Data processing method, device, and machine device
By selectively processing and storing feature quantities from time-series data, mechanical devices with limited resources can accurately predict future states, optimizing resource usage and improving predictive accuracy.
Patent Information
- Application Number
- JP2023074339
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-04-28
- Publication Date
- 2024-11-08
AI Technical Summary
Mechanical devices with microcomputers face challenges in processing large volumes of time-series data due to limited resources, making it difficult to predict future states accurately.
A data processing method that involves storing only the latest data in memory, calculating feature quantities, and inputting them into a trained model to predict future values, while optimizing memory and processor usage by selectively calculating feature amounts based on predetermined conditions.
This approach reduces memory and processor resource consumption while enabling accurate prediction of future states, even with limited resources, by storing only a portion of the time-series data and calculating feature amounts efficiently.
Smart Images

Figure 2024158804000001_ABST
Abstract
Description
[Technical field]
[0001] The subject matter disclosed herein relates to data processing methods, devices and machinery. [Background technology]
[0002] Conventionally, mechanical devices are equipped with many sensors. These sensors output detection values indicating the state of an object at very short time intervals. In other words, the output signal of the sensor is time-series data consisting of a plurality of detection values. For example, Patent Document 1 describes a method of diagnosing the state of a mechanical device using time-series data measured by a sensor. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent No. 6935046 Summary of the Invention [Problem to be solved by the invention]
[0004] By the way, some mechanical devices are equipped with dedicated microcomputers (microcontrollers) near each sensor, in addition to a computer (central processing unit) that controls the entire device. In this case, the output signals from the sensors are processed by the microcomputer, which reduces the computational burden on the central processing unit.
[0005] In addition, by predicting future values using time-series data measured by sensors, it is possible to realize control that responds to future changes in the state of machinery. It is possible to use characteristic values (hereinafter referred to as "feature values") found in a series of time-series data as input variables for predicting such future values.
[0006] However, compared to a computer serving as a central processing unit, a microcomputer has fewer resources such as a processor and a memory unit, and therefore, when the time-series data output from a sensor is large in size, it may be difficult for the microcomputer to process the entire time-series data.
[0007] An object of the present disclosure is to provide a technique that can appropriately predict future values from time-series data even when a device has limited resources. [Means for solving the problem]
[0008] In order to solve the above problems, a first aspect is a data processing method for processing time series data using a device having a processor and a memory, comprising: a) storing the latest data of the time series data in the memory and deleting some of the past data from the memory; b) calculating features of multiple data stored in the memory; and c) inputting the features into a trained model to predict future values.
[0009] A second aspect is a data processing method of the first aspect, wherein the step b) includes: b-1) a step of determining whether or not it is the timing to satisfy a specified condition; and b-2) a step of calculating the feature amount when the step b-1) determines that it is the timing to satisfy the specified condition.
[0010] A third aspect is the data processing method according to the first or second aspect, wherein the time series data includes data indicating a discharge flow rate of a treatment liquid discharged from a nozzle.
[0011] A fourth aspect is the data processing method of the third aspect, in which the feature amount includes an increase rate of the discharge flow rate.
[0012] A fifth aspect is the data processing method according to the third or fourth aspect, wherein the feature amount includes an increase rate of the discharge flow rate with respect to an elapsed time after the discharge flow rate exceeds a first flow rate.
[0013] A sixth aspect is the data processing method according to the first or second aspect, in which the time-series data is data measured by a sensor installed in a device that sends a treatment liquid in a tank to a heater.
[0014] A seventh aspect is a data processing method of the sixth aspect, wherein the time series data includes data indicating a flow rate of the processing liquid sent from the tank to the heater, and the feature includes an integrated value of the flow rate over a predetermined period of time.
[0015] An eighth aspect is a data processing method according to the sixth or seventh aspect, wherein the time series data includes a temperature of the processing liquid in the tank, and the feature amount includes a moving average of the temperature over a predetermined period of time.
[0016] A ninth aspect is a data processing method of the sixth or seventh aspect, wherein the time series data includes data indicating the start and stop of liquid transfer from the tank to the heater, and the feature includes the elapsed time since the liquid transfer started or the elapsed time since the liquid transfer was stopped.
[0017] A tenth aspect is a device for processing time series data, comprising a processor and a memory, wherein the processor executes the steps of: a) storing the latest data of the time series data in the memory and deleting some of the past data from the memory; b) calculating features of multiple data stored in the memory; and c) inputting the features into a trained model to predict future values.
[0018] An eleventh aspect is a mechanical apparatus comprising: a time series data output unit that outputs time series data; a device electrically connected to the time series data output unit, the device having a processor and a memory; and a central processing unit electrically connected to the device, wherein the device executes the steps of: a) storing the latest data of the time series data in the memory while deleting some of the past data from the memory; b) calculating features of multiple data stored in the memory; and c) inputting the features into a trained model to predict future values.
[0019] A twelfth aspect is a mechanical apparatus comprising: a time series data output unit that outputs time series data; a device electrically connected to the time series data output unit, the device having a processor and a memory; and a central processing unit electrically connected to the device, wherein the device executes the steps of: A) storing the latest data of the time series data in the memory while deleting some of the past data from the memory; B) calculating features of multiple data stored in the memory; and C) outputting the features to the central processing unit, and the central processing unit executes the step of: D) inputting the features into a trained model and predicting future values. Effect of the Invention
[0020] According to the data processing method of the first aspect, only a part of the time series data is stored in the memory, not the entire time series data. This allows memory resources to be saved. Also, feature values are calculated based on a plurality of data stored in the memory, not the entire time series data. This allows processor resources to be saved. Therefore, even if the device has limited resources, future values can be appropriately predicted from the time series data.
