JP2022157881A - Engine oil property predictive model generating program, and engine oil property predictive model generated thereby - Google Patents

Abstract

To provide a program and a server for generating a predictive model that can accurately predict a deterioration of engine oil with high versatility regardless of engine structures or types of engine oil.SOLUTION: A program is to: machine-learn, as explanatory variables, at least one of first vehicle parameters including vehicle type, engine displacement, vehicle size, maximum oil load, total vehicle mileage and mileage for a predetermined period of time, at least one of second vehicle parameters including engine oil usage mileage, and a property parameter indicating a property of the engine oil; generate a predictive model for predicting the property of the engine oil; verify the predictive model using at least one parameter of the first vehicle parameters, at least one parameter of the second vehicle parameters, and the property parameter indicating the property of the engine oil; and determine the predictive model satisfying favorable requirements as a favorable predictive model.SELECTED DRAWING: Figure 3

Description

The present invention relates to a program for generating a predictive model for predicting deterioration of engine oil used in vehicle engines.

Vehicle engines use engine oil for lubrication and cooling. The engine oil is exposed to high temperatures in the engine, and the pistons of the engine move at high speed and are subject to wear. Therefore, as the mileage of the vehicle increases, various substances accumulate in the engine oil due to heat load and wear, and the engine oil deteriorates.

In the past, the remaining life of the engine oil was predicted from the temperature and contamination coefficient of the engine oil (see, for example, Patent Document 1), and the pentane insoluble content and total acid value were determined from the infrared absorbance and the mileage. There are methods for predicting the degree of deterioration of oil (see, for example, Patent Document 2), and methods for determining oil life from operating conditions using the annual mileage and period of use (see, for example, Patent Document 3).

Patent No. 6557549 JP-A-9-100712 Patent No. 6292544

Previous techniques for predicting engine oil degradation are for a specific engine design and can be difficult to predict for different engine designs. In addition, there are many cases in which prediction is made with a specific engine oil, and prediction with respect to different engine oils is difficult in some cases.

The present invention has been made in view of the above points, and its object is to increase versatility and accurately predict the deterioration of engine oil regardless of the engine structure or the type of engine oil. An object of the present invention is to provide an engine oil property prediction model generation program for generating an engine oil property prediction model.

Features of the engine oil property prediction model generation program according to the present invention are:
data for at least one of first vehicle parameters including one or more of vehicle type, engine displacement, vehicle size, maximum oil load, total vehicle mileage, and mileage over a predetermined period; and engine oil usage mileage. Data for at least one of the second vehicle parameters and data for property parameters indicating properties of the engine oil are machine-learned as explanatory variables to determine the properties of the engine oil. a machine learning step of generating at least one predictive model for predicting
Verification of data of at least one of the first vehicle parameters, data of at least one of the second vehicle parameters, and data of property parameters indicating properties of engine oil and a good prediction model determination step of verifying the at least one prediction model using verification data used in and determining a prediction model that satisfies good requirements as a good prediction model.

It is possible to provide an engine oil property prediction model generation program capable of generating a prediction model capable of accurately predicting deterioration of engine oil with high versatility regardless of the engine structure or the type of engine oil.

FIG. 4 shows a table (a) showing an example of vehicle characteristic parameters used in predicting the property of engine oil according to the present embodiment, and a table (b) showing an example of property parameters. 4 is a flow chart showing an overview of the overall flow of engine oil property prediction according to the present embodiment. FIG. 2 is a functional block diagram of property prediction of engine oil according to the present embodiment; FIG. 4 is a diagram showing the concept of correspondence in engine oil property prediction according to the present embodiment. FIG. 5 is a diagram showing an example of a list of relationships between property parameters and vehicle characteristic parameters obtained from an engine oil property prediction model according to the present embodiment; 4 is a table showing examples of combinations of vehicle-related parameters; 4 is a table showing examples of combinations of vehicle-related parameters; 4 is a table showing examples of combinations of vehicle-related parameters;

<<<<outline of the present embodiment>>>>
<<First Embodiment>>
According to the first embodiment,
data for at least one of first vehicle parameters including one or more of vehicle type, engine displacement, vehicle size, maximum oil load, total vehicle mileage, and mileage over a predetermined period; and engine oil usage mileage. Data for at least one of the second vehicle parameters and data for property parameters indicating properties of the engine oil are machine-learned as explanatory variables to determine the properties of the engine oil. a machine learning step of generating at least one predictive model for predicting
Verification of data of at least one of the first vehicle parameters, data of at least one of the second vehicle parameters, and data of property parameters indicating properties of engine oil A good prediction model determination step of verifying the at least one prediction model using the verification data used in and determining a prediction model that satisfies the good requirements as a good prediction model. An engine oil property prediction model generation program is provided. be done.

The engine oil property prediction model generation program is a program for generating a prediction model for predicting the property of engine oil. The properties of engine oil can be predicted by the generated prediction model. The engine oil property prediction model generation program includes a machine learning process and a good prediction model determination process.

