JP2019121088A - State determination device, currency processing machine state determination system, state determination method, and program - Google Patents

Abstract

To make it possible to update state determination processing for a currency processing machine using data at the operation time of the currency processing machine.SOLUTION: A state determination device includes an acquisition unit configured to acquire sensor data relating to transportation of currency in a currency processing machine, a storage unit configured to store a model that outputs an output value indicating an abnormality degree of the currency processing machine with sensor data relating to the transportation of currency in the currency processing machine as an input value, and a state determination unit configured to determine the state of the currency processing machine using the sensor data acquired by the acquisition unit and the model stored in the storage unit.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a state determination device, a money handling machine state determination system, a state determination method, and a program.

As an error countermeasure in the money handling machine, there is a method of providing a mechanism for removing the cause of the error.
For example, Patent Document 1 discloses a coin handling machine provided with a jammed coin removal mechanism at the entrance of a coin passage. In this coin processing machine, the rotary disk and the coin passage are arranged such that the inlet of the coin passage is located at the peripheral portion of the rotary disk. The coins on the rotary disk are sequentially sent out to the coin passage by the centrifugal force of the rotation of the rotary disk.

Here, a thickness control member is provided at the entrance of the money passage for preventing entry into the coin passage in a state where a plurality of coins overlap. For example, when a plurality of coins overlap in a wedge shape and enter once between the thickness regulating member and the rotary disk, jam may occur at the entrance of the money passage.
Therefore, in this money handling machine, a jammed coin removing mechanism is provided at the entrance of the money passage, and when it is detected that a coin is jammed, the coin is removed. Thus, in this money handling machine, when a coin is stuck at the entrance of the money passage, an error can be generated and automatically eliminated without having to manually eliminate the stuck coin.

Unexamined-Japanese-Patent No. 6-44433

  As another measure against errors in the money handling machine, there is a method of performing maintenance of the money handling machine such as cleaning inside the money handling machine and replacement of parts. While maintenance of equipment such as a money handling machine is generally performed on a regular basis, the frequency of occurrence of errors due to contamination in the equipment or consumption of parts depends on the use situation of the equipment. Once an error occurs, the device may not be used until the error is eliminated. Therefore, it is preferable that maintenance be performed before the error occurs to eliminate the cause of the error.

  In order to perform maintenance before an error occurs, it is conceivable to determine the state of the money handling machine. If the state of the money handling machine is known, the user or maintenance personnel can determine the necessity of maintenance based on the state. However, it is difficult to analyze the vast amount of sensor data obtained from the money handling machine to determine the state of the money handling machine.

  The present invention provides a state determination device capable of accurately determining the state of a money handling machine, a money handling machine state determination system, a state determination method, and a program.

  According to the first aspect of the present invention, the state determination device is an acquisition unit for acquiring sensor data regarding the transportation of money in the money handling machine, and the money using sensor data regarding the transportation of money in the money handling machine as an input value The state of the money processing machine is determined using a storage unit that stores a model that outputs an output value indicating the degree of abnormality of the processing machine, sensor data acquired by the acquisition unit, and a model stored in the storage unit. And a state determination unit.

  The storage unit may store a model for each denomination of money, and the state determination unit may determine the presence or absence of an abnormality for each denomination of money.

  The state determination unit may determine the presence or absence of an abnormality of the money handling machine for all the money of a plurality of denominations based on the output value from the model for each denomination of money.

  When an error occurs in the money handling machine, among the sensor data obtained by measurement by the sensor of the money handling machine where the error occurred, the learning data based on the sensor data in the past first period after the error occurrence is abnormal. Sensor data obtained by measurement by the sensor by performing machine learning using learning data based on sensor data in the second period prior to the first period as learning data in the normal state. The machine learning processing unit may be configured to update a model that outputs a determination value of the degree of abnormality of the money handling machine with respect to the input of

  A notification processing unit that uses the model to determine the necessity of a notification prompting maintenance on the money handling machine, and an output unit that outputs the notification when the notification processing unit determines that the notification is necessary It may be provided.

  The money processing machine comprises a state determination unit that determines the presence or absence of an abnormality in the money handling machine using the model, the machine learning processing unit generates an error in the money handling machine, and the state determination unit determines the money handling machine The model may be updated when it is determined that there is no abnormality in.

  According to a second aspect of the present invention, the money handling machine state determination system includes a state determination device and a money handling machine, and the state determination device acquires sensor data regarding conveyance of money in the money handling machine. A storage unit for storing a model that outputs an output value indicating an abnormality degree of the money handling machine as an input value, sensor data regarding conveyance of money in the money handling machine, sensor data obtained by the obtaining unit, And a state determination unit that determines the state of the money handling machine using the model stored in the storage unit.

  According to the third aspect of the present invention, the state determination method includes the steps of acquiring sensor data regarding the transportation of money in the money handling machine, the obtained sensor data, and sensor data regarding the transportation of money in the money handling machine Determining the state of the money handling machine using a model that outputs an output value indicating an abnormality degree of the money handling machine with the input value as the input value.

  According to a fourth aspect of the present invention, a program includes the steps of: acquiring, on a computer, sensor data relating to the transport of money in a money handling machine, the obtained sensor data, and a sensor relating to the transport of money in the money handling machine It is a program for executing the step of determining the state of the money handling machine using a model that uses data as an input value and outputs an output value that indicates the degree of abnormality of the money handling machine.

  According to the present invention, the state of the money handling machine can be determined with high accuracy.

It is a schematic block diagram showing functional composition of a money processor state judging system concerning a first embodiment of the present invention. It is a figure which shows the example of installation of the sensor in the conveyance path | route of the coin which concerns on the embodiment. It is a figure which shows the example of a data structure of the data for learning which the state determination apparatus which concerns on the embodiment uses for learning of a model. It is a figure which shows the example of the model which the model memory | storage part which concerns on the embodiment memorize | stores. It is a figure which shows the example of the determination method of the state of a money processor with respect to the data for learning at the time of additional learning by the same embodiment. It is a flowchart which shows the example of the procedure of the pre-processing which the machine learning process part which concerns on the embodiment performs. It is a flowchart which shows the example of the process sequence which the machine learning process part which concerns on the embodiment generate | occur | produces an initial model. It is a flowchart which shows the example of the processing procedure which the machine learning process part which concerns on the embodiment performs additional learning. It is a flowchart which shows the example of the process sequence which the state determination apparatus which concerns on the embodiment determines the necessity of alerting | reporting, and the necessity of additional learning. It is a schematic block diagram showing a functional configuration of a money handling machine state determination system according to a second embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described, but the following embodiments do not limit the invention according to the claims. Moreover, not all combinations of features described in the embodiments are essential to the solution of the invention.

