JP2021135231A - Wear-out state prediction method, wear-out state prediction device, wear-out state prediction program, and prediction model generation method - Google Patents

Abstract

To provide a wear-out state prediction method, wear-out state prediction device, wear-out state prediction program and prediction model generation method that use a wear-out speed of an aircraft tire as supervised data to be input when generating a model.SOLUTION: A wear-out state prediction device 10 is provided with a controller 11 that generates a model predicting a wear-out speed of an aircraft tire 32 as an objective variable, using a prescribed algorithm with preliminarily acquired temperature data on an aircraft 31 as an explanatory variable. The controller 11 is configured to use the wear-out speed as supervised data associated with the explanatory variable and input when generating the model. The wear-out speed is a parameter obtained by dividing a wear volume of the aircraft tire 32 by an accumulation value about flight.SELECTED DRAWING: Figure 1

Description

The present invention relates to a wear state prediction method, a wear state prediction device, a wear state prediction program, and a prediction model generation method.

Conventionally, a technique for predicting a wear state of an aircraft tire has been known (Patent Document 1). The method described in Patent Document 1 applies a plurality of wear energies corresponding to a plurality of running states (for example, a touchdown running state, a deceleration running state after touchdown, a taxi running state, etc.) classified according to usage conditions. Obtain and predict the wear state of aircraft tires based on the acquired wear energy.

Japanese Unexamined Patent Publication No. 2013-113724

In recent years, big data analysis has become active due to the development of AI (for example, machine learning). However, there is little research on prediction methods that utilize big data regarding the wear state of aircraft tires.

Therefore, the present invention has been made in view of such a situation, and is a wear state prediction method, a wear state prediction device, and wear that use the wear rate of an aircraft tire as training data input when generating a model. An object of the present invention is to provide a state prediction program and a prediction model generation method.

The wear state prediction method according to one aspect of the present invention uses the wear rate of an aircraft tire as the training data that is associated with the explanatory variables and input when the model is generated.

According to the present invention, it is possible to generate a model having high prediction performance.

FIG. 1 is a schematic view showing the relationship between the wear state predictor, the network, and the airline company. FIG. 2 is a schematic configuration diagram of a wear state prediction device according to the present embodiment of the present invention. FIG. 3 is a diagram illustrating an example of machine learning according to the present embodiment of the present invention. FIG. 4 is a diagram illustrating an example of associating the learning data and the teacher data according to the embodiment of the present invention. FIG. 5 is a sequence diagram illustrating an operation example of the wear state prediction system according to the embodiment of the present invention. FIG. 6 is a sequence diagram illustrating an operation example of the wear state prediction system according to the embodiment of the present invention. FIG. 7 is a diagram illustrating an example of a method in which the air temperature after the day when the wear rate is predicted is taken into consideration.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the description of the drawings, the same parts are designated by the same reference numerals and the description thereof will be omitted.

(Configuration example of wear condition prediction system)
An example of the configuration of the wear state prediction system 1 will be described with reference to FIG. As shown in FIG. 1, the wear state prediction system 1 includes a wear state prediction device 10, a network 20, and an airline company 30.

The wear state prediction device 10 performs two-way communication with the airline company 30 via the network 20. Specifically, the wear state prediction device 10 acquires data on the aircraft 31 from the airline company 30 via the network 20. The wear state prediction device 10 generates a machine learning model using the acquired data, and predicts the wear state of the aircraft tire 32 mounted on the aircraft 31 using the generated machine learning model. Details of the data acquired by the wear state predictor 10 from the airline company 30 will be described later. The network 20 is a communication network capable of transmitting and receiving data. For example, the network 20 is composed of various communication lines such as a dedicated line, a public switched telephone network, a satellite communication line, and a mobile communication line installed by a telecommunications carrier.

In the present embodiment, the aircraft tire 32 includes a tire mounted on the main gear and a tire mounted on the nose gear. A plurality of (for example, 6) tires are mounted on the main gear. Also, a plurality of (for example, two) tires are mounted on the nose gear.