The machine learning process performs machine learning using the learning data as explanatory variables to generate at least one prediction model for predicting properties of the engine oil. A prediction model generated by machine learning is a candidate prediction model. The machine learning process is not the process of generating the final predictive model (good predictive model determination), but the process of generating at least one candidate predictive model for selecting the final predictive model. The learning data is data used for machine learning among the data of the property parameters and the vehicle feature parameters.

The vehicle characteristic parameter has a first vehicle parameter and a second vehicle parameter. The first vehicle parameter is one or more of vehicle type, engine displacement, vehicle size, maximum oil load, total vehicle mileage, and mileage for a predetermined period of time. A second vehicle parameter is a parameter that includes engine oil usage mileage. The second vehicle parameters do not include the first vehicle parameters. The second vehicle parameter may include the engine oil usage distance. That is, the vehicle characteristic parameters always include the engine oil usage mileage and consist of a combination of the engine oil usage mileage and other vehicle parameters.

That is, in order to generate a prediction model for predicting the property of engine oil, machine learning is performed using a combination of data of engine oil usage distance and other vehicle parameters, and property parameters as explanatory variables. In other words, machine learning is performed by using the engine oil usage distance as a basic vehicle parameter and combining it with other vehicle parameters as a parameter for supplementing the engine oil usage distance. By combining not only the engine oil usage distance but also other vehicle parameters according to the property of the engine oil, it is possible to generate a prediction model that can more accurately predict the property of the engine oil to be predicted.

The good prediction model determination step includes data of at least one parameter among the first vehicle parameters, data of at least one parameter of the second vehicle parameters, and data of property parameters indicating properties of the engine oil. , verification data used for verification is used. The verification data may be different from all of the learning data, or may partially overlap.

At least one prediction model is verified by the verification data, and a prediction model satisfying a good condition is determined as a good prediction model. That is, at least one candidate prediction model is verified to finally determine a good prediction model. For example, the good prediction model determination process calculates the coefficient of determination between the property parameter value obtained using the prediction model generated in the machine learning process and the property parameter value of the verification data, and determines whether it is good or not from the coefficient of determination. It may be determined whether

In this way, it is possible to provide an engine oil property prediction model generation program for generating an engine oil property prediction model, and an engine oil property prediction model generated by the engine oil property prediction model generation program.

<<Second Embodiment>>
A second embodiment is the first embodiment,
The machine learning process is at least one of random forest, linear regression, decision tree, support vector machine, Gaussian process regression, and ensemble learning with boosting.

A machine learning algorithm capable of generating an optimum prediction model may be appropriately selected according to the combination of the engine oil usage distance and other vehicle parameters.

<<Third Embodiment>>
A third embodiment is the first embodiment,
The machine learning step includes:
A bootstrap sampling module that randomly extracts from the learning data while allowing duplication;
a decision tree generation module that generates a plurality of decision trees using the explanatory variables;
an evaluation module that evaluates a prediction model composed of a set of decision trees;
have

The machine learning process has a bootstrap sampling module, a decision tree generation module and an evaluation module.

The bootstrap sampling module is a module for randomly extracting sample data, which is a subset, from learning data, which is a population, in machine learning (reconstruction sampling).

The decision tree generation module is a module that makes predictions based on conditions that can be determined with YES or NO. The decision tree determines the condition for the value of the feature value itself.

The evaluation module is a module for inputting data having actual measured values of oil properties that have been actually analyzed to a prediction model generated using data for learning and making trial calculations of actual measured values of oil properties.

In addition, the evaluation module can evaluate the dependence of the vehicle parameters on the selected explanatory variables, the stability of the OOB error rate, and the like.

<<Fourth Embodiment>>
A fourth embodiment is the first to third embodiments,
The property parameter includes at least one unit characteristic parameter indicating a unit characteristic of engine oil,
At least one parameter selected from the first vehicle parameters and at least one parameter selected from the second vehicle parameters are associated with one unit characteristic parameter, and one unit characteristic parameter is associated with At least one corresponding prediction model is generated.

The property parameters include at least one unit property parameter indicating unit properties of the engine oil. For example, the unit property can be base number, Cu, Fe, residual carbon content, kinematic viscosity, Pb, and the like. The unit characteristic may be anything as long as it indicates the characteristic of the engine oil. The unit characteristic may be appropriately determined according to the type of engine oil.

At least one parameter selected from the first vehicle parameters and at least one parameter selected from the second vehicle parameters are associated with one unit characteristic parameter, and as a result of machine learning, one A predictive model is generated for at least one candidate corresponding to a unit characteristic parameter of . At least one candidate predictive model is generated for each unit characteristic parameter.

Further, a good prediction model is determined from at least one candidate prediction model by the good prediction model determination step. That is, the machine learning step generates at least one candidate prediction model for each unit characteristic parameter, and the good prediction model determination step determines a good prediction model for each unit characteristic parameter. In this way, an appropriate good predictive model can be generated for each of the unit characteristic parameters.

