Machine learning device, picture prediction device and control device

Abstract

A machine learning device, a screen prediction device, and a control device are provided. The machine learning device comprises: a state observation unit, which obtains a state vector column of a set of state vectors of the operation screen transition due to a series of operations or changes in the machine state, including screen type data of each operation screen used through a series of operations in the operation screen of the display unit of the control device for controlling the industrial machine and machine state data of the machine state when the operation screen is displayed; a training data acquisition unit, which uses a partial state vector column of N consecutive state vectors starting with an arbitrary state vector in the state vector column as input data, and obtains training data using the screen type data of the operation screen after the state vector at the end of the partial state vector column as label data; and a learning unit, which performs supervised machine learning on the training data to generate a learned model for predicting the screen type data to be used by the user next.

Description

Technical Field

[0001] The present invention relates to a machine learning device, a picture prediction device and a control device. Background Art

[0002] As the number of functions added to numerical control devices for controlling machine tools increases, the amount of information displayed on output devices such as LCDs and the types of operation screens also increase. As a result, users need to frequently switch screens, and the increase in switching time has a negative impact on the productivity of on-site users.

[0003] Therefore, when the numerical control device is activated, the screen data of all operation screens used in the screen display process are loaded into a memory such as a RAM (Random Access Memory), thereby achieving faster screen switching.

[0004] However, due to the high functionality of the operation screen itself, the data size of the operation screen increases, and the startup time of the control device increases. As a result, waiting time is generated when the power is turned on and off during machine adjustment, and there is a problem of reduced productivity during machine adjustment. In addition, since the screen data of the operation screen that has not been used once is also loaded, the possibility of insufficient memory for storing screen data increases.

[0005] Therefore, as a method for suppressing the amount of data of the loaded operation screen, a method is known in which, when the operation screen is a multi-window, the screen data of the child window is pre-loaded when the parent window is displayed. Alternatively, a method is known in which, until the scheduled operation is completed, when the next operation is about to start, the screen data of the next operation screen determined by the program is loaded, as in a computer game.

[0006] In addition, a machine learning technique is known that determines the display order of menu items in a menu display based on status data including information indicating a processing status of a processing and information indicating a selected menu item. Alternatively, a machine learning technique is known that detects an operator and learns the display of an operation menu based on the information of the detected operator. For example, refer to Patent Documents 1 and 2.

[0007] Patent Document 1: Japanese Patent No. 6290835

[0008] Patent Document 2: Japanese Patent Application Publication No. 2017-138881 Summary of the invention

[0009] However, the purpose of each operation screen is assumed, for example, it can be distinguished according to each state such as "manual operation", "editing operation", "automatic operation", "alarm generation", etc. However, the state is dynamically switched from "manual operation" to "automatic operation", from "manual operation" or "automatic operation" to "alarm generation", that is, the switching of the operation screen depends on the state of the machine tool. In addition, the switching of the operation screen also depends on the machine design related to display and operation, the operation rules of the machine tools of each factory, etc. Therefore, it is difficult to predetermine the order of displaying the operation screen.

[0010] Therefore, in order to limit the data of the operation screen to be loaded in advance and thus reduce the amount of data of the operation screen to be loaded, it is desirable to predict the operation screen that the user will use through a series of operations.

[0011] (1) One embodiment of the machine learning device of the present invention includes: a state observation unit that obtains a state vector x corresponding to a transition of an operation screen corresponding to a series of operations by a user or a change in a machine state of an industrial machine. n The time series data of the collection is the state vector column {x n :n≥1}, as the state vector x n The screen type data y indicates the screen type of each operation screen used by the user when the user performs a series of operations based on the operation screen displayed on the display unit of the control device for controlling the industrial machine. n and machine state data z indicating the state of the machine controlled by the control device when the operation screen is displayed n ; A training data acquisition unit, which includes the state vector column obtained by the state observation unit, with an arbitrary state vector x n (n≥1) as the first set of N (N≥2) consecutive partial state vector sequences {x n 、x n+1 ,...x n+N-1} As input data, obtain training data, which will include a state vector x at the end of the column representing the partial state vector of the input data n+N-1 Screen type data y of the type of screen to be displayed next to the corresponding operation screen n+N-1 The screen type data of the screen to be displayed next after the operation screen is used as label data; and a learning unit, which uses the input data and the label data as training data to perform supervised machine learning, and generates a learned model for predicting the screen type data that the user will use next based on a partial state vector column that includes N consecutive state vectors traced back from the current state vector to the past, when the user performs a series of operations based on the operation screen displayed on the display unit.

[0012] (2) One embodiment of the picture prediction device of the present invention comprises:

[0013] The state acquisition unit acquires information related to the operation screen corresponding to a series of operations by the user, including the current state vector x n (n≥N) traces back to the past N consecutive state vectors of the partial state vector sequence {x n-N+1 ,...x n}; and a prediction unit, which is based on the learned model generated by the machine learning device of (1), according to the partial state vector sequence {x n-N+1 ,...x n}, predicting screen type data of an operation screen that the user will use next.