The wear state prediction device 10 is, for example, a general-purpose computer, and includes a CPU, a ROM (Read Only Memory), and a RAM (Random Access Memory). The CPU reads the program stored in the ROM or the like into the RAM and executes it. The wear state prediction device 10 may be a stationary terminal device or a portable terminal device (for example, a smartphone) that is easy to carry. Further, the wear state prediction device 10 may be configured as a server of the management center.

Next, a configuration example of the wear state prediction device 10 will be described with reference to FIG. As shown in FIG. 2, the wear state prediction device 10 includes a controller 11 (for example, a CPU) and a storage device 14. The controller 11 includes a first processing function 12 and a second processing function 13 as a plurality of information processing functions. Further, the first processing function 12 is classified into a data acquisition unit 121, a data processing unit 122, an algorithm selection unit 123, and a model generation unit 124. The second processing function 13 is classified into a new data acquisition unit 131, an application model selection unit 132, and a wear state prediction unit 133.

The data acquisition unit 121 acquires data from the airline company 30 via a communication unit (not shown). The communication unit is an interface installed in the wear state prediction device 10 that connects to the network 20 and transmits / receives data to / from the airline company 30. The data acquired by the data acquisition unit 121 from the airline company 30 includes temperature data of the aircraft 31, acceleration data of the aircraft 31, weight data of the aircraft 31, and data indicating the position where the aircraft tire 32 is mounted (hereinafter, simply mounting position). (Sometimes called data), etc. are included.

The temperature data of the aircraft 31 is the temperature data measured by the sensor installed in the aircraft 31.

The acceleration data of the aircraft 31 includes the acceleration data of the aircraft 31 in the XYZ directions. The X direction is the front-rear direction of the aircraft 31. The Y direction is the left-right direction of the aircraft 31. The Z direction is the vertical direction of the aircraft 31.

The data processing unit 122 processes the data acquired by the data acquisition unit 121. The detailed contents of the process will be described later.

The algorithm selection unit 123 selects an algorithm to be used to generate a machine learning model. Algorithms include, for example, linear multiple regression, lasso regression, non-linear SVM, random forest, XGBost and the like.

The model generation unit 124 generates a machine learning model by machine learning the data generated by the data processing unit 122 using the algorithm selected by the algorithm selection unit 123. The model generation unit 124 generates a machine learning model for each algorithm. Therefore, a plurality of machine learning models are generated. The generated machine learning model is stored in the storage device 14.

The new data acquisition unit 131 acquires new data to be input to the machine learning model generated by the model generation unit 124 from the airline company 30. The new data acquired by the new data acquisition unit 131 from the airline company 30 has the same attributes (temperature, acceleration, etc.) and structure (every second described later) as the data acquired by the data acquisition unit 121 from the airline company 30. However, the data itself is different.

The applied model selection unit 132 reads the machine learning model generated by the model generation unit 124 from the storage device 14, and evaluates the prediction performance of the read machine learning model. The applied model selection unit 132 selects the machine learning model having the highest prediction performance based on the evaluation result.

The wear state prediction unit 133 inputs new data acquired by the new data acquisition unit 131 into the machine learning model selected by the application model selection unit 132, and predicts the wear state. The wear state includes the amount of wear and the rate of wear. Details of the amount of wear and the rate of wear will be described later. The data regarding the predicted wear state is stored in the storage device 14.

Next, a method of generating a machine learning model and a method of predicting a wear state using the generated machine learning model will be described with reference to FIGS. 3 to 4.

As shown in FIG. 3, machine learning is classified into a learning phase and a prediction phase. In the learning phase, features are extracted from a huge amount of data (big data) and a machine learning model is generated. In the prediction phase, new data is input to the generated machine learning model and the results are output. In FIG. 3, as described above, the acquired data is data acquired from the airline company 30, such as temperature data of the aircraft 31, acceleration data of the aircraft 31, weight data of the aircraft 31, and mounting position data.

In the field of statistics, the data input for machine learning is sometimes called an explanatory variable. In the present embodiment, the temperature data of the aircraft 31, the acceleration data of the aircraft 31, the weight data of the aircraft 31, the mounting position data, and the like are explanatory variables. In addition, the data output from the trained model may be called an objective variable. In the present embodiment, the wear state of the aircraft tire 32 is the objective variable.