<<Fifth Embodiment>>
According to the fifth embodiment,
A server capable of providing the engine oil property prediction model generation program of the first embodiment is provided. Since the engine oil property prediction model generation program can be provided, it is possible to provide the engine oil property prediction model generation program for generating an appropriate engine oil property prediction model according to the engine structure and the type of engine oil.

<<Sixth Embodiment>>
According to the sixth embodiment,
A server capable of providing an engine oil property prediction model generated by the engine oil property prediction model generation program of the first embodiment is provided.

<<<<details of the present embodiment>>>>
The engine oil property prediction model generation program and the engine oil property prediction model generated by the program according to the present embodiment predict oil property and performance of engine oil from various types of vehicle information. Although the engine oil property prediction model generation program will be mainly described below, a server capable of providing the engine oil property prediction model generation program and an engine oil property prediction model generated by the engine oil property prediction model generation program can be provided. server is also explained.

Embodiments will be described below with reference to the drawings. FIG. 1 shows a table (a) showing an example of vehicle characteristic parameters used in predicting the property of engine oil according to the present embodiment, and a table (b) showing an example of property parameters. FIG. 2 is a flow chart showing an outline of the overall flow of engine oil property prediction according to the present embodiment. FIG. 3 is a functional block diagram of engine oil property prediction according to the present embodiment. FIG. 4 is a diagram showing an example of a correspondence relationship in engine oil property prediction according to the present embodiment. FIG. 5 is a diagram showing an example of a list of relationships between property parameters and vehicle characteristic parameters obtained from the engine oil property prediction model according to the present embodiment. 6A to 6C are tables showing examples of combinations of vehicle-related parameters.

<<<flow of engine oil in internal combustion engine>>>
Engine oil is supplied to various parts of the internal combustion engine by an engine oil pump. For example, engine oil is supplied in the vicinity of various movable members such as cylinders and crankshafts.

Engine oil circulates during operation of the internal combustion engine. Therefore, the engine oil gradually deteriorates due to metal friction of moving parts of the internal combustion engine, temperature, and the like. Depending on the degree of deterioration, the function of the engine oil may not be fully exhibited.

<Functions of engine oil>
Engine oil has various functions such as a lubricating function, a sealing function, a cooling function, a cleaning and dispersing function, and an antirust function. The lubricating function is a function for reducing metal friction and moving movable members smoothly. The sealing function is a function for sealing the gap between the piston and the piston ring to maintain airtightness. A cooling function is a function for absorbing and releasing heat generated by combustion or the like. The cleaning and dispersing function is a function for taking in dirt inside the internal combustion engine generated by combustion. The antirust function is a function for protecting the internal combustion engine from rust and corrosion caused by moisture and acid in the internal combustion engine.

Like the lubricating and cleaning/dispersing functions of engine oil, engine oil becomes contaminated as it is used. That is, as the vehicle runs, the engine oil deteriorates. As the engine oil deteriorates, it becomes unable to fully exhibit various functions such as lubricating function and cleaning/dispersing function. Therefore, it is necessary to manage the state of the engine oil (degree of deterioration, etc.).

<<<vehicle characteristic parameters and property parameters>>>
In the property prediction of engine oil according to the present embodiment, data of various parameters required for property prediction of engine oil are accumulated, and machine learning is performed using the accumulated data to generate a prediction model. Parameters used for property prediction of engine oil include vehicle characteristic parameters and property parameters.

The vehicle characteristic parameters are parameters characterized by vehicle specifications, such as vehicle engine displacement and vehicle total mileage, vehicle travel history, and the like. Vehicle characteristic parameters determined by vehicle specifications can be obtained in advance. Further, vehicle characteristic parameters, such as vehicle travel history, which are determined by the operation of the engine, can be obtained from a travel recording device or the like.

A property parameter is a parameter characterized by various substances accumulated in the engine oil. Various substances such as metals are gradually eluted into the engine oil due to wear of movable members and the like. The properties (degree of deterioration) of engine oil are determined by the types and amounts of substances eluted in the engine oil. The properties of the engine oil can be acquired when the engine is not in operation, such as when the vehicle is not in business, on holidays, or during maintenance. By extracting engine oil while the engine is not in operation and analyzing the extracted engine oil with an analyzer, it is possible to obtain the types and amounts of substances eluted in the engine oil.

The vehicle characteristic parameter and property parameter data thus acquired are collected at a predetermined timing in a storage device such as an engine oil property prediction system or device (for example, a data storage unit 100 to be described later), and gradually accumulated in By accumulating, data of explanatory variables for machine learning can be constructed.

Vehicle feature parameters function as explanatory variables in machine learning, which will be described later. A property parameter functions as a target variable in machine learning.

FIG. 1(a) is a table showing an example of vehicle characteristic parameters used in predicting the property of engine oil according to the present embodiment. FIG. 1(b) is a table showing an example of property parameters used in predicting the property of engine oil according to the present embodiment.

<<Vehicle Characteristic Parameters>>
Vehicle characteristic parameters include parameters that do not change over time and parameters that change over time.