First Embodiment
FIG. 1 is a schematic block diagram showing a functional configuration of the money handling machine state determination system according to the first embodiment of the present invention. As shown in FIG. 1, the money handling machine state determination system 1 includes a money handling machine 100 and a state determination device 200. The money handling machine 100 includes a money handling communication unit 110, a money handling unit 120, a money handling storage unit 180, and a money handling control unit 190. The money processing side storage unit 180 includes a sensor information storage unit 181, a denomination information storage unit 182, an error information storage unit 183, and a maintenance information storage unit 184. The state determination device 200 includes a server communication unit 210, an output unit 220, a server storage unit 280, and a server control unit 290. The server-side storage unit 280 includes a model storage unit 281. The server-side control unit 290 includes a machine learning processing unit 291, a state determination unit 292, and a notification processing unit 293.
Further, the money handling machine 100 and the state determination device 200 are communicably connected via the communication network 900. The number of money handling machines 100 connected in communication with the state determination device 200 may be one or more.

  The communication network 900 mediates communication between the money handling machine 100 and the state determination device 200. Various communication networks can be used as the communication network 900. For example, the communication network 900 may use the Internet or a telephone line, or a combination of these. Alternatively, as the communication network 900, a communication network dedicated to the money handling machine state determination system 1 may be constructed.

The money handling machine state determination system 1 performs money processing and state management of the money handling machine 100 that executes money processing.
The money processor 100 processes money. Below, although the case where money processing machine 100 is a coin wrapping machine is explained to an example, it is not limited to this. As the money handling machine 100, for example, a cash handling machine or a bill coin counting machine performs processing of money, and an error (clogging of money, transport failure, etc.) may occur in processing money, and it relates to processing of money Various devices to be maintained can be used.

The money processing communication unit 110 communicates with another device. In particular, the money handling communication unit 110 communicates with the server communication unit 210 of the state determination device 200 via the communication network 900.
The money processing unit 120 executes money processing. For example, when the money handling machine 100 is a coin wrapping machine, the money handling unit 120 receives an input of coins, sorts out the designated coins from the inserted coins, and packages 50 sorted coins at a time. Generate a coin.
The money processing unit 120 includes a sensor. The machine learning processing unit 291 of the state determination device 200 performs machine learning based on the sensor value by the sensor of the money processing unit 120, thereby generating and updating a model for determining the necessity of maintenance of the money handling machine 100. The sensor included in the money processing unit 120 will be described with reference to FIG. 2 by taking an example of a coin transport path.

  FIG. 2 is a view showing an example of installation of sensors in the coin transport path. The money handling machine 100 sends the inserted coins one by one to the transport path 121. When the money handling machine 100 is a coin wrapping machine, the conveyance path 121 receives insertion of coins one by one from the entrance 122 side, determines the denomination of the coin, excludes objects determined not to be coins, and excludes packaging When the number of coins to be packaged reaches 50 sheets (the number of sheets for packaging), the next coins are not passed until the packaging is completed. The coin referred to here is a 6 denomination coin of 1 yen, 5 yen, 10 yen, 50 yen, 100 yen and 500 yen. The money handling machine state determination system 1 treats coins other than these six denominations as non-coins.

In the configuration of FIG. 2, the sorting module 123 determines whether or not the sorting module 123 itself is a coin or not, and if it is a coin, determines the denomination of the coin.
The stop timing sensor 131 includes a light emitting unit 131a and a light receiving unit 131b, and detects the presence or absence of passage of a coin depending on whether the light from the light emitting unit 131a reaches the light receiving unit 131b. Thereby, the stop timing sensor 131 detects the timing at which the next coin is stopped when the number of coins to be packaged reaches the number of packages.

For example, a mechanism for stopping a coin is provided in the vicinity of the stop timing sensor 131 on the transport path 121 and upstream (on the side of the inlet 122). When the stop timing sensor 131 detects the 50th passage of the coin to be packaged, this mechanism operates to prevent the next coin from passing.
The stop timing sensor 131 is not limited to a specific type of sensor, and may be any sensor that can detect the presence or absence of a coin passing through the area of the stop timing sensor 131 itself. The area of the sensor here is a position where the sensor can detect a passing object such as a coin.

The removal timing sensor 132 detects the timing for removing an object determined not to be a coin. When the removal timing sensor 132 detects that an object determined not to be a coin has reached the removal timing sensor 132 itself, the removal port 124 opens. As a result, an object determined not to be a coin is excluded from the transport path 121 from the outlet 124.
The exclusion timing sensor 132 is configured by a combination of an exclusion timing sensor left 132a and an exclusion timing sensor right 132b. Each of the exclusion timing sensor left 132a and the exclusion timing sensor right 132b is configured using an optical sensor, and detects the presence or absence of an object passing through the area of the sensor itself. In order to prevent detection omission, the exclusion timing sensor left 132a and the exclusion timing sensor right 132b are arranged side by side in the width direction of the transport path 121 (a direction perpendicular to the traveling direction of the currency).
The exclusion timing sensor 132 is not limited to a specific type of sensor, and may be any sensor that can detect the presence or absence of a coin passing through the exclusion timing sensor 132 itself.

  The sorting timing sensor 133 detects the timing of sorting the coins to be packaged. When the sorting timing sensor 133 detects that a coin determined to be a coin other than the one to be packaged has reached the sorting timing sensor 133 itself, the sorting port 125 is opened. Thereby, coins other than those to be packaged are discharged from the sorting port 125 to the outside of the transport path 121. On the other hand, even if the coin determined to be the coin to be packaged reaches the sorting timing sensor 133, the sorting port 125 remains closed. As a result, the coins to be packaged are delivered to the downstream side of the transport path 121, and the 50 coins are collected and packaged.

The sorting timing sensor 133 is configured by a combination of a sorting timing sensor left 133a and a sorting timing sensor right 133b. Each of the sorting timing sensor left 133a and the sorting timing sensor right 133b is configured using an optical sensor, and detects the presence or absence of an object passing through the area of the sensor itself. In order to prevent detection omission, the sorting timing sensor left 133 a and the sorting timing sensor right 133 b are arranged side by side in the width direction of the transport path 121.
The sorting timing sensor 133 is not limited to a specific type of sensor, as long as it can detect the presence or absence of a coin passing through the sorting timing sensor 133 itself.
The stop timing sensor 131, the exclusion timing sensor 132, and the sorting timing sensor 133 are collectively referred to as a sensor 130.

The sensor data used for learning by the machine learning processing unit 291 of the state determination device 200 may be the sensor value (measurement value by the sensor 130) itself or may be obtained by calculation based on the measurement result of the sensor Good.
As shown in FIG. 2, since the machine learning processing unit 291 performs learning using a sensor that is originally installed in the money handling machine 100, it is not necessary to use a separate sensor for learning. In this point, the hardware configuration of the money handling machine 100 does not become complicated.
The sensor used by the machine learning processing unit 291 for learning is not limited to a specific one, as long as sensor data can be obtained according to the presence or absence of an abnormality of the money handling machine 100.

FIG. 3 is a diagram showing an example data structure of learning data used by the state determination apparatus 200 for learning a model.
In the example of FIG. 3, the learning data is configured to include an “input” field indicating input data to the model and an “output” field indicating output data from the model. Each value in the "input" column is normalized to a range of 0 to 1 in order to prevent the range of values (possible maximum and minimum magnitudes) from affecting learning.