Machine learning includes supervised learning using supervised learning and learning without supervised data. In this embodiment, supervised learning is used. Teacher data means so-called "correct answer data". Supervised learning is a method using a learning data set (indicated by reference numeral 40 in FIG. 3) in which "input data" and "correct answer data" are set. As shown in FIG. 3, in the present embodiment, big data is processed to generate a learning data set 40 before machine learning is performed. Such data processing before machine learning is referred to as data preprocessing below.

Next, data preprocessing will be described. In the present embodiment, the data preprocessing means the aggregation of data and the association between the aggregated data and the teacher data. Such association is sometimes called labeling. The data acquisition unit 121 acquires data from the airline company 30 for each flight. When data is acquired from a sensor mounted on an aircraft as a data structure for one flight, the data is stored, for example, every second. That is, temperature data, acceleration data, and the like are stored every second. Therefore, the amount of data is enormous even for one flight.

Usually, the number of flights from the time when the aircraft tire 32 is attached to the airframe to the time when the tire 32 is taken out is several hundred times. In the following, the number of flights from the time when the aircraft tire 32 is attached to the airframe to the time when it is taken out may be expressed as all flights. In the present embodiment, the data to be machine-learned is data of all flights (hundreds of flights) from the time when the aircraft tire 32 is attached to the body to the time when it is taken out. For such big data, the teacher data (wear rate based on the amount of wear) to be linked is one value. The amount of wear means the amount of wear of the aircraft tire 32 from the time when the aircraft tire 32 is attached to the airframe to the time when it is taken out, and the unit is mm. When the groove depth of the aircraft tire 32 when new is X (mm) and the groove depth when the aircraft tire 32 is taken out is Y (mm), XY (mm) is the amount of wear. Is. Usually, the amount of wear is measured after the aircraft tire 32 is taken out, so there is only one value.

In this embodiment, the wear rate is used as the teacher data. Here, the relationship between the amount of wear and the wear rate will be described. In the present embodiment, the wear rate is defined as the amount of wear of the aircraft tire 32 in one landing. The unit of wear rate is expressed as mm / LD. The wear rate is calculated by dividing the amount of wear measured after the aircraft tire 32 is taken out by the total number of landings. The wear rate calculated in this way is used as the teacher data. In this embodiment, the number of landings and the number of flights are synonymous.

However, the definition of wear rate is not limited to the above. The wear rate may be defined as the amount of wear relative to the taxi mileage. Taxi traveling means that the aircraft 31 travels on the ground (mainly a runway) using the power of the aircraft 31. The wear rate is calculated by dividing the amount of wear measured after the aircraft tire 32 is taken out by the taxi mileage on all flights. The taxi mileage in one flight is the total value of the taxi mileage at the departing airfield and the taxi mileage at the arriving airfield. This total value is usually several kilometers.

The wear rate may also be defined as a parameter obtained by dividing the amount of wear by the cumulative value for flight. Cumulative values for flights include, for example, the number of landings and taxi mileage described above. Further, the cumulative value related to the flight may include the total value of the representative values (for example, the average value) of the acceleration of each flight and the total value of the weight of each flight.

In the following, as an example, the wear rate will be described as the amount (mm / LD) that the aircraft tire 32 wears in one landing. Of course, as described above, the wear rate may be a value obtained by dividing the amount of wear by the taxi mileage, or may be a parameter obtained by dividing the amount of wear by the cumulative value related to flight. Further, as described above, since the amount of wear is one value, the wear rate calculated by the amount of wear is also one value.

Here, the reason why the wear rate is used as the teacher data in this embodiment will be described. The inventor's research has revealed that when wear is used as teacher data, all the data may be in a horizontal line, making accurate prediction difficult. In addition, the inventor's research has revealed that when the number of landings is used as teacher data, it may be difficult to make an accurate prediction because the variation in the amount of residual groove is not reflected. On the other hand, the inventor's research has shown that accurate predictions are possible when wear rates are used as teacher data. Therefore, in this embodiment, the wear rate is used as the teacher data.