<Parameters that do not change over time>
Parameters that do not change over time are parameters that are uniquely determined by specifications such as vehicle type and model. The parameters that do not change over time are, for example, the vehicle size, vehicle type, engine displacement, maximum oil load, and payload among the vehicle characteristic parameters shown in FIG. 1(a).

<Vehicle size and vehicle type>
Vehicle sizes can be small, medium, large, and the like. Vehicle types include trucks, buses, sedans, and SUVs. The vehicle size and vehicle type are uniquely determined when the vehicle specifications are determined. The vehicle size and vehicle type are not limited to these classifications, and can be further subdivided or classified in other ways. Vehicle size and vehicle type can be related to engine oil degradation.

<Engine displacement>
The engine displacement is a specific numerical value that is uniquely determined by the vehicle type and model. It should be noted that the engine displacement may be large, medium, small, etc., instead of a specific numerical value. In general, the larger the engine displacement, the greater the load on the engine oil, and the smaller the engine displacement, the smaller the load on the engine oil. Therefore, engine displacement is related to engine oil degradation.

<Maximum oil load>
In this embodiment, maximum oil load is defined as rated power/volume of engine oil. Rated power is the mechanical power that can be extracted from an internal combustion engine. Rated power can be the maximum power output of an internal combustion engine that can be safely achieved under specified conditions. The rated output is uniquely determined when the vehicle specifications are determined. Generally, when the rated power is low, the load on the engine oil is small, and when the rated power is high, the load on the engine oil is large. Maximum oil load is related to engine oil degradation.

The amount of engine oil is the total amount of engine oil that can be supplied to the internal combustion engine. The amount of engine oil is uniquely determined when the specifications of the vehicle, such as the capacity of the engine oil tank and the route for holding the engine oil, are determined. Generally, when the amount of engine oil is large, the load on the engine oil is small, and when the amount of engine oil is small, the load on the engine oil is large.

Maximum oil load is defined as rated power/volume of engine oil and is the maximum power that can be drawn from the internal combustion engine for the total amount of engine oil. The maximum oil load is a parameter that indicates whether the total amount of engine oil is designed with a sufficient margin relative to the rated output or a design with a reduced margin. In this way, the maximum oil load is a parameter that reflects vehicle design concepts such as vehicle type, model, target customer, price range, vehicle concept, and vehicle manufacturer.

<Load capacity>
The payload is the weight of the load that can be loaded on the vehicle. Generally, when the load is small, the load on the engine oil is small, and when the load is large, the load on the engine oil is large. Load capacity can be related to engine oil degradation.

<Parameters that change over time>
Parameters that change over time are parameters that change as the vehicle travels (uses). This is information that can be determined according to the running state of the vehicle. For example, among the vehicle characteristic parameters shown in FIG. 1(a), there are an oil usage distance, a total engine mileage, a monthly mileage, and the like. Data on parameters that change over time can be acquired by a travel recording device or the like mounted on the vehicle.

<Oil usage distance>
The oil usage distance is the distance traveled while continuously using the engine oil based on the time when the engine oil was replaced. If the oil usage distance is short, the deterioration of the engine oil has not progressed so much, and if the oil usage distance is long, the deterioration of the engine oil has progressed. In general, the oil usage distance is shorter than the total engine mileage, which will be described later, and longer than the monthly mileage.

<Engine total mileage>
Total engine mileage is the distance the vehicle has traveled since it was manufactured. The total engine mileage indicates the state of the vehicle from the time of manufacture to the present, and mainly indicates the degree of use due to running of the vehicle. In general, the total engine mileage is longer than the oil usage distance described above and the monthly mileage described later.

If the total mileage of the engine is short, the vehicle is close to a new vehicle, so the wear of the components of the internal combustion engine has not progressed, and various substances are less likely to be eluted into the engine oil. On the other hand, if the total engine mileage is long, the vehicle has been used, the wear of the internal combustion engine members is progressing, and various substances are likely to be eluted into the engine oil. In the case of a vehicle with a long total engine mileage, various substances tend to be easily eluted even when the engine oil is changed.

<Monthly mileage>
The monthly travel distance is the distance traveled by the vehicle in one month. The monthly mileage indicates the state of the vehicle for the most recent month, and mainly indicates the usage frequency of the vehicle for the month. If the monthly mileage is short, the frequency of use is low, and there is a high possibility that the engine oil has not deteriorated in a short period of time. On the other hand, if the monthly mileage is long, the frequency of use is high and the engine oil is constantly exposed to high temperatures, so there is a high possibility that the engine oil will deteriorate in a short period of time.

When the total mileage of the engine is short and the mileage per month is long, the vehicle is close to a new vehicle, but the frequency of use is high, and the engine oil tends to deteriorate rapidly in a short period of time. In addition, when the total engine mileage is long and the monthly mileage is short, although the vehicle has been used, the frequency of use is low, and there is a possibility that the engine oil will not rapidly deteriorate.