The time (normalized value) from the rise of the sensor value of the stop timing sensor 131 to the rise of the sensor value of the exclusion timing sensor 132 is stored in the “first section up” field of the “input” column. . The rising of the sensor value here means that the coin reaches the area of the sensor and the sensor detects the coin. The fall of the sensor value is that the coin has passed the area of the sensor and the sensor does not detect the coin. The “first section” indicates a section from the stop timing sensor 131 to the removal timing sensor 132. "Up" indicates the rise of the sensor value.
In the “second section up” column, the time (normalized value) from the rise of the sensor value of the exclusion timing sensor 132 to the rise of the sensor value of the sorting timing sensor 133 is stored. The “second section” indicates a section from the exclusion timing sensor 132 to the sorting timing sensor 133.

The “first section down” column stores the time (normalized value) from when the sensor value of the stop timing sensor 131 falls to when the sensor value of the exclusion timing sensor 132 falls. “Down” indicates the fall of the sensor value.
The “second section down” column stores the time (normalized value) from when the sensor value of the exclusion timing sensor 132 falls to when the sensor value of the sorting timing sensor 133 falls.

The time (the value obtained by normalizing the time until the sensor value of the stop timing sensor 131 falls) after the sensor value of the stop timing sensor 131 rises is stored in the “131 stay” column. Here, the coin being located in the area of the sensor is referred to as the coin staying in the sensor. The value in the “131 staying” column indicates the time when the coin has stayed in the stop timing sensor 131. The prediction accuracy can be enhanced by using, as an input value, a residence time correlated with the movement time as well as the movement time for the coin to move between the sensors.
In the "132 staying" column, the time (normalized value) from the rise of the sensor value of the exclusion timing sensor 132 to the fall of the sensor value of the exclusion timing sensor 132 is stored. The value in the “132 staying” column indicates the time when the coin is staying in the removal timing sensor 132.

In the “133 stagnation” column, the time (normalized value) from the rise of the sensor value of the sorting timing sensor 133 to the fall of the sensor value of the sorting timing sensor 133 is stored. The value in the “133 staying” column indicates the time when the coin is staying in the sorting timing sensor 133.
A predetermined value corresponding to the state of the transport path 121 is stored in the “output” column. Specifically, when the transport path 121 is normal, "0.9" is stored in the output field. If the transport path 121 is abnormal, “0.1” is stored in the output field.

In the development stage, the production stage and the test stage of the money handling machine 100, it is conceivable to conduct experiments using a real machine of the money handling machine 100 to acquire learning data. In this case, it is possible to acquire data at the time of abnormality of the conveyance path 121 by acquiring data in a state in which dirt is attached to the conveyance path 121 or in a simulated abnormal state in which foreign matter is inserted in the conveyance path 121. it can. On the other hand, by acquiring data in a normal state where a simulated abnormality is not generated in the conveyance path 121, it is possible to acquire data when the conveyance path 121 is normal.
The normal state of the transport path 121 corresponds to an example of the normal state of the money processing unit 120. The normal state of the money processing unit 120 is also referred to as the normal state of the money processing machine 100. An abnormal state of the transport route 121 corresponds to an example of an abnormal state of the money processing unit 120. The abnormal state of the money processing unit 120 is also referred to as an abnormal state of the money processing machine 100.

When learning data is acquired at the time of operation of the money handling machine 100, it is not always clearly determined whether the transport path 121 is normal or abnormal unlike the time of experiment. In this case, as will be described later, for example, the error occurrence to two days before is made abnormal, and the error occurrence is made two days to four days before being normal. May be determined.
Even when the money handling machine 100 handles bills, as in the case of handling coins, it is possible to use, as sensor data, the moving time for the bills to move between sensors and the staying time for the bills to stay at the sensors. Furthermore, when the money handling machine 100 handles bills, the degree of skewing (inclination to the transport direction) of the bills and the rotational speed of the transport motor can also be used as sensor data.

The money processing side storage unit 180 stores various data. The money handling side storage unit 180 is configured using a storage device provided in the money handling machine 100.
The sensor information storage unit 181 stores sensor data in association with the type of coin (denomination) and stores each piece of money.
In addition, when sensor data are obtained by calculation from the measurement result of a sensor, you may make it the sensor information storage part 181 memorize | store the measurement result of a sensor. In this case, the state determination device 200 may generate sensor data from the measurement results of the sensor.

The frequency at which the money handling machine 100 transmits sensor data to the state determination device 200 can be set to various frequencies. For example, the money handling machine 100 may transmit sensor data at a short frequency, such as transmitting sensor data to the state determination device 200 for each operation. One operation here is a group of operations performed by the money handling machine 100. When the money handling machine 100 is a coin wrapping machine, a series of operations for winding one bar can be one operation. Alternatively, a series of operations in one transaction, such as one deposit to the money handling machine 100, can be one operation.
Among the sensor data stored in the sensor information storage unit 181, sensor data that has been transmitted to the state determination device 200 can be deleted. Since the money handling machine 100 transmits sensor data to the state determination device 200 at a short frequency, the storage capacity of the sensor information storage unit 181 may be small.

The denomination information storage unit 182 stores data on each denomination. For example, the denomination information storage unit 182 stores information for the sorting module 123 to determine the type (denomination) of the coin, such as the size or magnetic information of the coin for each denomination.
The error information storage unit 183 stores history information of errors generated by the money processing unit 120. The error referred to here is a notification of abnormality which is performed by an apparatus such as the money handling machine 100 by detecting an abnormality in the apparatus itself.
The maintenance information storage unit 184 stores history information of maintenance performed on the money processing unit 120.
The money processing side control unit 190 controls each unit of the money processing machine 100 to execute various processes. The money processing side control unit 190 is configured, for example, by a CPU (Central Processing Unit, central processing unit) included in the money processing machine 100 reading out a program from the money processing side storage unit 180 and executing it.

  The state determination device 200 manages the state of the money handling machine 100. The state determination device 200 acquires sensor data from the money handling machine 100, and determines the necessity of a notification prompting maintenance based on the acquired sensor data. Further, the state determination device 200 updates the model for notification using the obtained sensor data. The state determination device 200 is configured using, for example, a computer such as a workstation.

The state determination device 200 may be configured as a server, and a model may be commonly used for the plurality of money processors 100. As a result, the state determination device 200 can acquire a large number of sensor data, and can increase the model update frequency more than when providing a model for each money handling machine 100. The state determination apparatus 200 can obtain a highly accurate model in that learning can be performed using a large number of learning models.
Alternatively, the state determination device 200 may include a model for each money handling machine 100. As a result, the state determination device 200 can reflect the characteristics of each money handling machine 100 on the model, and in this respect, a model with high accuracy can be obtained.

The server communication unit 210 communicates with other devices. In particular, the server side communication unit 210 communicates with the money processing side communication unit 110 of the money handling machine 100 via the communication network 900.
The server side communication unit 210 corresponds to an example of an acquisition unit that acquires sensor data related to the transport of money in the money handling machine 100.
The output unit 220 outputs notification information notifying that maintenance of the money handling machine 100 is necessary. For example, a display and a speaker included in the state determination device 200 may be used as the output unit 220, a message indicating that maintenance of the money handling machine 100 is necessary may be displayed on the display, and a notification sound may be output from the speaker . Alternatively, the server side communication unit 210 may be used as the output unit 220 to transmit notification information to a terminal device used by the maintenance worker, such as a smartphone (Smartphone) or a personal computer (PC) of the maintenance worker. It is also good.