[0014] (3) One embodiment of the picture prediction device of (2) includes the machine learning device of (1).

[0015] (4) One embodiment of the control device of the present invention includes the picture prediction device of (2) or (3).

[0016] According to one embodiment, the operation screen used by the user in a series of operations is predicted, and the data of the operation screen to be loaded is limited to the data of the predicted operation screen, thereby reducing the amount of data of the operation screen to be loaded. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a functional block diagram showing a functional configuration example of a numerical control system according to an embodiment.

[0018] Figure 2 This is a diagram showing an example of state vectors included in the state vector sequence.

[0019] Figure 3 Yes Description Figure 1 A diagram of an example of a process for generating a learned model performed by a machine learning device.

[0020] Figure 4 This is a diagram schematically showing the structure of an RNN involved in one embodiment.

[0021] Figure 5 This is a flowchart for explaining the screen prediction processing of the numerical control system in the operation stage.

[0022] Figure 6 is a diagram showing an example of a learned model.

[0023] Figure 7 is a diagram showing an example of a learned model.

[0024] Figure 8 This is a diagram showing an example of a learned model for predicting screen type data of operation screens up to the next three times.

[0025] Fig. 9 This is a diagram showing an example of the configuration of a numerical control system.

[0026] Fig.10 This is a diagram showing an example of the configuration of a numerical control system.

[0027] Description of reference numerals:

[0028] 100-machine tool; 200-control device; 211-screen data loading unit; 300-machine learning device; 311-state vectorization unit; 312-state observation unit; 314-learning unit; 350-learned model; 400-screen prediction device; 411-state acquisition unit; 412-prediction unit. DETAILED DESCRIPTION

[0029] Hereinafter, one embodiment will be described using the drawings.

[0030] <One Implementation Method>

[0031] Figure 1 1 is a functional block diagram showing an example of a functional configuration of a numerical control system according to an embodiment of the present invention. Figure 1 As shown, the numerical control system includes a machine tool 100 , a control device 200 , a machine learning device 300 , and a screen prediction device 400 .

[0032] The machine tool 100, the control device 200, the machine learning device 300, and the screen prediction device 400 may be directly connected to each other via a connection interface not shown. In addition, the machine tool 100, the control device 200, the machine learning device 300, and the screen prediction device 400 may be connected to each other via a network not shown, such as a LAN (Local Area Network) or the Internet.

[0033] The machine tool 100 is a machine for machining that is operated by numerical control of a control device 200. The machine tool 100 feeds back information indicating an operating state based on an operating command of the control device 200 to the control device 200.

[0034] The present embodiment is not limited to the machine tool 100, but is widely applicable to all industrial machines. Industrial machines include various machines such as machine tools, industrial robots, service robots, forging machines, and injection molding machines.

[0035] The control device 200 is a numerical control device well known to those skilled in the art, and controls the operation of the machine tool 100. In addition, when the machine tool 100 is a robot or the like, the control device 200 may be a robot control device or the like.

[0036] In the following description, the state of the machine tool 100 controlled by the control device 200 is simply referred to as a “machine state”.

[0037] The screen prediction device 400 is a computer device or the like. The screen prediction device 400 obtains a partial state vector sequence, which includes N (N≥2) consecutive state vectors that are preset and traced back from the current state vector to the past corresponding to the transition of the operation screen corresponding to a series of user operations or machine state changes. The screen prediction device 400 inputs the obtained partial state vector sequence into the learned model generated by the machine learning device 300 described later, thereby predicting the screen type data of the operation screen that the user will use next. In addition, the screen prediction device 400 outputs the prediction result to the control device 200.

[0038] Thus, the control device 200 only needs to load the screen data of the operation screen indicated by the next screen type data predicted by the screen prediction device 400, and the amount of data of the operation screen to be loaded in advance can be reduced.

[0039] Before describing the picture prediction device 400 , machine learning for generating a learned model will be described.

[0040] <Control Device 200 in Learning Phase>

[0041] In the learning phase, the control device 200 collects the supervisory data required for machine learning. Therefore, when the user performs a series of operations based on the operation screen displayed on the display unit 212, the control device 200 obtains data including screen type data indicating the screen type of each operation screen used by the user and machine state data indicating the machine state when the operation screen is displayed as a state vector. The control device 200 stores the obtained state vector and outputs the obtained state vector to the machine learning device 300 via a communication unit (not shown).

[0042] The control device 200 in the learning phase includes a display unit 212 and a storage unit 213 .

[0043] The display unit 212 is, for example, a liquid crystal display, and displays an operation screen in response to a user operation.

[0044] The storage unit 213 is, for example, an SSD (Solid State Drive), an HDD (Hard Disk Drive), or the like.