<<property parameter>>
The property parameters include, for example, the kinematic viscosity of the engine oil and the types and amounts of substances eluted in the engine oil. Materials include Cu, Fe, residual carbon, Pb, and the like. A base number is also included in the property parameters.

As described above, engine oil has a lubricating function and a cleaning/dispersing function. Therefore, as the internal combustion engine operates, worn metals, carbon generated by combustion, and the like are taken into the engine oil. If the amount of these substances is small, the engine oil can sufficiently exhibit the effects of lubricating function and cleaning/dispersing function, but if the amount of these substances is large, the engine oil cannot sufficiently exhibit the effects.

The degree of deterioration can be determined by the property parameter, and the speed of deterioration can be determined by the change rate of deterioration.

<Prediction of deterioration of engine oil in unspecified vehicle>
Even if the total vehicle mileage is the same, the amount of elution into the engine oil due to wear and the tendency of change differ depending on the vehicle type. In particular, different vehicle manufacturers and different vehicle types have different design concepts, so there is a high possibility that the amount of elution into the engine oil and the tendency of change will also differ. By building a prediction model generation program and generation method that can predict the deterioration of engine oil, even for unspecified vehicles, data on vehicle characteristic parameters and property parameters can be collected and machine-learned for prediction. Can generate models. By constructing a prediction model generation program and generation method, it is possible to predict deterioration of engine oil for unspecified vehicles.

<<<Engine oil property prediction model generation process>>>
FIG. 2 is a flow chart showing the overall flow of engine oil property prediction model generation processing according to the present embodiment. Although the processing shown in FIG. 2 is shown as a program executed entirely by the central processing unit of the computer, some processing may be performed manually. Manual processing can be defined accordingly.

First, vehicle-related parameter data and property parameter data are collected (step S311). As previously described, vehicle-related parameter data and property parameter data are collected when the engine is not running or, if possible, is running. The collected data can be used to generate prediction models by machine learning, and can be used to verify the generated prediction models. For example, 80% of the collected data is used to generate the prediction model and the remaining 20% is used to validate the prediction model. Note that this ratio is not limited to these, and can be appropriately changed according to the total number of data, the type of machine learning, the characteristics of the data, and the like. Further, among the collected data, the data used for generating the prediction model and the data used for verification may be handled completely separately or partially overlapped.

Next, a combination of vehicle-related parameters and property parameters is selected for machine learning (step S313).

For example, vehicle-related parameters include oil usage distance, total engine mileage, engine displacement, vehicle size, monthly mileage, vehicle type, maximum oil load, etc., as shown in FIG. 1(a). The property parameters are base number, Cu, Fe, residual carbon content, kinematic viscosity, Pb, etc., as shown in FIG. 1(b).

In the present embodiment, all combinations of vehicle-related parameters are associated with each of the base number, Cu, Fe, residual carbon content, kinematic viscosity, and Pb, which are property parameters. 6A to 6C show some combinations (A-1-1) to (H-1-1) of vehicle-related parameters as examples of combinations of vehicle-related parameters. Preferred vehicle-related parameters for predicting each of the property parameters by associating all combinations of vehicle-related parameters with base number, Cu, Fe, residual carbon content, kinematic viscosity, and Pb, and performing machine learning. A parameter combination can be obtained.

(A-1-1) to (A-7-1) shown in FIG. 6A are part of the combination (A-1-1) that all include the oil usage distance. In this embodiment, in the process of step S313, all combinations of vehicle-related parameters including the oil usage distance are selected. In other words, each of the base number, Cu, Fe, residual carbon content, kinematic viscosity, and Pb, which are property parameters, is associated with all combinations of vehicle-related parameters, including the oil usage distance, and machine-learned. In this way, one vehicle-related parameter (oil usage distance) is always combined to search for other vehicle-related parameters related to deterioration, including the oil usage distance that is likely to be related to engine oil deterioration. be able to. Furthermore, by always combining one vehicle-related parameter (oil usage distance), the time required for machine learning can be reduced.

Each time step S313 is executed, it selects one of all combinations of vehicle-related parameters including oil usage mileage. The process of step S313 is repeatedly executed until all combinations of vehicle-related parameters including the oil usage distance are executed by the determination of step S327, which will be described later.

Next, data for machine learning of the combination of parameters selected in the process of step S313 is read (step S315), and an engine oil property candidate prediction model is generated by the random forest method using the read data (step S317). .

For example, in the process of step S313, when two vehicle-related parameters, the oil usage distance and the total engine mileage, are selected and associated with the base number, which is the property parameter, the base number and the oil usage distance An engine oil property candidate prediction model for predicting the base number using the oil usage distance and the total engine mileage is generated from the data for machine learning and the total engine mileage. In this way, engine oil property candidate prediction models are generated for each combination of parameters selected and associated in the process of step S313.

Next, data for verifying the combination of parameters selected in the process of step S313 is read out (step S319), and the value of the property parameter is predicted (calculated) using the engine oil property candidate prediction model generated in step S317. ) (step S321). For example, the base number is calculated from the engine oil property candidate prediction model for predicting the generated base number by reading data for verification of the base number, oil usage distance, and total engine travel distance.