The server-side storage unit 280 stores various data. The server-side storage unit 280 is configured using a storage device included in the state determination device 200.
The model storage unit 281 stores a model for determining the presence or absence of an abnormality in the money handling machine 100. The model storage unit 281 stores a model for each denomination.
The model storage unit 281 corresponds to an example of a storage unit that stores a model that outputs an output value indicating an abnormality degree of the money handling machine 100 using sensor data related to conveyance of money in the money handling machine 100 as an input value.

FIG. 4 is a diagram showing an example of a model stored in the model storage unit 281. As shown in FIG.
The model shown in FIG. 4 is composed of a three-layer neural network of an input layer 511, an intermediate layer 513 and an output layer 515. The input layer 511 includes m (m is a positive integer) input nodes 511-1 to 511-m. The middle layer 513 includes n (n is a positive integer) middle nodes 513-1 to 513-n. The output layer 515 comprises one output node 515-1.

Each of the input nodes 511-1 to 511-m and each of the intermediate nodes 513-1 to 513-n are connected by input-intermediate paths 512-1-1 to 511-m-n. The input-intermediate paths 512-1 to 512-m-n are collectively referred to as an input-intermediate path 512.
Each of the intermediate nodes 513-1 to 513-n and the output node 515-1 are connected by an intermediate-output route 514-1-1 to 514-n-1. The intermediate-output paths 514-1 to 514-n-1 are collectively referred to as an intermediate-output path 514.

Each of these paths is constituted by a sigmoid function, and a weight (weighting coefficient) is set to each path. In the model shown in FIG. 4, when input of values is received at each node of the input layer 511, the input value is multiplied by the load at the input-intermediate path 512 and output to each node of the intermediate layer 513. At each node of the middle layer 513, the input values are averaged.
Further, in the model shown in FIG. 4, the value input to the intermediate layer 513 is multiplied by the load in the intermediate-output path 514 and output to the node (output node 515-1) of the output layer 515. The nodes of the output layer 515 average the input values.
In this configuration, the model shown in FIG. 4 receives real-valued input data in the range of 0 to 1 at each node of the input layer 511, and the real value in the range of 0 to 1 from the node of the output layer 515 Output the output data of
The value of the output data is compared with the threshold, and when the value of the output data is larger than the threshold, it can be determined that the money handling machine 100 has an abnormality.

The number of nodes in the input layer 511 can be, for example, seven. The number of nodes in the middle layer 513 can be, for example, 50. The number of nodes in the output layer 515 may be one, for example.
However, the model used by the state determination device 200 is not limited to that shown in FIG. For example, two or more intermediate layers may be provided. In addition, the number of nodes in each layer is not limited to the number described above. Furthermore, the model used by the state determination device 200 may be a learnable model, and is not limited to the neural network.

The server-side control unit 290 controls each unit of the state determination device 200 to perform various processes. The server-side control unit 290 is configured by the CPU included in the state determination device 200 reading and executing a program from the server-side storage unit 280.
The machine learning processing unit 291 generates and updates each of the models for each denomination stored in the model storage unit 281 by machine learning. Specifically, the machine learning processing unit 291 adjusts each load of the input-intermediate path 512 and each load value of the intermediate-output path 514 using the learning data.

Since the machine learning processing unit 291 determines the state of the money handling machine 100 using a neural network, the determination time can be shortened compared to the case where the determination is performed manually, and it can not be predicted by humans. It is possible to determine whether or not maintenance is necessary by making a determination taking into consideration the correlation among various data.
As a learning method used by the machine learning processing unit 291, a combination of the error back propagation method and the momentum method can be used. In this case, the momentum constant is set to, for example, 0.9. However, the learning method used by the machine learning processing unit 291 is not limited to a specific method, and various learning methods can be applied.

The machine learning processing unit 291 generates a model by initial learning, and repeats model updating by additional learning. In the initial learning, the machine learning processing unit 291 uses the learning data obtained in the experiment using the money handling machine 100 described above to determine each weight of the input-intermediate path 512 and each intermediate-output path 514. Set the load value. In the additional learning, the machine learning processing unit 291 uses the learning data obtained at the time of operation of the money handling machine 100 and uses the values of each load of the input-intermediate path 512 and each load of the intermediate-output path 514 Update.
In the case of additional learning, it is not always clear whether the money handling machine 100 is normal or abnormal at each time when learning data is obtained. Therefore, the machine learning processing unit 291 determines the state of the money handling machine 100 to be either normal or abnormal based on the occurrence of an error, and sets an output value in learning data.

FIG. 5 is a diagram showing an example of a method of determining the state of the money handling machine 100 with respect to learning data at the time of additional learning.
In the example of FIG. 5, the machine learning processing unit 291 learns learning data obtained at an error from the time when an error occurs to the timing two days before the time when an error occurs in the money processor 100 as a reference time. The data for learning is determined, and the data for learning obtained at the timing from 2 days to 4 days before the error occurrence is determined as the data for learning at normal time.

The machine learning processing unit 291 determines the learning data from the error occurrence time to the learning threshold as data at the abnormal time based on the learning threshold, and from the learning threshold to a predetermined period in the past (for example, from the error occurrence to the learning threshold) Data for the same period as in (1) is determined as normal data. The learning threshold mentioned here is a threshold serving as a reference for distributing learning data for additional learning into data at normal time and data at abnormal time. In the example of FIG. 5, the learning threshold is set two days before the error occurrence.
The period from the error occurrence to the two days before the occurrence of the error shown in FIG. 5 corresponds to an example of the first past period from the error occurrence. The period from two days before the error occurrence to four days before corresponds to an example of a second period that is earlier than the first period.

  In the case where the notification processing unit 293 determines that there is no need for notification, the machine learning processing unit 291 determines that there is an error in the money handling machine 100, and the state determining unit 292 determines that there is no abnormality in the money handling machine 100 To update the model. If the notification processing unit 293 determines that the notification is necessary, it is considered that the current model can detect an abnormality and it is not necessary to update the model in any case where the state determination unit 292 determines that there is an abnormality. Be In this case, the load of the machine learning processing unit 291 may be reduced by suppressing the machine learning processing unit 291 from updating the model.

  The state determination unit 292 determines the presence or absence of abnormality in the money handling machine 100 using the model generated and updated by the machine learning processing unit 291. In particular, the state determination unit 292 determines the presence or absence of an abnormality for each piece of money and for each of a plurality of pieces of money. Even if it is determined that each piece of money is judged to be normal, there is a possibility that an abnormal tendency strongly predicts the occurrence of an error when viewed as a whole of a plurality of money or a plurality of denominations of money. The state determination unit 292 determines the presence or absence of an abnormality for each coin and for each of a plurality of pieces of money, whereby the tendency of the abnormality is strong in the whole plurality of money or the plurality of money types. It can be determined that the case is abnormal.