Since the teacher data (wear rate) is one value, it is necessary to aggregate each of the various data associated with the teacher data into one value. Therefore, the data processing unit 122 aggregates the temperature data, the acceleration data, the weight data, and the like into one value. Since the mounting position data is a fixed value, it does not need to be aggregated.

First, an example of a method for aggregating temperature data will be described. The data processing unit 122 averages the temperature data in one flight and calculates the average temperature data in one flight. The data processing unit 122 repeats this process to calculate the average temperature data for each flight (first flight, second flight ... 100th flight ...). Then, the data processing unit 122 averages the average temperature data for each flight and calculates the average temperature data (one value) for all flights (first to several hundredth times). It should be noted that such an average process is an example of an aggregation method and is not an essential process. As will be described later, the representative value of the explanatory variable is not limited to the average value, and may be the median value or the mode value.

As mentioned above, when acquiring data from a sensor mounted on an aircraft, temperature data is stored every second. That is, the stored air temperature data includes the air temperature data when the aircraft 31 is stopped, the air temperature data when the aircraft 31 is running, the air temperature data when the aircraft 31 is flying, and the like. Is done. When aggregating the temperature data, the data processing unit 122 may select and use a part of the data according to the state of the aircraft 31. For example, the temperature data when the aircraft 31 is in flight need not be used. In other words, the data processing unit 122 may calculate the average temperature data (one value) by averaging the temperature data of a part of each flight.

Next, an example of an acceleration data aggregation method will be described. As described above, the acceleration data includes the respective acceleration data in the XYZ directions. Therefore, the data processing unit 122 aggregates each of the acceleration data in the X direction, the acceleration data in the Y direction, and the acceleration data in the Z direction. Here, as a representative, acceleration data in the X direction will be described.

The data processing unit 122 calculates the average acceleration data (X direction) in one flight by squaring the acceleration data (X direction) in one flight and then averaging them. The data processing unit 122 repeats this process to calculate the average acceleration data (X direction) for each flight (first flight, second flight ... 100th flight ...). Then, the data processing unit 122 averages the average acceleration data (X direction) for all flights and calculates the average acceleration data (one value in the X direction) for all flights (first to several hundred times). do. Since the same applies to the Y direction and the Z direction, description thereof will be omitted.

Next, an example of a weight data aggregation method will be described. The data processing unit 122 extracts the maximum value from the weight data in one flight. The data processing unit 122 repeats this process to extract the maximum value of the weight data for each flight (first flight, second flight ... 100th flight ...). Then, the data processing unit 122 averages the maximum values of the weight data for each flight and calculates the average weight data (one value) for all the flights (first to several hundredth times).

In this way, the data processing unit 122 calculates the average temperature data, average acceleration data (X direction, Y direction, Z direction), and average weight data of all flights as one value. As shown in FIG. 4, one of these values (learning data 41) is associated with the teacher data 42 (wear rate) to generate a learning data set 40. Since the training data set 40 is generated for each aircraft tire, a plurality of training data sets 40 are usually generated.

The algorithm selection unit 123 selects an algorithm to be used to generate a machine learning model. Algorithms include linear multiple regression, lasso regression, non-linear SVM, random forest, XGBost, etc., as described above. However, the present invention is not limited to these, and any algorithm that enables supervised learning may be used. Since these algorithms are well known, detailed description thereof will be omitted.

As shown in FIG. 4, the model generation unit 124 generates a machine learning model by machine learning the learning data set 40 generated by the data processing unit 122 by using the algorithm selected by the algorithm selection unit 123. In this embodiment, the above-mentioned linear multiple regression, lasso regression, non-linear SVM, random forest, and XGBost are used as the algorithm.

In general, the prediction performance by a machine learning model is evaluated by one or a combination of items indicating the performance of the algorithm such as the ^{coefficient of determination (R 2) and the root mean square error (RMSE).} In the present embodiment, the closer the value of the wear rate, which is the objective variable, to the actually measured value, the higher the performance is evaluated. The applied model selection unit 132 selects the machine learning model having the highest prediction performance based on the evaluation result.