Next, the coefficient of determination between the value of the property parameter calculated in step S321 and the value of the actual verification data is calculated (step S323), and the various coefficients of the generated prediction model and the coefficients calculated in step S321 The coefficient of determination is stored (step S325). The coefficient of determination is a coefficient indicating the degree of difference between the value calculated from the engine oil property candidate prediction model and the actually measured value (scale of fit). That is, it is a coefficient that indicates how much the measured value can be explained by the engine oil property candidate prediction model. The coefficient of determination has a value of 0 or more and 1 or less, and the closer it is to 1, the better the engine oil property prediction model fits.

Next, it is determined whether or not other combinations remain, that is, whether or not engine oil property candidate prediction models have been generated for all combinations of vehicle-related parameters associated with each property parameter (step S327). ). If other combinations remain (NO), the process returns to step S313.

If engine oil property candidate prediction models have been generated for all combinations of vehicle-related parameters associated with each property parameter (YES), the coefficients of determination stored in step S325 are read out and the coefficients of determination are compared. , the best engine oil property prediction model for each of the property parameters (step S329).

For example, for one property parameter, the base number, an engine oil property prediction model that is closest to the measured value is determined. That is, from among a plurality of candidate engine oil property prediction models generated for every combination of vehicle-related parameters including the oil usage distance, one engine oil property prediction model that is closest to the measured base value is determined. In other words, one best engine oil property prediction model is selected from a plurality of candidate engine oil property prediction models.

In addition to selecting the best one engine oil property prediction model, several good engine oil property prediction models may be selected. For example, all the engine oil property prediction models with coefficients of determination of 0.8 or more may be selected as good engine oil property prediction models. An engine oil property prediction model can be appropriately selected and predicted in accordance with changes in property parameters and vehicle-related parameters over time and changes over time.

In general, the optimal combination of vehicle-related parameters differs depending on the property parameters.

Next, the coefficients of the best engine oil property prediction model determined in step S329 are stored (step S331), and the process ends.

By executing the generated best engine oil property prediction model, property parameters can be accurately predicted.

<<<Engine oil property prediction function>>>
FIG. 3 is a functional block diagram of engine oil property prediction according to the present embodiment.

The engine oil property prediction process has a data storage unit 100 , a prediction model generation unit 200 and a communication unit 300 .

<<data storage unit 100>>
The data storage unit 100 stores data values of vehicle characteristic parameters and property parameters. In the present embodiment, the vehicle characteristic parameters are oil usage distance, total engine mileage, engine displacement, vehicle size, monthly mileage, vehicle type, maximum oil load, load capacity, and the like. The property parameters are base number, Cu, Fe, residual carbon content, kinematic viscosity, Pb and the like. As previously mentioned, actual measurements of these parameters are obtained and collected while the internal combustion engine is running or not. The collected measured values are stored and accumulated in the data storage unit 100 . Machine learning can be performed using accumulated data.

<<prediction model generation unit 200>>
Prediction model generation unit 200 has parameter selection unit 210 , candidate prediction model generation unit 220 , candidate prediction model evaluation unit 230 , candidate prediction model verification unit 240 , and best property prediction model storage unit 250 .

<Parameter selection unit 210>
The parameter selection unit 210 selects property parameters and vehicle characteristic parameters. Specifically, the property parameter selection is a process of selecting one property parameter from the base number, Cu, Fe, residual carbon content, kinematic viscosity, Pb, and the like. Selecting the vehicle characteristic parameter includes selecting at least one vehicle characteristic parameter from oil usage distance, total engine mileage, engine displacement, vehicle size, monthly mileage, vehicle type, maximum oil load, load capacity, and the like. processing.

Specifically, vehicle characteristic parameters are selected from the combinations shown in FIGS. 6A-6C. In this embodiment, it is selected from all combinations of vehicle-related parameters including oil usage distance (see FIG. 6A).

A combination of vehicle-related parameters and property parameters is selected by the parameter selection unit 210, and at least one vehicle feature parameter can be associated with one property parameter.

The parameter selection unit 210 corresponds to the process of step S313 in FIG.

<Candidate predictive model generating unit 220>
Candidate predictive model generating section 220 generates an engine oil property candidate predictive model using the combination data of the property parameters and vehicle characteristic parameters selected by parameter selecting section 210 .

Candidate prediction model generation unit 220 generates engine oil property candidate prediction models using vehicle characteristic parameters as explanatory variables. In the property prediction of engine oil according to the present embodiment, a random forest is used as a machine learning algorithm. Machine learning by random forest generates an engine oil property candidate prediction model.

Candidate prediction model generation unit 220 reads data for machine learning corresponding to combinations of vehicle-related parameters and property parameters, and uses the read data to generate engine oil property candidate prediction models by the random forest method. Candidate prediction model generating section 220 generates one engine oil property candidate prediction model for one combination of vehicle-related parameters and property parameters.