[0045] The storage unit 213 has a data area 220 for storing screen type data indicating the screen type of each operation screen used by the user and machine status data indicating the machine status when the operation screen is displayed. In addition, the storage unit 213 stores screen data 230(1)-230(M) of M operation screens (M is an integer greater than or equal to 2).

[0046] <Machine Learning Device 300>

[0047] The machine learning device 300, for example, obtains the screen type data and machine status data stored in the storage unit 213 of the control device 200 via a communication unit (not shown). The machine learning device 300 may, for example, represent the obtained screen type data and machine status data as one-hot as described later, and vectorize them into a combined state vector. The machine learning device 300 may obtain, from the vectorized state vectors, a set of state vectors corresponding to the transition of the operation screen corresponding to a series of user operations or changes in the machine status as time series data, i.e., a state vector column.

[0048] In addition, the machine learning device 300 may obtain N (N≥2) consecutive partial state vector sequences that are included in the state vector sequence and start with an arbitrary state vector as input data. The machine learning device 300 may obtain screen type data of the operation screen that is displayed next after the operation screen that corresponds to the state vector at the end of the partial state vector sequence of the acquired input data as label data.

[0049] The machine learning device 300 uses the acquired input data and label data as training data to perform supervised machine learning. When the user performs a series of operations based on the operation screen displayed on the display unit 212, a learned model can be generated to predict the type of screen data that the user will use next based on a partial state vector column that includes N consecutive state vectors traced back from the current state vector.

[0050] In this way, the machine learning device 300 can provide the learned model 350 to the picture prediction device 400 .

[0051] The machine learning device 300 will be described in detail.

[0052] In addition, unless otherwise specified, the following description will be made of a case where "N" is set to "3" and the screen type data to be used by the user "next" is predicted. "N" may be a value greater than 2 or 4. In addition, the screen prediction device 400 may predict the screen type data to be used by the user next (for example, the second or third time including next).

[0053] In addition, the learning process of the machine learning device 300 can be executed by a background task or a subprocessor when a series of user operations or a change in the machine state occur.

[0054] like Figure 1 As shown, the machine learning device 300 includes a state vectorization unit 311 , a state observation unit 312 , a training data acquisition unit 313 , a learning unit 314 , and a storage unit 315 .

[0055] In the learning phase, the state vectorization unit 311 obtains the screen type data y from the control device 200 via a communication unit (not shown). n and machine status data n The state vectorization unit 311, as described later, converts the acquired picture type data y n and machine status data n Vectorized into a state vector x combined with one-hot representation n .

[0056] The state observation unit 312 obtains the vectorized state vector x n The state vector x corresponding to the transition of the operation screen corresponding to a series of user operations or changes in the machine state n The collection is the time series data, that is, the state vector column {x n :n≥1}. The state observation unit 312 outputs the acquired state vector sequence to the training data acquisition unit 313 and the storage unit 315.

[0057] Figure 2 is the state vector x included in the state vector column n An example of a graph. Figure 2 As shown, the state vectors x1, x2, x3, etc. have screen type data y of the operation screen used by the user at time t1, t2, t3, etc. according to the operation performed by the user or the change of the machine state. n , and machine status data n Also, the machine state data z of the state vectors x1, x2, x3, etc. n It has the components of the mode and alarm described later.

[0058] Here, in the screen type data y nIn the case of "manual operation", there are various types of operation screens according to each state such as "manual operation", "editing operation", "automatic operation", and "alarm generation". For example, in the case of "manual operation", there are operation screens such as "MDI program" for inputting a temporary short NC program, "measurement" for measuring the size of the workpiece and tool, "tool compensation" for setting the distance from the tool center point to the tool tip, and "workpiece origin compensation" for setting the distance from the origin of the machine to the origin of the workpiece. In addition, in the case of "editing operation", there are operation screens such as "program editing" for editing the NC program for automatic operation, "cutting simulation" for executing the NC program while keeping the machine stopped and drawing the tool trajectory and workpiece shape as a graph. In addition, in the case of "automatic operation", there are operation screens such as "tool position display" for displaying the coordinates of the tool position, "program check" for displaying the content of the NC program being executed, "processing drawing" for drawing the tool trajectory and workpiece shape as a graph for the NC program being executed, and "load meter" for displaying the percentage of the cutting load involved in the motor. In addition, in the case of "alarm", there are operation screens such as "diagnosis" that displays the internal status of the CNC, "parameter" for inputting various parameters of the CNC, "alarm message" that displays alarm messages, and "alarm history" that displays alarm history.

[0059] In addition, in the machine status data z n The "modes" include: editing and executing MDI programs, "MDI" for manual data input, "EDIT" for making the NC program for automatic operation editable, "MEM (AUTO)" for automatic operation of the NC program, "HANDLE" for axis feed by handle operation, "JOG" for axis feed by button operation, "REFERENCE" for restoring the origin, and "REMOTE (TAPE)" for executing NC programs on external devices.