The notification processing unit 293 uses the model generated and updated by the machine learning processing unit 291 to determine the necessity of a notification prompting maintenance on the money handling machine 100. The notification processing unit 293 determines the presence or absence of an abnormality for each piece of money and each of a plurality of pieces of money using the model generated and updated by the machine learning processing unit 291 for each denomination.
Both the state determination unit 292 and the notification processing unit 293 perform the determination by applying a threshold to the same output from the model. On the other hand, the determination threshold used by the state determination unit 292 for determining the necessity of model update and the notification threshold used by the notification processing unit 293 can be set to different values. For example, a fixed value such as 0.5 may be set as the determination threshold, and the value of the notification threshold may be set variably in accordance with the request of the customer operating the money handling machine 100. Specifically, in the case of a customer who desires frequent maintenance, by setting the notification threshold value to a smaller value such as 0.4, the state determination device 200 promotes maintenance at a relatively minor abnormality stage. It will notify. On the other hand, in the case of a customer who desires that the number of times of maintenance is small, the notification threshold value is set to a large value such as 0.6 to lower the frequency of maintenance notification by the state determination apparatus 200, and the tendency for abnormality Maintenance can be carried out if it is strengthened.

Next, the operation of the state determination device 200 will be described with reference to FIGS.
FIG. 6 is a flowchart illustrating an example of a pre-processing procedure performed by the machine learning processing unit 291. In this pre-processing, the machine learning processing unit 291 generates learning data from sensor data.
In the process of FIG. 6, the machine learning processing unit 291 calculates the value of the item of learning data from the sensor data value (step S101). For example, the machine learning processing unit 291 calculates the difference between the rise time and the fall time of the sensor value in a certain sensor as the residence time of money in the sensor. As described above, the rise time of the sensor value is the time when the coin reaches the sensor and the sensor value switches from the off (non-detection) value to the on (detection) value. The fall time of the sensor value is the time when the money passes through the sensor and the sensor value switches from the on (detection) value to the off (non-detection) value. The residence time of money in the sensor is the time from when the money reaches the sensor until it passes through the sensor.
However, the money handling machine 100 may perform the process of step S101. In this case, the value of the item of the learning data is transmitted from the money handling machine 100 to the state determination device 200, and the state determination device 200 does not perform the process of step S101.

Next, the machine learning processing unit 291 performs normalization to set the value of each item of the learning data to a value in the range of 0 to 1 (step S102). This is to avoid the difference in the size of the range (range of possible values) of each item of the learning data from being different in the degree of influence on the learning.
For example, the machine learning processing unit 291 subtracts a predetermined value determined as the minimum value from the value of the item of the learning data. Furthermore, the machine learning processing unit 291 divides the obtained subtraction result by a value obtained by subtracting a predetermined value determined as the minimum value from a predetermined value determined as the maximum value of the item of the learning data.

Next, the machine learning processing unit 291 performs batting processing to remove the combination of the same value from the learning data (step S103).
Here, when the same combination of values is included in the learning data, whether the combination of the values is treated as normal or abnormal may differ depending on the operation of the money handling machine 100. If the same combination of values is assigned to both normal and abnormal, there is a possibility that appropriate learning can not be performed. Therefore, the machine learning processing unit 291 performs batting processing to avoid that the combination of the same value is allocated to both normal and abnormal.

For example, the machine learning processing unit 291 counts the number of times normally allocated and the number of times abnormally allocated for the occurrence of the combination of the same value in the learning data, and the combination of the larger value is used. You may leave only one. Alternatively, the machine learning processing unit 291 may delete all combinations of the same value in the learning data.
After step S103, the process of FIG. 6 ends.

FIG. 7 is a flowchart illustrating an example of a processing procedure in which the machine learning processing unit 291 generates an initial model.
In the process of FIG. 7, the machine learning processing unit 291 acquires sensor data (step S201).
Next, the machine learning processing unit 291 performs the preprocessing described with reference to FIG. 6 (step S202). Thereby, the machine learning processing unit 291 acquires learning data from the sensor data.
Then, the machine learning processing unit 291 performs learning using the obtained learning data, and constructs a model (step S203). The machine learning processing unit 291 causes the model storage unit 281 to store the obtained model as a learned model (step S204).
After step S204, the process of FIG. 7 ends.

FIG. 8 is a flowchart illustrating an example of a processing procedure in which the machine learning processing unit 291 performs additional learning. As described later with reference to FIG. 9, the machine learning processing unit 291 performs the process of FIG. 8 when an error occurs in the money handling machine 100.
In the process of FIG. 8, the machine learning processing unit 291 sets a learning threshold two days before the occurrence of an error (step S311). As described above, the learning threshold is a threshold serving as a reference for distributing learning data for additional learning into normal data and abnormal data.

Here, when learning data is acquired by experiment or simulation, the state of the money handling machine 100 is set on the assumption that it is normal or abnormal, and the obtained learning data is distributed to either normal or abnormal. be able to.
On the other hand, when an error occurs during operation, it is not always determined whether the data in the period before the error should be the data at the normal time but the data at the abnormal time. Therefore, the machine learning processing unit 291 sets the learning threshold so that the data in the period before the error is used for learning (supervised learning), and uses the set learning threshold to use the data in the period before the error. Assign to either normal or abnormal.

The specific value of the learning threshold is not limited to two days before the error occurrence shown in FIG. For example, the learning threshold may be set in units of days other than two days, such as the machine learning processing unit 291 setting the learning threshold one day before the error occurrence. Alternatively, the machine learning processing unit 291 may set the learning threshold in units other than daily units, such as setting the learning threshold six hours before the occurrence of an error. The time used for the learning threshold may be a simple time (a time on a calendar) or an operating time of the money handling machine 100.
Alternatively, the machine learning processing unit 291 may set a learning threshold based on a standard other than time, such as 1000 sheets before an error occurs in the number of conveyed money in the money processing machine 100.

  The machine learning processing unit 291 that sets the learning threshold in step S311 acquires sensor data that is a basis of learning data (step S312). For example, the machine learning processing unit 291 acquires sensor data for four days which is twice of two days which is setting of a learning threshold. However, the period during which the machine learning processing unit 291 acquires sensor data is not limited to this. For example, when the learning threshold value is set to 1000 sheets before the occurrence of an error in the number of conveyed money sheets in the money handling machine 100, the machine learning processing unit 291 acquires sensor data from 2000 sheets before to the time of error occurrence. You may do it.

Next, the machine learning processing unit 291 divides the obtained sensor data into normal data and abnormal data (step S313). The normal data here is data that is treated as if the money handling machine 100 is in the normal state. The abnormal data is data to be treated as if the money handling machine 100 is in an abnormal state.
Next, the machine learning processing unit 291 performs the preprocessing described with reference to FIG. 7 (step S314). Thereby, the machine learning processing unit 291 acquires learning data based on the sensor data.
Then, the machine learning processing unit 291 performs machine learning using the obtained data for learning, and updates the learned model (step S315).
After step S315, the process of FIG. 8 ends.

FIG. 9 is a flowchart illustrating an example of a processing procedure in which the state determination device 200 determines whether notification is necessary and whether additional learning is necessary.
In the process of FIG. 9, the server-side control unit 290 acquires sensor data and performs normalization (step S401). The sensor data obtained in step S401 is used to determine whether notification is necessary and whether model updating is necessary.

