

[0045] For example, the initial value of confidence level DC may be changed. Specifically, from the perspective of evaluating the estimation accuracy of the first model data DM1, the initial value of confidence level DC may be any value. In this configuration, one example of an initial value for confidence level DC is zero.

[0046] The configuration of the information processing system 100 may be changed. For example, the first estimation device 10 and the second estimation device 60 may be configured as a single unit. In this case, one of the CPUs 21 and 71 may be omitted. The other CPU 21 and 71 corresponds to the execution device. Similarly, one of the memory 22 and 72 may be omitted. The other memory 22 and 72 corresponds to the storage device. [Explanation of Symbols]

[0047] DM1…First model data DM2…Second model data 10…First estimation device 20…First device main unit 21…CPU 22…Memory 22A…Control program 30…Input device 40…Display 60…Second estimation device 70…Second device main unit 71…CPU 72…Memory 72A…Control program 80…Connection cable 100…Information processing system 200…Vehicle 210…Control device

Claims

1. It comprises an execution device and a storage device, The aforementioned storage device is Using first driving data, which is various data acquired by sensors mounted on the vehicle, and combinations of failure causes of the power transmission system mounted on the vehicle linked to the first driving data, as training data, a first model data is pre-generated model data by machine learning, and Second driving data, which is various data acquired by sensors mounted on the vehicle, and second model data, which is pre-generated model data based on a rule-based system of combinations of failure causes of the power transmission system linked to the second driving data, I remember that, The first model data is model data using a decision tree, and when the first driving data is input, it can output a first estimation result which is the estimation result of the cause of failure of the power transmission device. The second model data can output a second estimation result, which is an estimation result of the cause of failure of the power transmission device, when the second driving data is input. The execution device is To acquire the aforementioned first driving data, The acquired first driving data is input into the first model data to output the first estimation result, To acquire the aforementioned second driving data, The acquired second driving data is input into the second model data to output the second estimation result, When the outputted first estimation result and the second estimation result match, the reliability of the first model data is increased. If the outputted first estimation result and the second estimation result do not match, the predetermined branching conditions for the nodes from the root node to the leaf node showing the first estimation result among the multiple nodes included in the first model data are corrected to increase the likelihood of leading to a different leaf node than the leaf node, Execute Information processing system.

2. The first driving data is, Data at predetermined timings regarding the amount of accelerator pedal operation performed by the driver of the vehicle, the speed of the vehicle, and the temperature of the oil circulating in the power transmission system, The power transmission device comprises time-series data including the specified timing for the pressure of the oil circulating in the power transmission device. The information processing system according to claim 1.

3. This applies to information processing systems that include an execution device and a memory device. The aforementioned storage device is Using first driving data, which is various data acquired by sensors mounted on the vehicle, and combinations of failure causes of the power transmission system mounted on the vehicle linked to the first driving data, as training data, a first model data is pre-generated model data by machine learning, and Second driving data, which is various data acquired by sensors mounted on the vehicle, and second model data, which is pre-generated model data based on a rule-based system of combinations of failure causes of the power transmission system linked to the second driving data, I remember that, The first model data is model data using a decision tree, and when the first driving data is input, it can output a first estimation result which is the estimation result of the cause of failure of the power transmission device. The second model data can output a second estimation result, which is an estimation result of the cause of failure of the power transmission device, when the second driving data is input. The execution device, To acquire the aforementioned first driving data, The acquired first driving data is input into the first model data to output the first estimation result, To acquire the aforementioned second driving data, The acquired second driving data is input into the second model data to output the second estimation result, When the outputted first estimation result and the second estimation result match, the reliability of the first model data is increased. If the outputted first estimation result and the second estimation result do not match, the predetermined branching conditions for the nodes from the root node to the leaf node showing the first estimation result among the multiple nodes included in the first model data are corrected to increase the likelihood of leading to a different leaf node than the leaf node, Make it run Information processing program.

4. This applies to information processing systems that include an execution device and a memory device. The aforementioned storage device is Using first driving data, which is various data acquired by sensors mounted on the vehicle, and combinations of failure causes of the power transmission system mounted on the vehicle linked to the first driving data, as training data, a first model data is pre-generated model data by machine learning, and Second driving data, which is various data acquired by sensors mounted on the vehicle, and second model data, which is pre-generated model data based on a rule-based system of combinations of failure causes of the power transmission system linked to the second driving data, I remember that, The first model data is model data using a decision tree, and when the first driving data is input, it can output a first estimation result which is the estimation result of the cause of failure of the power transmission device. The second model data can output a second estimation result, which is an estimation result of the cause of failure of the power transmission device, when the second driving data is input. The execution device, To acquire the aforementioned first driving data, The acquired first driving data is input into the first model data to output the first estimation result, To acquire the aforementioned second driving data, The acquired second driving data is input into the second model data to output the second estimation result, When the outputted first estimation result and the second estimation result match, the reliability of the first model data is increased. If the outputted first estimation result and the second estimation result do not match, the predetermined branching conditions for the nodes from the root node to the leaf node showing the first estimation result among the multiple nodes included in the first model data are corrected to increase the likelihood of leading to a different leaf node than the leaf node, Execute Information processing methods.

Images