[0060] In addition, in the machine status data z n The "alarms" include: NC program errors, namely "P / S alarms", axis movement exceeding the limit value, "overtravel alarms", indicating overheating of the control device or motor, servo motor-related alarms, namely "servo alarms", spindle motor-related alarms, namely "spindle motor alarms", and alarms defined by equipment manufacturers, namely "external alarms".

[0061] In addition, machine status data n The "alarm" is an example of a situation where the operation screen displayed on the display unit 212 changes when the machine state changes. In addition, the "alarm" may include types of pop-up displays for notifying regular maintenance and software update notifications, for example.

[0062] In addition, as described later, in the machine status data z n It is also possible to have operating status or signals in addition to modes and alarms.

[0063] like Figure 2 As shown, for example, when the operation screen of the display unit 212 is changed from the screen of "tool position display" to the screen of "program check" by the user's operation at time t1 in the machine state where the mode is "MDI" and there is no alarm, the screen type data y1 of the state vector x1 is represented by one-hot, the component of the screen of "tool position display" (for example, the first component) is "1", and the components of other operation screens are "0". In addition, the "mode" of the machine state data z1 of the state vector x1 is represented by one-hot, the component of "MDI" (for example, the third component) is "1", and the components of other modes are "0". In addition, the "alarm" of the machine state data z1 of the state vector x1 is represented by one-hot, the component of "no alarm" (for example, the last component) is "1", and the components of other alarms are "0".

[0064] At the next time t2, when the operation screen of the display unit 212 is changed from the "Program Check" screen to the "Load Scale" screen by the user's operation while the mode is kept at "MDI" and no alarm is present, the screen type data y2 of the state vector x2 is expressed by one-hot, the component (e.g., the second component) of the "Program Check" screen is "1", and the components of the other operation screens are "0". Furthermore, at time t3, when the operation screen of the display unit 212 is changed from the "Load Scale" screen to the "Drawing During Processing" screen by the user's operation while the mode is kept at "MDI" and no alarm is present, the screen type data y3 of the state vector x3 is expressed by one-hot, the component (e.g., the third component) of the "Load Scale" screen is "1", and the components of the other operation screens are "0". In addition, the patterns of the machine state data z2 and z3 of the state vectors x2 and x3 and the components of the alarm are the same as the machine state data z1 of the state vector x1.

[0065] The training data acquisition unit 313 acquires the state vector sequence included in the acquired state vector sequence, and the state vector x n (n≥1) is the first three consecutive partial state vector sequences {x n 、x n+1 、x n+2} is obtained as input data.

[0066] In addition, the training data acquisition unit 313 obtains the state vector x at the end of the partial state vector sequence corresponding to the input data. n+2Screen type data y of the type of the operation screen to be displayed next to the corresponding operation screen n+2 The training data acquisition unit 313 outputs the acquired input data and label data to the storage unit 315 .

[0067] The learning unit 314 uses the input data and label data stored in the storage unit 315 as training data to perform supervised machine learning. When the user performs a series of operations based on the operation screen displayed on the display unit 212, the learning unit 314 generates a learned model 350 for predicting the type of screen data that the user will use next based on a partial state vector column that includes three consecutive state vectors traced back from the current state vector.

[0068] Figure 3 Yes Description Figure 1 A diagram of an example of a process of generating a learned model 350 performed by a machine learning device 300.

[0069] like Figure 3 As shown, the learned model 350, for example, includes a partial state vector sequence {xp 0} that includes three consecutive state vectors traced back from the state vector xp0 corresponding to the current time t0. -2 、xp -1 , xp0} is set as the input layer, the RNN (Recurrent Neural Network) 10(1)-10(3) of each vector input of the input layer is set as the intermediate layer, and the screen type data yp0 used by the user next is set as the output layer. In addition, the state vector xp -1 Indicates the screen type data and machine status data of the most recent operation screen transition. -2 Representation and state vector xp -1 The screen type data and machine status data at the time of transition from the previous operation screen are compared.

[0070] In addition, below, when there is no need to distinguish RNN10(1)-10(3) individually, they are collectively referred to as "RNN10".

[0071] Figure 4 Schematically shows the structure of RNN10 involved in this embodiment. Figure 4 As shown, the output of each perceptron in the middle layer is input to its own perceptron and the perceptron in the same layer.

[0072] The learning unit 314 receives the partial state vector sequence {x1, x2, x3}, {x2, x3, x4} ... {x N-2 、xN-1 、x N} input data, screen type data y3, y4...y N The learning unit 314 replaces the partial state vector column with {x p-2 、x p-1 、x p0}, and replace the screen type data with y p0 Using the received training data, for example, supervised learning based on the well-known error back propagation method is performed. Through this supervised learning, the learning unit 314 optimizes parameters such as the weight vector in the RNN 10 to generate a learned model 350.