Next, the server-side control unit 290 determines whether an error has occurred in the money handling machine 100 (step S402). Specifically, the money handling machine 100 transmits information of the generated error to the state determination device 200 in addition to the sensor data. The server-side control unit 290 determines the presence or absence of an error occurrence in the money handling machine 100 by determining the presence or absence of the generated error information transmitted by the money handling machine 100.
When it is determined that an error has occurred (step S402: YES), the state determination unit 292 selects a model to be used for the state determination of the money handling machine 100 (step S411). Specifically, the state determination unit 292 selects a model of the denomination to be determined among the models stored by the model storage unit 281 for each denomination.

Next, the state determination unit 292 inputs the sensor data obtained in step S401 into the selected model, and calculates an output value from the model (step S412). An output value obtained by inputting sensor data to a model for each denomination is also referred to as a denomination output value.
Then, the state determination unit 292 determines whether the calculated denomination output value is less than or equal to the determination threshold (step S404). When it is determined that the denomination output value is equal to or less than the determination threshold (step S413: NO), the state determination unit 292 determines whether the determination has been performed for data of all denominations (step S431). If it is determined that there is a denomination for which the determination has not been made (step S431: NO), the process returns to step S411, and the processing for the denomination for which the determination has not been made is continued.
On the other hand, if it is determined in step S431 that determination has been made for all denominations (step S431: YES), the processing of FIG. 9 is ended.

On the other hand, when it is determined in step S413 that the denomination output value is larger than the determination threshold (step S413: YES), the machine learning processing unit 291 performs the additional learning described with reference to FIG. 8 (step S421). In this case, while detecting the tendency of an error and calculating the denomination output value of a small value, since the actually obtained denomination output value is larger than the judgment threshold, the model is updated. The machine learning processing unit 291 performs additional learning.
After step S421, the process proceeds to step S431.

  On the other hand, when it is determined in step S402 that no error has occurred (step S402: NO), the notification processing unit 293 selects a model and a notification threshold to be used for determining the necessity of notification (step S411). Specifically, the notification processing unit 293 selects a model of the denomination to be determined among the models stored by the model storage unit 281 for each denomination. Also, there are two types of notification threshold: denomination notification threshold and overall notification threshold, and the server-side storage unit 280 stores the denomination notification threshold for each denomination, and the entire notification threshold is common to each denomination. Remember. The notification processing unit 293 selects the denomination notification threshold of the denomination to be determined among the denomination notification thresholds stored by the server-side storage unit 280 for each denomination.

Next, the notification processing unit 293 inputs the sensor data obtained in step S401 into the selected model, and calculates an output value from the model (step S442). The process at step S442 is similar to the process at step S412.
Then, the notification processing unit 293 determines whether the calculated denomination output value is equal to or less than the denomination notification threshold (step S443).
When it is determined that the denomination output value is equal to or less than the denomination notification threshold (step S443: YES), the notification processing unit 293 notifies the maintenance worker of preventive maintenance (step S471). Preventive maintenance here is maintenance for preventing problems such as occurrence of an error.
After step S471, the process of FIG. 9 ends.

  On the other hand, when it is determined in step S443 that the denomination output value is larger than the denomination notification threshold (step S443: NO), the notification processing unit 293 determines whether or not the determination has been performed on data of all denominations. (Step S451). If it is determined that there is a denomination for which the determination has not been made (step S451: NO), the process returns to step S441, and the processing for the denomination for which the determination has not been made is continued.

On the other hand, when it is determined in step S451 that the determination has been performed for all denominations (step S451: YES), the notification processing unit 293 calculates the overall output value from the denomination output value using the denomination ratio ( Step S461).
Then, the notification processing unit 293 determines whether the calculated entire output value is equal to or less than the entire notification threshold (step S462). The entire notification threshold is set to a constant value in advance, for example, to 0.5.

If it is determined that the entire output value is equal to or less than the entire notification threshold (step S462: YES), the process proceeds to step S471.
On the other hand, when it is determined in step S462 that the entire output value is larger than the entire notification threshold (step S462: NO), the process of FIG. 9 is ended.
On the other hand, if it is determined in step S431 that the denomination output value is equal to or greater than the determination threshold (step S431: NO), the process proceeds to step S451.

As described above, the server-side communication unit 210 acquires sensor data regarding the transport of money in the money handling machine 100. The model storage unit 281 stores a model that uses sensor data relating to the transport of money in the money handling machine 100 as an input value and outputs an output value indicating an abnormality degree of the money handling machine 100. The state determination unit 292 determines the state of the money handling machine 100 using the sensor data acquired by the server side communication unit 210 and the model stored in the model storage unit 281.
The state determination unit 292 can determine the state of the money handling machine 100 using the model in this manner, so that the state of the money handling machine 100 can be determined with high accuracy. For example, based on the relationship between the sensor data and the money handling machine 100 that the state determination unit 292 determines the state of the money handling machine 100 using the model obtained by machine learning, the determination by a human is not noticeable. There is a possibility that the tendency of abnormality of the money handling machine 100 can be detected.

In addition, the model storage unit 281 stores a model for each denomination of money. The state determination unit 292 determines the presence or absence of an abnormality for each denomination of money.
As described above, the state determination unit 292 can detect the abnormality when the abnormality occurs in the money processor 100 for a specific denomination by determining the presence or absence of the abnormality for each denomination of money, and the abnormality can be detected. The denomination in which the abnormality occurs can be identified. According to the state determination device 200, the state of the money handling machine 100 can be accurately determined at this point.

Further, the state determination unit 292 determines the presence or absence of an abnormality of the money handling machine 100 for all the money of a plurality of denominations based on the output value from the model for each denomination of money.
Even in the case where it is determined that the denominations of money are judged to be normal, there is a possibility that the tendency of an abnormality of the money handling machine 100 strongly predicts the occurrence of an error when viewed across multiple denominations. Since the state determination unit 292 determines the presence or absence of abnormality of the money handling machine 100 for all the money of a plurality of denominations, it can be determined as abnormal even when the tendency of the abnormality is strong in the whole plurality of denominations .

Further, when an error occurs in the money processor 100, the machine learning processing unit 291 detects sensor data obtained by measurement by the sensor of the money processor 100 in which the error occurred, in the first past period from the error occurrence. By using learning data based on sensor data as learning data at the time of abnormality, machine learning is performed by using learning data based on sensor data in the second period past the first period as learning data at the normal time. Update.
Thus, the state determination device 200 can update the state determination process of the money handling machine 100 using the data at the time of operation of the money handling machine 100. Here, when operating the money handling machine 100, it is not always clear whether the state of the money handling machine 100 is normal or abnormal. On the other hand, in the state determination apparatus 200, the machine learning processing unit 291 sets the learning data in the first period as learning data in the abnormal state, and the learning data in the second period as learning data in the normal state. Do. Thus, the state determination device 200 can update the state determination process of the money handling machine 100 using data at the time of operation of the money handling machine 100.