The candidate predictive model generation unit 220 corresponds to the processing of steps S315 and S317 in FIG.

The candidate predictive model generator 220 has a bootstrap sampling module, a decision tree generation module, and an evaluation module.

The bootstrap sampling module allows duplication from the training data and generates a plurality of sample data that are subsets.

The decision tree generation module generates a decision tree from each piece of sample data, and generates decision trees as many as the number of sample data. By generating a predetermined number of decision trees, a prediction model consisting of a set of decision trees is generated.

The evaluation module estimates the number of failures by inputting data having a track record of the number of failures generated in an actually developed program for the candidate prediction model generated using the learning data. The generated candidate prediction models are evaluated by comparing the calculated number of occurrences of defects with the actual number of occurrences of defects.

Although the random forest is used as the machine learning algorithm in this embodiment, other machine learning algorithms may be used. For example, machine learning algorithms such as decision trees, support vector machines, Gaussian process regression, and ensemble learning with boosting can be used as appropriate. Machine learning can be performed by appropriately changing the machine learning algorithm according to the combination of property parameters and vehicle feature parameters and the collected data. In addition, as the data accumulates, the tendency of the data may change, and the algorithm may be switched appropriately according to the accumulation of data.

Alternatively, a plurality of machine learning algorithms may be used to generate engine oil property candidate prediction models, and the machine learning algorithm that generated the optimum prediction model may be selected. The selection of the machine learning algorithm may be appropriately determined according to the accuracy of prediction, hardware resources such as memory, processing time, and the like.

<Candidate prediction model evaluation unit 230>
Candidate prediction model evaluation unit 230 calculates property parameter values using the engine oil property candidate prediction models generated by candidate prediction model generation unit 220, and determines the calculated values and actual verification data values. Calculate the coefficient.

As described above, in the present embodiment, an engine oil property candidate prediction model is generated for each of all combinations of vehicle-related parameters including the oil usage distance. Candidate prediction model evaluation unit 230 calculates the coefficient of determination of each of the generated engine oil property candidate prediction models. Candidate prediction model evaluation unit 230 stores the coefficient of the generated engine oil property candidate prediction model and the coefficient of determination.

Candidate prediction model evaluation unit 230 corresponds to the processing of steps S319 to S325 in FIG.

<Candidate prediction model verification unit 240>
Candidate prediction model verification unit 240 determines the best prediction model for each property parameter from the coefficient of determination calculated for each engine oil property candidate prediction model.

Candidate prediction model verification unit 240 corresponds to the process of step S329 in FIG.

<Best Property Prediction Model Storage Unit 250>
The best property prediction model storage unit 250 stores the coefficients and constants of the engine oil property prediction model determined to be the best for each property parameter by the candidate prediction model verification unit 240 as the best engine oil property prediction model.

<<communication unit 300>>
The communication unit 300 can transmit and receive with an external device. For example, it is communicably connected to an external device via a network. The external device consists of a computer or the like.

Communication unit 300 transmits the engine oil property prediction model generated by prediction model generation unit 200 to an external device. A program for executing the generated engine oil property prediction model, coefficients and constants necessary for execution, etc. are transmitted. As a result, the engine oil property predictive model generated by the predictive model generator 200 can be executed in an external device.

Also, the engine oil property prediction model generation program itself for executing the processing of the prediction model generation unit 200 is transmitted to an external device. As a result, the engine oil property prediction model generation program can be executed in the external device, and the engine oil property prediction model can be generated in the external device.

<<Characteristics Judgment>>
The property determination unit uses the property prediction model generated by the prediction model generation unit 200 to input actual vehicle characteristic parameter data and determines the property of the engine oil. That is, it determines the degree of deterioration of the engine oil.

The property of the engine oil determined by the property judging section can be displayed on a display or the like. Alternatively, the information may be transmitted to a terminal device or the like by communication means. Furthermore, depending on the properties, information regarding the necessity of engine oil replacement may be determined and displayed.

<<Relationship between combinations of vehicle characteristic parameters and property parameters>>
FIG. 5 is a table showing combinations of vehicle characteristic parameters suitable for prediction of each property parameter obtained for one type of engine oil by executing the engine oil property prediction model according to the present embodiment. FIG. 5 is a table showing one type of engine oil, and in the case of other types of engine oil, combinations different from those in FIG. 5 may be used. That is, when the engine structure or the type of engine oil changes, the type of property parameters may change, or the combination of vehicle feature parameters suitable for prediction of property parameters may change. Since the engine oil property prediction model generation program according to the present embodiment generates an engine oil property prediction model for each engine structure and type of engine oil, it is possible to generate an engine oil property prediction model according to the engine structure and engine oil. can be done. In FIG. 5, the coefficient of determination is shown at the bottom.