[0073] In addition, for the above-mentioned supervised learning, the learning unit 314 can perform it through online learning, batch learning, or small batch learning.

[0074] Online learning is a learning method that performs supervised learning immediately each time supervised data is created. In addition, batch learning is a learning method that collects multiple supervised data corresponding to the repeated creation of supervised data and uses all the collected supervised data to perform supervised learning. Furthermore, mini-batch learning is a learning method that performs supervised learning each time a certain amount of supervised data between online learning and batch learning is accumulated.

[0075] Then, the learning unit 314 provides the optimized learned model 350 to the picture prediction device 400 .

[0076] In addition, when the learning unit 314 obtains new supervisory data after generating the learned model 350 , it can update the generated learned model 350 by further performing supervised learning on the learned model 350 .

[0077] The storage unit 315 is an SSD or HDD, etc., and stores the state vector sequence acquired by the state observation unit 312, the input data and label data acquired by the training data acquisition unit 313, the learned model 350 generated by the learning unit 314, etc.

[0078] The machine learning for generating the learned model 350 included in the picture prediction device 400 has been described above.

[0079] Next, the picture prediction device 400 will be described.

[0080] <Screen Prediction Device 400>

[0081] like Figure 1 As shown, the picture prediction device 400 is configured to include a state acquisition unit 411 , a prediction unit 412 , and a storage unit 413 .

[0082] The state acquisition unit 411 acquires the screen type data y from the control device 200 during the operation phase. n and machine status data n The state acquisition unit 411 acquires the screen type data y n and machine status data n Represented as one-hot and vectorized into a combined state vector x n The state acquisition unit 411 acquires the current state vector x related to the operation screen corresponding to a series of operations by the user. n The partial state vector sequence {x n-2 、x n-1 、x n} as a partial state vector column {xp -2 、xp -1 , xp0}.

[0083] The prediction unit 412 calculates the partial state vector sequence {xp -2 、xp -1 , xp0}, and predicts the screen type data yp0 of the operation screen to be used next by the user. The prediction unit 412 outputs the prediction result to the control device 200.

[0084] The storage unit 413 is a RAM or the like, and stores the learned model 350 generated by the machine learning device 300 .

[0085] <Control Device 200 in Operation Phase>

[0086] like Figure 1 As shown, the control device 200 in the operation phase is configured to include a screen data loading unit 211, a display unit 212, and a storage unit 213. Here, the display unit 212 and the storage unit 213 are as described in the learning phase.

[0087] The screen data loading unit 211 acquires the screen type data yp0 of the operation screen predicted by the screen prediction device 400 to be used next by the user via a communication unit (not shown). The screen data loading unit 211 loads the screen data 230 of the operation screen corresponding to the acquired screen type data yp0 from the storage unit 213 to a RAM (not shown) or the like.

[0088] In addition, after the screen data 230 of the operation screen of the predicted screen type data yp0 is loaded into the screen data loading unit 211, if the user selects another operation screen, the screen data 230 of the other selected operation screen can be loaded from the storage unit 213, and the screen data 230 of the operation screen of the predicted screen type data yp0 can be unloaded from the RAM.

[0089] <Screen prediction processing of numerical control systems in the operation stage>

[0090] Next, the operation related to the screen prediction processing of the numerical control system according to the present embodiment will be described.

[0091] Furthermore, the prediction process of the picture prediction device 400 and the loading process of the control device 200 may be executed by a background task or a subprocessor when a series of operations by the user or when the machine status changes.

[0092] Figure 5 This is a flowchart for explaining the screen prediction processing of the numerical control system in the operation stage.

[0093] In step S11, the state acquisition unit 411 of the picture prediction device 400 obtains the picture type data y from the control device 200. n and machine status data n , get the current state vector x n The partial state vector sequence {x n-2 、x n-1 、x n} as a partial state vector column {xp -2 、xp -1 , xp0}.

[0094] In step S12, the prediction unit 412 of the picture prediction device 400 converts the partial state vector sequence {xp -2 、xp -1 , xp0} is input to the learned model 350, and the screen type data yp0 of the operation screen to be used next by the user is predicted.

[0095] In step S13 , the screen data loading unit 211 of the control device 200 loads the screen data 230 of the operation screen corresponding to the screen type data yp0 predicted in step S12 from the storage unit 213 into a RAM (not shown) or the like.

[0096] According to the above, the screen prediction device 400 involved in one embodiment inputs a partial state vector column of three consecutive state vectors traced back from the current state vector to the past related to the operation screen corresponding to a series of operations of the user into the learned model 350, so as to at least predict the operation screen that the user will use next based on the flow of a series of operations of the user. As a result, the screen prediction device 400 can suppress the amount of data of the operation screen loaded by the control device 200 to a minimum. In addition, the screen prediction device 400 can predict the optimal operation screen corresponding to the machine state, the machine design related to display / operation, the operation rules of the machine tool 100 of each factory, etc.