In addition, as the internal state of the money handling machine 100 changes, the money handling machine 100 in a good condition such as after maintenance gradually deteriorates due to adhesion of dirt and consumption of parts over time, leading to an error occurrence It is conceivable.
In this case, the machine learning processing unit 291 performs machine learning using learning data in the past first period near the time of error occurrence as learning data at the time of abnormality, a period in which the state of the money processor 100 is relatively deteriorated. Machine learning can be performed by using the learning data in 3 as learning data at the time of abnormality. In addition, the machine learning processing unit 291 performs machine learning using learning data in the second period before the first period as learning data in the normal state, so that the period in which the state of the money handling machine 100 is relatively good. Machine learning can be performed by using the learning data in step S2 as learning data in the normal state. In this respect, it is expected that the machine learning processing unit 291 can perform machine learning with high accuracy, and obtain a high accuracy model.

Further, the notification processing unit 293 determines the necessity of the notification for prompting maintenance on the money handling machine 100 using the model stored in the model storage unit 281. When the notification processing unit 293 determines that the notification is necessary, the output unit 220 outputs the notification.
According to the state determination device 200, the occurrence of an error in the money handling machine 100 can be reduced by the maintenance worker performing maintenance on the money handling machine 100 according to the notification output from the output unit 220.

  In addition, the hardware configuration can be changed more easily than when the money handling machine 100 side performs this processing by performing the determination of the necessity of the notification for prompting maintenance by the state determination device 200. For example, when processing a large amount of data, it is relatively easy to upgrade the CPU of the state determination device 200 and add a storage medium such as a hard disk. The same applies to the state determination device 200 performing the state determination of the money handling machine 100.

The state determination unit 292 also determines the presence or absence of an abnormality in the money handling machine 100 using the model stored in the model storage unit 281. The machine learning processing unit 291 updates the model when an error occurs in the money handling machine 100 and the state determination unit 292 determines that there is no abnormality in the money handling machine 100.
Here, when the state determination unit 92 determines that there is an abnormality in the money handling machine 100 when an error occurs in the money handling machine 100, it is considered that model updating is unnecessary in that it is correctly determined. In this case, the load of the machine learning processing unit 291 may be reduced by suppressing the machine learning processing unit 291 from updating the model.

Note that the processing performed by the state determination unit 292 in steps S402 to S451 and the processing performed in steps S461 to S471 can be various types of processing for a unit in which the processing in steps S461 to S471 is larger. .
For example, the state determination unit 292 may perform determination for each piece of money in steps S402 to S451, and may perform determination for each of a plurality of coins in steps S461 to S471. When the money handling machine 100 is a coin wrapping machine, the determination may be performed every 50 sheets (every one bar) in steps S461 to S471.
Alternatively, the state determination unit 292 may perform determination for each denomination in steps S402 to S451, and perform determination for each of a plurality of denominations (for each of a plurality of denominations) in steps S461 to S471. . For example, when the state determination unit 292 performs determination on all six types of denominations in steps S461 to S471, the entire output may be calculated based on Formula (1) in step S461.

Here, n1, n5, n10, n50, n100 and n500 are the average values of sensor data for 1-yen, 5-yen, 10-yen, 50-yen, 500-yen and 500-yen, respectively. . p1, p5, p10, p50, p100 and p500 are the ratio of 1-yen, 5-yen, 10-yen, 50-yen, 500-yen and 500-yen, respectively, to the total number of balls .
The state determination unit 292 performs weighted averaging of sensor data according to the number of sheets for each denomination, using Equation (1).
Alternatively, the state determination unit 292 may perform the determination for each denomination in steps S402 to S451, and may perform the determination for each operation in steps S461 to S471.

Second Embodiment
In the first embodiment, the case where the state determination device 200 performs notification has been described as an example. On the other hand, the money processor may notify. In the second embodiment, a case where the money handling machine performs notification will be described.
FIG. 10 is a schematic block diagram showing a functional configuration of the money handling machine state determination system according to the second embodiment of the present invention. As shown in FIG. 10, the money handling machine state determination system 2 includes a money handling machine 300 and a model management device 400. The money handling machine 300 includes a money handling unit 120, an output unit 220, a money handling IF (Interface) unit 310, a money handling storage unit 380, and a money handling control unit 390. The money processing storage unit 380 includes a sensor information storage unit 181, a denomination information storage unit 182, an error information storage unit 183, a maintenance information storage unit 184, and a learned model storage unit 385. The money processing side control unit 390 includes a state determination unit 292 and a notification processing unit 293. The model management device 400 includes a server-side storage unit 280, a server-side IF unit 410, and a server-side control unit 490. The server-side storage unit 280 includes a model storage unit 281. The server-side control unit 490 includes a machine learning processing unit 291.
Among parts in FIG. 10, parts having the same functions as those in FIG. 1 have the same reference numerals (120, 181, 182, 183, 184, 220, 280, 281, 291, 292, 293). The explanation is omitted.

  In the money handling machine state determination system 1 shown in FIG. 1, the money handling machine 100 and the state determination device 200 have a so-called online configuration in which communication connection is made via the communication network 900. On the other hand, in the money handling machine state determination system 2 shown in FIG. 10, the money handling machine 300 and the model management device 400 exchange data via the portable storage medium 910, so-called an off-line configuration. There is. To that end, the money handling machine 300 includes a money handling IF unit 310, and the model management device 400 includes a server side IF unit 410.

  The portable storage medium 910 is a portable storage medium, and mediates the exchange of data between the money handling machine 300 and the model management device 400. The portable storage medium 910 is connected to the money handling machine 300 to write data such as sensor data, and then connected to the model management apparatus 400 to read out the data written to the money handling machine 300. In addition, the portable storage medium 910 is connected to the model management device 400 to write data such as a learned model, and then connected to the money handling machine 300 to read the data written to the model management device 400. . The connection referred to here may be any connection that can input and output data. For example, the portable storage medium 910 may be physically connected to the money handling machine 300 or the model management device 400, such as being inserted into the port of the money handling machine 300 or the model management device 400. Alternatively, the portable storage medium 910 may be communicably connected to the money handling machine 300 or the model management device 400 by near field communication or the like.

The portable storage medium 910 may be, for example, a universal serial bus (USB) flash memory, but is not limited thereto.
The money handling IF unit 310 and the server IF unit 410 each input / output data to / from the portable storage medium 910. For example, when the portable storage medium 910 is a USB flash memory, the money processing IF unit 310 and the server IF unit 410 are each configured to include a USB port.

  The money handling machine 300 may be provided with the money handling communication unit 110 in addition to the money handling IF unit 310. The model management apparatus 400 may include the server side communication unit 210 in addition to the server side IF unit 410. Further, in the configuration of FIG. 1, the money handling machine 100 may be provided with a money handling IF unit 310 in addition to the money handling communication unit 110. The state determination device 200 may include the server side IF unit 410 in addition to the server side communication unit 210.

  In the money handling machine state determination system 1, the state determination device 200 includes the output unit 220, and the server-side control unit 290 of the state determination device 200 includes the state determination unit 292 and the notification processing unit 293. In the processor state determination system 2, the money processor 300 includes an output unit 220, and the money processing control unit 390 of the money processor 300 includes a state determination unit 292 and a notification processing unit 293. Further, the money processing side storage unit 380 includes a learned model storage unit 385 in addition to the respective units included in the money processing machine 100 of FIG. 1.