As shown in FIG. 3, the wear state prediction unit 133 inputs new data into the machine learning model (learned model) selected by the application model selection unit 132, and predicts the wear rate. The new data is acquired by the new data acquisition unit 131. The new data is data from the time when the aircraft tire 32 is attached to the airframe to the time when the wear rate is predicted.

In this embodiment, average temperature data, average acceleration data, average weight data, mounting position data, and the like are used as explanatory variables for machine learning, but in these explanatory variables, the contribution rate to the objective variable is not uniform. .. The contribution rate indicates the magnitude of the influence of each explanatory variable on the objective variable, and the greater the influence on the objective variable, the more important the explanatory variable can be said to be.

The explanatory variables (average temperature data, average acceleration data, average weight data, mounting position data, etc.) in the present embodiment are non-cumulative explanatory variables. In other words, in this embodiment, the cumulative explanatory variables are not used, and only the non-cumulative explanatory variables are used. The cumulative explanatory variable is, for example, the taxi mileage described above. The reason for not using the cumulative explanatory variables is that the accuracy of the machine learning model is reduced when the cumulative explanatory variables are used.

The inventor analyzed the contribution of each of the explanatory variables to the objective variable. As a result, the inventor found that among the explanatory variables, the contribution rate of the average temperature data was the highest compared to other data. The inventor also found that the explanatory variable with the second highest contribution rate after the average temperature data is the average acceleration data. This indicates that a highly accurate model can be obtained by using the average temperature data as an explanatory variable when generating a machine learning model for predicting the wear rate. Moreover, it is shown that a more accurate model can be obtained by adopting the average acceleration data as an explanatory variable in addition to the average temperature data.

Next, an operation example of the wear state prediction system 1 will be described with reference to the sequence diagrams of FIGS. 5 to 6.

In step S101, the temperature data of the aircraft 31, the acceleration data of the aircraft 31, the weight data of the aircraft 31, the mounting position data, and the like are transmitted from the airline company 30.

In step S103, the data acquisition unit 121 acquires the data transmitted in step S101. The process proceeds to step S105, and the data processing unit 122 processes the data acquired in step S103. Specifically, the data processing unit 122 aggregates temperature data, acceleration data, weight data, and the like into one value. The data processing unit 122 creates the learning data set 40 by associating the aggregated data (learning data 41) with the teacher data 42.

The process proceeds to step S107, and the algorithm selection unit 123 selects an algorithm (linear multiple regression, Lasso regression, non-linear SVM, random forest, XGBost, etc.) to be used for generating the machine learning model.

The process proceeds to step S109, and the model generation unit 124 generates a machine learning model by machine learning the learning data set 40 generated in step S105 using the algorithm selected in step S107.

In step S201, the temperature data of the aircraft 31, the acceleration data of the aircraft 31, the weight data of the aircraft 31, the mounting position data, and the like are transmitted from the airline company 30. The data transmitted in step S201 is new data used to predict the wear rate.

In step S203, the new data acquisition unit 131 acquires the new data transmitted in step S201. The process proceeds to step S205, and the application model selection unit 132 evaluates the prediction performance of the machine learning model generated in step S109 and selects the machine learning model having the highest prediction performance.

The process proceeds to step S207, and the wear state prediction unit 133 inputs the new data acquired in step S203 into the machine learning model selected in step S207, and predicts the wear rate.

(Action effect)
As described above, according to the wear state prediction method according to the present embodiment, the following effects can be obtained.

The wear state prediction device 10 uses the wear rate as the teacher data that is associated with the explanatory variables and is input when the model (machine learning model) is generated. The wear state prediction device 10 inputs the data of the aircraft 31 into the generated model and predicts the wear rate. As mentioned above, the inventor's research has shown that accurate predictions are possible when wear rates are used as teacher data. Therefore, by using the wear rate as the teacher data, it is possible to generate a machine learning model with high prediction performance. Then, by using a machine learning model having high prediction performance, it becomes possible to accurately predict the wear rate. The wear rate is a value obtained by dividing the amount of wear of the aircraft tire 32 by the number of landings, or a value obtained by dividing the amount of wear of the aircraft tire 32 by the taxi mileage.