[0097] Although one embodiment has been described above, the control device 200 , the machine learning device 300 , and the picture prediction device 400 are not limited to the above-described embodiment, and include modifications and improvements within a range that can achieve the purpose.

[0098] <Modification 1>

[0099] In the above-described embodiment, the machine learning device 300 is exemplified as a device different from the control device 200 and the picture prediction device 400 . However, the control device 200 or the picture prediction device 400 may have some or all of the functions of the machine learning device 300 .

[0100] <Modification 2>

[0101] In the above-described embodiment, the picture prediction device 400 is exemplified as a device different from the control device 200 , but the control device 200 may include a part or all of the functions of the picture prediction device 400 .

[0102] Alternatively, for example, a server may include part or all of the state acquisition unit 411, prediction unit 412, and storage unit 413 of the screen prediction device 400. In addition, each function of the screen prediction device 400 may be realized on the cloud using a virtual server function or the like.

[0103] Furthermore, the picture prediction device 400 may be implemented as a distributed processing system in which the functions of the picture prediction device 400 are distributed to an appropriate number of servers.

[0104] <Variation 3>

[0105] In addition, for example, in the above-mentioned embodiment, Figure 3 The learned model 350 is composed of three RNNs 10 in the middle layer, but is not limited to this. Figure 6As shown, the learned model 350 uses RNN10A(1)-10A(3k), has an intermediate layer configured with multiple layers of three 1 groups as one layer, and is deep-learned (k is an integer greater than 2). Figure 3 Compared with the case of , it can more accurately capture the characteristics of the screen type data and machine state data of the past state vector, and can more accurately predict the screen type data of the next operation screen. In addition, RNN10A(1)-10A(3k) are respectively Figure 3 The RNN10 corresponds to .

[0106] Or, if Figure 7 As shown, instead of Figure 3 The learned model 350 can configure the RNN with input gate / output gate / forget gate, that is, LSTM (Long Short-Term Memory) 20(1)-20(3) in the middle layer. In this case, LSTM20(1)-20(3), a full combination layer 30 and argmax40 are configured in the middle layer of the learned model 350. Argmax40 outputs the picture type data with the highest probability value among the multiple picture type data output from the full combination layer 30 as the picture type data yp0. Thus, Figure 3 Compared with the case of the previous state vector, the screen prediction device 400 can more accurately capture the characteristics of the screen type data and machine state data of the past state vector, and can more accurately predict the screen type data of the subsequent operation screen. Figure 7 In the example, the number of LSTM20 can be set to three, but it can also be more than one.

[0107] <Variation 4>

[0108] Furthermore, for example, in the above-described embodiment, the screen prediction device 400 predicts the operation screen that the user will use next, but it is also possible to predict the operation screen that the user will use next thereafter.

[0109] Figure 8 3 is a diagram showing an example of a learned model 350 for predicting the screen type data yp0, yp1, and yp2 of the operation screen until the third time after the next one. Figure 8 As shown, the learned model 350 sets the partial state vector sequence including three consecutive state vectors traced back from the current state vector xp0 to the past as {xp -2 、xp -1, xp0} input layer, RNN10B(1)-10B(5) are set as the intermediate layer, and the screen type data yp0, yp1, yp2 used by the user for the third time after that are set as the output layer.

[0110] In this case, the machine learning device 300 receives the partial state vector sequence {x1, x2, x3}, {x2, x3, x4} . . . {x N-2 、x N-1 、x N} input data, screen type data {y3, y4, y5}, {y4, y5, y6}...{y N-2 ,y N-1 ,y N}’s labeled data is used as training data for supervised learning.

[0111] In addition, RNN10B(1)-10B(5) are respectively Figure 3 The RNN10 corresponds to .

[0112] <Variation 5>

[0113] In addition, for example, in the above-mentioned embodiment, the screen prediction device 400 uses the learned model 350 generated by the machine learning device 300 to predict the operation screen that the user will use next, but the present invention is not limited to this. Fig. 9 As shown, the server 500 may store the learned model 350 generated by the machine learning device 300, and share the learned model 350 with m picture prediction devices 400A(1)-400A(m) connected to the network 50 (m is an integer greater than or equal to 2). Thus, even if a new machine tool, control device, and picture prediction device are configured, the learned model 350 can be applied.

[0114] Furthermore, the picture prediction devices 400A(1)-400A(m) are connected to the control devices 200A(1)-200A(m), respectively, and the control devices 200A(1)-200A(m) are connected to the machine tools 100A(1)-100A(m), respectively.

[0115] In addition, the machine tools 100A(1)-100A(m) are respectively Figure 1 The control devices 200A(1)-200A(m) correspond to the machine tool 100. Figure 1 The picture prediction devices 400A(1)-400A(m) correspond to the control device 200. Figure 1 The picture prediction device 400 corresponds to.