In the money handling machine state determination system 1, the state determination device 200 determines the state of the money handling machine 100 to perform notification, while in the money handling machine state determination system 2, the money handling machine 300 itself has the above configuration. Determines the state of the money processor 300 and gives notification.
In the money handling machine state determination system 2, the model management device 400 learns a model, and the money handling machine 300 performs state determination and notification using the model. For that purpose, the portable storage medium 910 transmits the model generated or updated by the machine learning processing unit 291 of the model management device 400 to the money handling machine 300. In the money handling machine 300, a learned model storage unit 385 stores the model. The state determination unit 292 reads the model from the learned model storage unit 385, and determines the state of the money handling machine 300.

For example, when an error occurs in the money handling machine 300 and a maintenance worker is called, the maintenance worker brings the latest model copied from the model management device 400 to the portable storage medium 910 and goes to the site . The maintenance worker writes (stores) the latest model copied to the portable storage medium 910 in the learned model storage unit 385 at the site.
For that purpose, the sensor information storage unit 181 of the money handling machine 300 stores sensor data for the last 4 days, and when an error occurs in the money handling machine 300, holds the data from 4 days before the error occurrence. .

The process performed by the state determination unit 292 of the money handling machine 300 is the same as the process performed by the state determination unit 292 of the state determination device 200 described with reference to FIG. Also in the second embodiment, the model management apparatus 400 performs model generation and update. In step S 441 of FIG. 9, the machine learning processing unit 491 of the model management device 400 updates the model by performing additional learning based on sensor data obtained from the money processor 300 via the portable storage medium 910. Do. The updated model is transmitted to the money handling machine 300 via the portable storage medium 910, and the learned model storage unit 385 stores the model.
In the second embodiment, the money processing unit 120 corresponds to an example of an acquisition unit that acquires sensor data regarding conveyance of money in the money processing machine in that the money processing unit 120 includes a sensor and acquires sensor data. The learned model storage unit 385 corresponds to an example of a storage unit that stores a model that outputs an output value indicating an abnormality degree of the money handling machine 300 using sensor data related to conveyance of money in the money handling machine 300 as an input value.
In the second embodiment, the money handling machine 300 corresponds to an example of the state determination device, and determines the state of the money handling machine 300 itself.

As described above, the money processing unit 120 acquires sensor data related to the transfer of money in the money processing machine 300. The learned model storage unit 385 stores a model that uses sensor data relating to the transport of money in the money handling machine 300 as an input value and outputs an output value indicating an abnormality degree of the money handling machine 300. The state determination unit 292 determines the state of the money processor 300 using the sensor data acquired by the money processing unit 120 and the model stored in the learned model storage unit 385.
The state determination unit 292 can determine the state of the money handling machine 300 using the model in this manner, so that the state of the money handling machine 300 can be determined with high accuracy. For example, based on the relationship between the sensor data and the money processor 300 that the state determination unit 292 determines the state of the money processor 300 using the model obtained by machine learning, the determination by a human does not notice. There is a possibility that the tendency of abnormality of the money handling machine 300 can be detected.

Further, in the configuration shown in FIG. 10, the present invention can be easily applied to a money handling machine having no communication function, in that the money handling machine 300 does not need to be connected to a communication network.
In addition, since the communication network is not used to exchange data between the money handling machine 300 and the model management device 400, the bandwidth of the communication network is not compressed.
In each of the above embodiments, the money handling machine 300 and the model management device 400 are configured as separate devices, but the money handling machine 300 may be configured as an integrated device including the respective units of the model management device 400.

Note that a program for realizing all or a part of functions of the money processing control unit 190, the server control unit 290, the money processing control unit 390, and the server control unit 490 is a computer readable recording medium. The processing may be performed by causing the computer system to read and execute the program recorded on the recording medium after recording. Here, the “computer system” includes an OS and hardware such as peripheral devices.
The term "computer-readable recording medium" refers to a storage medium such as a flexible disk, a magneto-optical disk, a ROM, a portable medium such as a ROM or a CD-ROM, or a hard disk built in a computer system. The program may be for realizing a part of the functions described above, or may be realized in combination with the program already recorded in the computer system.

  Although the embodiments of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and design changes and the like within the scope of the present invention are also included.

1, 2 Money handling machine state judgment system 100, 300 Money handling machine 110 Money handling side communication unit 120 Money handling unit 180, 380 Money handling side storage unit 181 Sensor information storage unit 182 Money type information storage unit 183 Error information storage unit 184 Maintenance information storage unit 190, 390 Money processing side control unit 200 State determination device 210 Server side communication unit 220 Output unit 280 Server side storage unit 281 Model storage unit 290, 490 Server side control unit 291 Machine learning processing unit 292 State determination unit 293 Notification processing unit 310 Money handling IF unit 385 Learned model storage unit 400 Model management device 410 Server side IF unit 900 Communication network 910 Portable storage medium

Claims (9)

An acquisition unit for acquiring sensor data relating to the transport of money in the money handling machine;
A storage unit for storing a model that outputs an output value indicating an abnormality degree of the money handling machine as sensor data regarding conveyance of money in the money handling machine as an input value;
A state determination unit that determines the state of the money handling machine using the sensor data acquired by the acquisition unit and the model stored in the storage unit. The storage unit stores the model for each denomination of money;
The state determination device according to claim 1, wherein the state determination unit determines the presence or absence of an abnormality for each denomination of money.   The state according to claim 2, wherein the state determination unit determines the presence or absence of an abnormality of the money processor for all money of a plurality of denominations based on output values from the model for each denomination of money. Judgment device.   When an error occurs in the money handling machine, among the sensor data obtained by measurement by the sensor of the money handling machine where the error occurred, the learning data based on the sensor data in the past first period after the error occurrence is abnormal. Machine learning processing unit for updating the model by performing machine learning using learning data based on sensor data in a second period prior to the first period as learning data in the normal state. The state determination device according to any one of claims 1 to 3, comprising: A notification processing unit that determines the necessity of a notification that promotes maintenance on the money handling machine using the model;
An output unit that outputs the notification when the notification processing unit determines that the notification is necessary;
The state determination device according to any one of claims 1 to 4, comprising: It comprises a state determination unit that determines the presence or absence of an abnormality in the money handling machine using the model;
The machine learning processing unit updates the model when an error occurs in the money handling machine and the state determination unit determines that there is no abnormality in the money handling machine.
The state determination device according to claim 4. An acquisition unit for acquiring sensor data relating to the transport of money in the money handling machine;
A storage unit for storing a model that outputs an output value indicating an abnormality degree of the money handling machine as sensor data regarding conveyance of money in the money handling machine as an input value;
A state determination unit that determines the state of the money handling machine using the sensor data acquired by the acquisition unit and the model stored in the storage unit.
Money processor status judgment system. Acquiring sensor data relating to the transport of money in the money handling machine;
The state of the money handling machine is determined using the obtained sensor data and a model that outputs an output value indicating the degree of abnormality of the money handling machine using sensor data on conveyance of money in the money handling machine as an input value State determination method including the steps of On the computer
Acquiring sensor data relating to the transport of money in the money handling machine;
The state of the money handling machine is determined using the obtained sensor data and a model that outputs an output value indicating the degree of abnormality of the money handling machine using sensor data on conveyance of money in the money handling machine as an input value A program to execute the steps and

Images