The base number, which is a property parameter, is associated with a combination of two vehicle characteristic parameters, oil usage distance and maximum oil load. Cu, which is a property parameter, is associated with a combination of three vehicle characteristic parameters: oil usage distance, total engine mileage, and engine displacement. The property parameter Fe is associated with a combination of five vehicle characteristic parameters: oil usage distance, total engine mileage, engine displacement, vehicle size, and maximum oil load. The residual carbon content, which is a property parameter, is associated with a combination of three vehicle characteristic parameters: oil usage distance, total engine mileage, and engine displacement. The kinematic viscosity, which is a property parameter, is associated with a combination of five vehicle characteristic parameters: oil usage distance, total engine mileage, engine displacement, monthly mileage, and vehicle type. The property parameter Pb is associated with a combination of four vehicle characteristic parameters: oil usage distance, total engine mileage, monthly mileage, and vehicle type.

As shown in FIG. 5, all property parameters can be associated with combinations with oil usage distances. As described above, the oil usage distance indicates the usage state of the engine oil. If the oil usage distance is short, the engine oil has not deteriorated much, and if the oil usage distance is long, the engine oil has deteriorated. ing. Therefore, a property parameter to indicate deterioration of engine oil can be well correlated with oil usage mileage.

Therefore, the property parameter can be associated with the oil usage distance as a basic vehicle characteristic parameter (see, for example, FIG. 4A). In addition, the remaining vehicle characteristic parameters other than the oil usage mile act as correction vehicle characteristic parameters for correcting the combinations that include the oil usage mileage. Combinations that include oil usage distance, which is a basic vehicle characteristic parameter, can be strengthened by corrected vehicle characteristic parameters (see, for example, FIG. 4(b)). In FIGS. 4A and 4B, for the sake of simplicity, the oil usage distance, which is the basic vehicle characteristic parameter, is indicated as α, and one of the corrected vehicle characteristic parameters other than the oil usage distance is indicated as β. Indicated.

<<<Modification 1>>>
In the above examples, base number, Cu, Fe, residual carbon content, kinematic viscosity, and Pb were used as the property parameters, but the parameters are not limited to these. In addition, flash point, water content, Ca, Zn, Cr, Sn, etc. can be used as property parameters that may be predicted. By including these, deterioration of engine oil can be predicted with greater versatility.

In addition to the degree of deterioration of engine oil, the speed of deterioration (for example, the rate of increase in the amount of substances eluted in engine oil) and the acceleration of deterioration (for example, the increase in the amount of substances eluted in engine oil) Acceleration of ) etc. may be used to determine the properties. It is possible to easily judge the deterioration of the engine oil in a state close to a new car and the deterioration of the engine oil in a state of advanced use.

<<<Modification 2>>>
<< Detection of engine oil property by sensor >>>
A vehicle may be provided with a sensor for detecting properties of the engine oil. By using the sensor, the properties of the engine oil can be detected while the internal combustion engine is running. To easily detect the properties of engine oil without extracting the engine oil when the engine is not in operation.

Various types of information obtained by detecting the engine oil by the sensor are stored in the storage device. The storage device includes a nonvolatile RAM (random access memory), HDD (hard disk drive), SSD (solid state drive), and the like. The storage device may be a recording medium or storage medium that can retain stored content even when the power is turned off.

<<<<Scope of Embodiment>>>>
While the embodiments have been described, as noted above, the description and drawings forming part of this disclosure should not be taken as limiting. Various embodiments and the like not described here are included.

100 data storage unit 200 prediction model generation unit 300 communication unit

Claims (6)

data for at least one of first vehicle parameters including one or more of vehicle type, engine displacement, vehicle size, maximum oil load, total vehicle mileage, and mileage over a predetermined period; and engine oil usage mileage. Data for at least one of the second vehicle parameters and data for property parameters indicating properties of the engine oil are machine-learned as explanatory variables to determine the properties of the engine oil. a machine learning step of generating at least one predictive model for predicting
Verification of data of at least one of the first vehicle parameters, data of at least one of the second vehicle parameters, and data of property parameters indicating properties of engine oil A good prediction model determination step of verifying the at least one prediction model using verification data used in and determining a prediction model that satisfies good requirements as a good prediction model. An engine oil property prediction model generation program. The engine oil property prediction model according to claim 1, wherein the machine learning process is at least one of random forest, linear regression, decision tree, support vector machine, Gaussian process regression, and ensemble learning by boosting. generation program. The machine learning step includes:
A bootstrap sampling module that randomly extracts from the learning data while allowing duplication;
a decision tree generation module that generates a plurality of decision trees using the explanatory variables;
an evaluation module that evaluates a prediction model composed of a set of decision trees;
The engine oil property prediction model generation program according to claim 2, comprising: The property parameter includes at least one unit characteristic parameter indicating a unit characteristic of engine oil,
At least one parameter selected from the first vehicle parameters and at least one parameter selected from the second vehicle parameters are associated with one unit characteristic parameter, and one unit characteristic parameter is associated with 4. The engine oil property prediction model generation program according to any one of claims 1 to 3, wherein at least one corresponding prediction model is generated. A server capable of providing the engine oil property prediction model generation program according to claim 1 . A server capable of providing an engine oil property prediction model generated by the engine oil property prediction model generation program according to claim 1.

Images