Although embodiments of the invention have been described above, the statements and drawings that form part of this disclosure should not be understood to limit the invention. Various alternative embodiments, examples and operational techniques will be apparent to those skilled in the art from this disclosure.

In the above-described embodiment, the new data (data at the time of prediction) input to the machine learning model is the data from the time when the aircraft tire 32 is mounted on the airframe to the time when the wear rate is predicted. I explained. On the other hand, it was explained that the data for generating the machine learning model (data at the time of learning) is the data from the time when the aircraft tire 32 is attached to the airframe to the time when it is taken out. The date on which the wear rate is predicted and the date on which the aircraft tire 32 is taken out do not always match. Therefore, the number of data from the day when the aircraft tire 32 is mounted on the airframe to the day when the wear rate is predicted, and the data from the day when the aircraft tire 32 is mounted on the airframe to the day when it is taken out. The numbers do not always match. As a result, highly accurate results may not be obtained.

Therefore, with respect to the average temperature data having the highest contribution rate to the objective variable, the temperature after the day when the wear rate is predicted may be considered. An example of a method in which the air temperature after the day when the wear rate is predicted is taken into consideration will be described with reference to FIG. 7.

January 8th shown in FIG. 7 is the day when the latest data exists, and January 11th is the day when the wear rate is predicted (hereinafter, may be simply referred to as the prediction implementation date). The latest data is the latest data transmitted by airline 30.

There is no data for January 9-11 when the wear rate is predicted, i.e. January 11. In this case, the data acquisition unit 121 first aggregates the data from the day when the aircraft tire 32 is mounted on the airframe to January 8. Among the aggregated data, the average temperature data (first temperature data) is 12.2 degrees as shown in FIG. The wear state prediction unit 133 inputs the aggregated data into the machine learning model and predicts the wear rate. The predicted wear rate is 0.0130 as shown in FIG.

The wear state prediction unit 133 calculates the amount of wear by multiplying the predicted wear rate (0.0130) by the number of flights (625 times). The wear state prediction unit 133 subtracts the calculated wear amount from the groove depth of the aircraft tire 32 when it is new, and calculates the remaining groove amount indicating the remaining groove depth. The calculated residual groove amount is 2.38 mm as shown in FIG. The wear state prediction unit 133 may calculate the amount of wear by multiplying the predicted wear rate by the cumulative value related to the flight.

When the amount of remaining grooves is larger than 0, the wear state prediction unit 133 generates data for the next day (January 9). As shown in FIG. 7, since the remaining groove amount (2.38 mm) on January 8 is larger than 0, the wear state prediction unit 133 generates the data for the next day (January 9). The wear state prediction unit 133 acquires the January average air temperature (second air temperature data) of the AAA airfield where the aircraft 31 arrives and stores it in the database. As a result, 8.9 degrees (see FIG. 7) is input to the average temperature of the AAA airfield on January 9. Next, the wear state prediction unit 133 uses the temperature data of 8.9 degrees to update the average temperature data from the day when the aircraft tire 32 is attached to the airframe to January 9. The updated average temperature data is 12.2 degrees as shown in FIG.

The wear state prediction unit 133 inputs data including the updated average temperature data into the machine learning model, and predicts the wear rate again. The predicted wear rate is 0.0130 as shown in FIG. The wear state prediction unit 133 calculates the amount of wear by multiplying the predicted wear rate (0.0130) by the number of flights (632 times). The wear state prediction unit 133 subtracts the calculated wear amount from the groove depth of the aircraft tire 32 when it is new, and calculates the remaining groove amount. The calculated residual groove amount is 2.28 mm as shown in FIG.

Since the remaining groove amount (2.28 mm) on January 9 is larger than 0, the wear state prediction unit 133 generates data for the next day (January 10). Then, the same process is repeated until the amount of the remaining groove becomes 0 or less. The average temperature of AAA airfield in February is 10.2 degrees Celsius.

As shown in FIG. 7, the day when the remaining groove amount becomes 0 or less is February 13 (the remaining groove amount is −0.02). After the day when the remaining groove amount becomes 0 or less (February 13) is calculated, the wear state prediction unit 133 calculates the day when the aircraft tire 32 is taken out. Next, an example of a method of calculating the date on which the aircraft tire 32 is taken out will be described.