[0116] Or, if Fig.10As shown, the server 500 operates as the screen prediction device 400, for example, and can predict the screen type data of the operation screen to be used next by the user for each of the control devices 200A(1)-200A(m) in the same factory connected to the network 50. Thus, even if a new machine tool and control device are installed, the learned model 350 can be applied.

[0117] <Variation 6>

[0118] In addition, for example, in the above-mentioned embodiment, the machine status data z n It has modes and alarms, but as mentioned above, it can also have modes and alarms, operating states, and signals.

[0119] In addition, in the machine status data z n The “operating status” includes, for example, “START (start status)” indicating the state in which the NC program is being executed, “STOP (stop status)” indicating the state in which the operation is ended and stopped, “HOLD (rest status)” indicating the state in which the operation is interrupted and stopped although there are still instructions to be executed, and “RESET (reset status)” indicating the initial state, or the state in which an emergency stop or the reset button is pressed.

[0120] In addition, in the machine status data z n The "signals" include "jog feed axis direction selection" which determines which axis / direction (±) of X, Y, and Z to move, "handle feed axis selection" which determines which axis / direction (±) of X, Y, and Z to move, and "single block" which stops the NC program after executing one line.

[0121] In addition, each function included in the control device 200, the machine learning device 300, and the picture prediction device 400 of one embodiment can be implemented by hardware, software, or a combination thereof. Here, implementation by software means that a computer reads a program and executes it.

[0122] Each component included in the control device 200 can be implemented by hardware (including electronic circuits, etc.), software or a combination thereof. In the case of implementation by software, the program constituting the software is installed in the computer (numerical control device 1). In addition, these programs can be recorded on a removable medium and distributed to users, and can be downloaded to the user's computer via a network for distribution. In addition, in the case of hardware configuration, for example, a part or all of the functions of each component contained in the above-mentioned control device can be constituted by an integrated circuit (IC) such as ASIC (Application Specific Integrated Circuit), gate array, FPGA (Field Programmable Gate Array), CPLD (Complex Programmable Logic Device), etc.

[0123] The program can be stored using various types of non-transitory computer readable media (Non-transitorycomputer readable medium) and provided to the computer. Non-transitory computer readable media include various types of tangible recording media (Tangible storage medium). Examples of non-transitory computer readable media include: magnetic recording media (e.g., floppy disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROM (Read Only Memory), CD-R, CD-R / W, semiconductor memory (e.g., mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM. In addition, the program can be supplied to the computer via various types of transitory computer readable media (Transitorycomputer readable medium). Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer readable medium supplies the program to the computer via wired communication paths such as electric wires and optical fibers or wireless communication paths.

[0124] Furthermore, the steps describing the program recorded in the recording medium include processing that is performed in time series in the order in which it is performed, processing that is not necessarily performed in time series, and processing that is performed in parallel or individually.

[0125] In other words, the machine learning device, picture prediction device, and control device of the present invention can take various embodiments having the following configurations.

[0126] (1) The machine learning device 300 of the present invention includes a state observation unit 312 that acquires a state vector x corresponding to a transition of an operation screen corresponding to a series of user operations or a change in the state of the device. n The time series data of the collection is the state vector column {x n :n≥1}, as the state vector x n The screen type data y indicates the screen type of each operation screen used by the user when the user performs a series of operations based on the operation screen displayed on the display unit 212 of the control device 200 that controls the industrial machine (machine tool 100). n and machine state data z representing the machine state of the industrial machine controlled by the control device 200 when the operation screen is displayed. n ; The training data acquisition unit 313, which includes the state vector column acquired by the state observation unit 312, with an arbitrary state vector x n (n≥1) as the first set of N (N≥2) consecutive partial state vector sequences {x n 、x n+1 ,...x n+N-1} As input data, obtain training data, which will include a state vector x at the end of the column representing the partial state vector of the input data n+N-1 Screen type data y of the type of screen to be displayed next to the corresponding operation screen n+N-1 The screen type data of the screen to be displayed next after the operation screen is used as label data; the learning unit 314 uses the input data and the label data as training data to perform supervised machine learning, and when the user performs a series of operations based on the operation screen displayed on the display unit 212, based on the partial state vector column including N consecutive state vectors traced back from the current state vector to the past, generates a learned model 350 for predicting the screen type data that the user will use next.

[0127] According to the machine learning device 300, the operation screen that the user will use next is predicted, and the data of the operation screen of the object loaded by the control device 200 is limited to the data of the predicted operation screen, thereby generating a learned model 350 for suppressing the amount of data of the loaded screen to a minimum.

[0128] (2) The learning unit 314 may perform machine learning on the flow of the series of operations using a recurrent neural network (RNN10).

[0129] In this way, the machine learning device 300 can generate a learned model 350 that can predict the screen type data of the operation screen that the user will use next based on the flow of a series of operations. (3) As a recurrent neural network, an LSTM 20, which is an RNN with input / output / forget gates, can be used.