First, the wear state prediction unit 133 uses the groove depth of the aircraft tire 32 when new and the wear speed (0.0120) on February 12, and the number of flights until just before the remaining groove amount becomes 0. (* 2 shown in FIG. 7) is calculated. The wear state prediction unit 133 has calculated the cumulative number of flights from the time when the aircraft tire 32 was installed on the prediction implementation date (January 11) from the number of flights until the remaining groove amount becomes 0 (shown in FIG. 7). * 1) is subtracted (875-662) to calculate the remaining number of possible landings (remaining LD number in FIG. 7: 213). The remaining number of landings means the number of times the aircraft tire 32 can be used.

The wear state prediction unit 133 divides the remaining number of possible landings by the number of flights per day (7 times) to calculate the number of days that the aircraft tire 32 can be used (the number of remaining days in FIG. 7: 30 days). do. Then, the wear state prediction unit 133 calculates February 10, which is 30 days after January 11 as, as the day when the aircraft tire 32 is taken out.

In this way, it is possible to accurately predict the wear rate by considering the air temperature after the day when the wear rate is predicted (after the time of prediction).

Further, by using a highly accurate wear rate, it is possible to accurately calculate the date when the aircraft tire 32 is taken out. Normally, when replacing an aircraft tire 32, the amount of wear is manually measured by an operator, and the tire 32 is replaced when the amount of wear reaches a predetermined value. However, if a certain number or more of aircraft tires 32 are replaced at the same time, the replacement work may be delayed or the inventory may be insufficient. On the other hand, according to the wear state prediction device 10, the date on which the aircraft tire 32 is taken out is calculated accurately, so that the replacement work can be efficiently operated.

Further, in the above-described embodiment, it has been explained that the temperature data of the aircraft 31 is used as an explanatory variable for generating the machine learning model. Then, it was explained that the temperature data of the aircraft 31 is the temperature data measured by the sensor installed in the aircraft 31. However, the explanatory variables used to generate the machine learning model are not limited to the temperature data measured by the sensors installed in the aircraft 31. The explanatory variables used to generate the machine learning model may be airfield temperature data.

The method of acquiring the air temperature data of the airfield is not particularly limited, but the air temperature data observed at a fixed point may be acquired, or the air temperature data published by an administrative agency (for example, the Japan Meteorological Agency in Japan) may be acquired. The method of aggregating the air temperature data of the airfield is the same as the method of aggregating the air temperature data of the aircraft 31 described above. That is, the air temperature data of the airfield is averaged and aggregated into one value in the same manner as the air temperature data of the aircraft 31.

However, the method of aggregating the air temperature data of the aircraft 31 and the air temperature data of the airfield is not limited to the average. The air temperature data of the aircraft 31 and the air temperature data of the airfield used as explanatory variables may be the median value or the mode value.

The air temperature data for the aircraft 31 includes both the temperature data measured by the sensors installed in the aircraft 31 and the air temperature data of the airfield.

Further, when the data processing unit 122 aggregates the air temperature data of the aircraft 31 and the air temperature data of the airfield, the period in which the air temperature data is stored may be divided and aggregated. For example, the data processing unit 122 may divide the period in which the temperature data is stored into the first half and the second half and aggregate them. As described above, the number of flights from the time when the aircraft tire 32 is attached to the airframe to the time when it is taken out is several hundred times. Here, it is assumed that the number of flights from the time when the aircraft tire 32 is attached to the airframe to the time when the tire 32 is taken out is 500 times.

In this case, the data processing unit 122 may divide and aggregate the period in which the temperature data is stored into the first half (1st to 250th times) and the second half (251st to 500th times). As a result, the average temperature data (one value) in the first half and the average temperature data (one value) in the second half can be obtained as the aggregated data. The data associated with the teacher data may be the average temperature data in the first half or the average temperature data in the second half. Even if the period in which the temperature data is stored is divided and aggregated in this way, the same effect as the above-mentioned effect can be obtained. The data to be divided is not limited to the temperature data, and the period in which the acceleration data is stored may be divided. Further, the division method is not limited to the first half and the second half, and may be divided into three.

Further, when the data processing unit 122 aggregates the air temperature data of the aircraft 31 and the air temperature data of the airfield, the data processing unit 122 classifies the period in which the air temperature data is stored into two or more different periods, and aggregates them using the classified plurality of periods. You may. As an example of the method of classifying into two or more periods, the data processing unit 122 may classify the period in which the temperature data is stored into the entire period (1st to 500th times) and the first half (1st to 250th times). It is possible. The data processing unit 122 averages the average temperature data of each of the plurality of classified periods (all periods and the first half), and acquires the average temperature data (one value) of the plurality of classified periods (the whole period and the first half). do. Even if the period in which the temperature data is stored is classified into a plurality of periods and aggregated in this way, the same effect as the above-mentioned effect can be obtained. The data to be classified is not limited to the temperature data, and the period in which the acceleration data is stored may be classified into a plurality of periods. That is, the data processing unit 122 may classify the periods in which the explanatory variables of the same attribute are stored into two or more different periods, and aggregate them using the plurality of classified periods.

1 Wear condition prediction system 10 Wear condition prediction device 11 Controller 12 1st processing function 13 2nd processing function 14 Storage device 20 Network 30 Aircraft 31 Aircraft 32 Aircraft tire 121 Data acquisition unit 122 Data processing unit 123 Algorithm selection unit 124 Model Generation unit 131 New data acquisition unit 132 Applicable model selection unit 133 Wear condition prediction unit

Claims (9)

It is a wear state prediction method that generates a model that predicts the wear rate of aircraft tires as an objective variable by using a predetermined algorithm that uses aircraft data acquired in advance as an explanatory variable.
The wear rate is used as the teacher data that is associated with the explanatory variables and input when the model is generated.
A wear state prediction method, wherein the wear rate is a parameter obtained by dividing the amount of wear of the aircraft tire by a cumulative value related to flight. The wear state prediction method according to claim 1, wherein the explanatory variable includes temperature data relating to the aircraft. The wear state prediction method according to claim 2, wherein the explanatory variable further includes acceleration data of the aircraft. The wear state prediction method according to any one of claims 1 to 3, wherein the wear rate is predicted by multiplying the predicted wear rate by a cumulative value related to the flight. Any one of claims 1 to 4, wherein the number of times the aircraft tire can be used is predicted by using the groove depth of the aircraft tire acquired in advance when the aircraft tire is new and the predicted wear rate. The wear condition prediction method described in the section. The temperature data for the aircraft is defined as the first temperature data.
Claims 2 and 2, wherein when predicting the wear rate, data including the second air temperature data after the time of prediction and the first air temperature data are input to the model, and the wear rate is predicted. The wear state prediction method according to any one of 5. A wear state prediction device including a controller that generates a model for predicting the wear rate of an aircraft tire as an objective variable by using a predetermined algorithm using aircraft data acquired in advance as an explanatory variable.
The controller uses the wear rate as the teacher data that is associated with the explanatory variables and is input when the model is generated.
The wear rate is a wear state prediction device, which is a parameter obtained by dividing the amount of wear of the aircraft tire by a cumulative value related to flight. A wear state prediction program that generates a model that predicts the wear rate of aircraft tires as an objective variable using a predetermined algorithm that uses aircraft data acquired in advance as an explanatory variable.
On the computer of the terminal device,
The step using the wear rate is executed as the teacher data that is associated with the explanatory variables and input when the model is generated.
The wear rate is a wear state prediction program characterized in that the wear rate is a parameter obtained by dividing the amount of wear of the aircraft tire by a cumulative value related to flight. It is a prediction model generation method that generates a model that predicts the wear rate of aircraft tires as an objective variable by using a predetermined algorithm that uses aircraft data acquired in advance as an explanatory variable.
The model is generated using the wear rate as the teacher data associated with the explanatory variables.
A method for generating a prediction model, wherein the wear rate is a parameter obtained by dividing the amount of wear of the aircraft tire by a cumulative value related to flight.

Images