[0130] In this way, the machine learning device 300 can generate the learned model 350 that can more accurately predict the screen type data of the operation screen that the user will use next.

[0131] (4) The screen prediction device 400 of the present invention includes a state acquisition unit 411 that acquires information related to the operation screen corresponding to a series of operations by the user, including the current state vector x n (n≥N) traces back to the past N consecutive state vectors of the partial state vector sequence {x n-N+1 ,...x n}; a prediction unit 412, which is based on the learned model 350 generated by the machine learning device 300, according to the partial state vector sequence {x n-N+1 ,...x n}, predict the screen type data of the operation screen that the user will use next.

[0132] According to the screen prediction device 400, the operation screen to be used next by the user is predicted, and the data of the operation screen to be loaded by the control device 200 is limited to the data of the predicted operation screen, thereby suppressing the amount of loaded screen data to a minimum.

[0133] (5) The picture prediction device 400 of (4) may further include a learned model 350 .

[0134] In this way, the same effect as the above-mentioned (4) can be achieved.

[0135] (6) The picture prediction device 400 may further include the machine learning device 300 .

[0136] In this way, the same effect as any one of the above (1) to (5) can be achieved.

[0137] (7) The picture prediction device 400 of (4) may further include a server 500 connected so as to be able to access the learned model 350 from the picture prediction device 400 via the network 50 .

[0138] In this way, even if a new machine tool 100 , control device 200 , and screen prediction device 400 are installed, the learned model 350 can be applied.

[0139] (8) The control device 200 of the present invention includes the screen data loading unit 211 that loads screen data corresponding to the screen type data of the operation screen that the user will use next, predicted by the screen prediction device 400.

[0140] According to control device 200, by limiting the data of the operation screen to be loaded to the data of the predicted operation screen, the amount of data of the loaded screen can be suppressed to a minimum.

[0141] (9) The control device 200 of (8) may further include a picture prediction device 400.

[0142] In this way, the same effect as above (8) can be achieved.

Claims

1. A machine learning device, It is characterized in that The machine learning device has: The state observation unit acquires a state vector x corresponding to a transition of the operation screen corresponding to a series of operations by the user or a change in the machine state of the industrial machine. n The time series data of the collection is the state vector column {x n :n≥1}, the state vector x n The screen type data y indicates the screen type of each operation screen used by the user when the user performs a series of operations based on the operation screen displayed on the display unit of the control device for controlling the industrial machine. n and machine state data z representing the machine state controlled by the control device when the operation screen is displayed n ; A training data acquisition unit is configured to acquire a training data including the state vector sequence acquired by the state observation unit and the state vector sequence having an arbitrary state vector x n As the beginning of the preset N consecutive partial state vector sequence {x n 、x n+1 ,...x n+N-1 } As input data, obtain training data, which will include a state vector x at the end of the column representing the partial state vector of the input data n+N-1 Screen type data y of the type of screen to be displayed next to the corresponding operation screen n+N-1 The screen type data of the screen displayed next to the operation screen is used as label data, wherein n≥1, N≥2; and A learning unit, which uses the input data and label data as training data to perform supervised machine learning, and generates a learned model for predicting the type of screen data that the user will use next based on a partial state vector column that includes N consecutive state vectors tracing back from the current state vector to the past when the user performs a series of operations based on the operation screen displayed on the display unit.

2. The machine learning device according to claim 1, It is characterized in that The learning unit performs machine learning on the flow of the series of operations using a recurrent neural network.

3. The machine learning device according to claim 2, It is characterized in that As the recurrent neural network, an RNN with an input gate, an output gate, and a forget gate, namely, LSTM, is suitable.

4. A picture prediction device, It is characterized in that The picture prediction device comprises: The state acquisition unit acquires information related to the operation screen corresponding to a series of operations by the user, including the current state vector x n The partial state vector sequence {x n-N+1 ,...x n }, where n ≥ N; and A prediction unit, which is based on the learned model generated by the machine learning device according to any one of claims 1 to 3, according to the partial state vector sequence {x n-N+1 ,...x n }, predicting screen type data of an operation screen that the user will use next.

5. The picture prediction device according to claim 4, It is characterized in that The picture prediction device includes the learned model.

6. The picture prediction device according to claim 5, It is characterized in that The picture prediction device includes the machine learning device according to any one of claims 1 to 3.

7. The picture prediction device according to claim 4, It is characterized in that The learned model is provided in a server connected so as to be accessible from the picture prediction device via a network.

8. A control device, It is characterized in that The control device includes a screen data loading unit that loads screen data corresponding to screen type data of an operation screen to be used next by the user, predicted by the screen prediction device according to any one of claims 4 to 7.

9. The control device according to claim 8, It is characterized in that The control device includes the picture prediction device according to any one of claims 4 to 7.

Images