[0021] According to the data processing method of the second aspect, the feature amount is calculated only when a predetermined condition is satisfied, so that calculation resources can be saved. [Brief description of the drawings]
[0022] [Figure 1] FIG. 1 is a diagram showing a configuration of a mechanical device according to an embodiment. [Diagram 2] 13 is a flowchart showing a processing procedure for time series data. [Diagram 3] 13 is a flowchart showing details of a first feature amount calculation process. [Figure 4] FIG. 10 is a diagram illustrating an example of time-series data. [Diagram 5] FIG. 2 is a diagram conceptually showing time-series data stored in a memory. [Figure 6] 13 is a flowchart showing details of a second feature amount calculation process. [Figure 7] 1 is a diagram showing a configuration of a liquid ejection device, which is an example of a mechanical device. [Figure 8] 5A to 5C are diagrams showing time-series data of a discharge flow rate output from a flow rate sensor and a change over time in a valve opening degree output from a valve controller. [Figure 9] FIG. 1 is a diagram showing a configuration of a printing device, which is another example of a mechanical device. [Figure 10] 11A and 11B are diagrams showing time-series data of the outlet temperature output from a temperature sensor and a liquid transfer on / off signal output from a pump control unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0023] Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings. Note that the components described in the embodiment are merely examples, and are not intended to limit the scope of the present invention. In the drawings, the dimensions and numbers of each part may be exaggerated or simplified as necessary for ease of understanding.
[0024] <1. Configuration of the machine> 1 is a diagram showing the configuration of a machine 1 according to an embodiment. The machine 1 is, for example, a substrate processing apparatus that processes substrates such as semiconductor wafers, substrates for flat panel displays, or printed circuit boards. Note that the machine 1 is not limited to substrate processing apparatuses, and may be, for example, a printing device. The machine 1 includes a plurality of sensors 10, a plurality of microcomputers (microcontrollers) 20, and a central processing unit 30.
[0025] The sensor 10 is a measuring instrument that measures a physical quantity that indicates the state of an object in the mechanical device 1. The "physical quantity that indicates the state of an object" is, for example, the position, speed, acceleration, temperature, pressure, or flow rate of the object. The sensor 10 detects the physical quantity at very short time intervals. This results in time series data D1 that indicates the state of the object. The time series data D1 is a plurality of measurement values (data) arranged at very short time intervals. The sensor 10 outputs the obtained time series data D1 to the microcomputer 20. The sensor 10 corresponds to a "time series data output unit." Note that the time series data output unit is not limited to a measuring instrument such as the sensor 10, but also includes a unit that can output a series of data that changes over time.
[0026] The microcomputer 20 is a device that processes the time-series data D1 output from the sensor 10. The microcomputer 20 is electrically connected to the sensor 10. The microcomputer 20 includes a processor 21 and a memory 22. The processor 21 is, for example, a central processing unit (CPU). The memory 22 is a computer-readable storage medium that stores various programs and various data. The memory 22 is a non-volatile memory or a volatile memory.
[0027] The processor 21 processes the time series data D1 according to a program stored in the memory 22. Specifically, the processor 21 processes the time series data D1 to calculate the feature amount of the time series data D1. The microcomputer 20 then predicts a future value using a trained model that uses the feature amount as an input. The future value is a predicted value of a physical quantity related to a physical quantity measured by the sensor 10. The future value may be the same physical quantity as the physical quantity measured by the sensor 10, or may be different from the physical quantity measured by the sensor 10. As the machine learning algorithm, for example, a random forest or a multilayer perceptron is used, but is not limited to these. In addition, the trained model may be built into the microcomputer 20 after being created by machine learning using a computer different from the microcomputer 20, or may be created by learning using a data set prepared in advance on the microcomputer 20. In addition, the feature amount newly acquired by the microcomputer 20 may be used for incremental learning of the machine learning model.
[0028] Furthermore, the microcomputer 20 appropriately transmits the predicted future value D2 to the central processing unit 30. For example, the microcomputer 20 may be configured to output the predicted future value to the central processing unit 30 when the future value is determined to be an abnormal value based on the comparison with a predetermined threshold value.
[0029] The central processing unit 30 is a computer that performs overall control of the entire machine 1. The central processing unit 30 is electrically connected to a plurality of microcomputers 20. The central processing unit 30 has a processor 31 and a memory 32. The processor 31 is configured by, for example, a CPU, a GPU (Graphics Processing Unit), an MCU (Micro-Control Unit), an MPU (Micro Processing Unit), an FPGA (Field Programmable Gate Array), a PLC (Programmable Logic Controller), or a system LSI (Large-scale Integrated Circuit).
[0030] In the example shown in FIG. 1, a plurality of sensors 10 are connected to one microcomputer 20, but only one sensor 10 may be connected to one microcomputer 20.
[0031] The resources of the microcomputer 20 are smaller than those of the central processing unit 30. Specifically, the processor 21 of the microcomputer 20 has a lower processing capacity than the processor 31 of the central processing unit 30. In addition, the memory 22 of the microcomputer 20 has a smaller storage capacity than the memory 32 of the central processing unit 30. It is difficult for the microcomputer 20 to hold a large amount of time-series data D1 output from the sensor 10. For this reason, the microcomputer 20 processes the time-series data D1 while saving resources.
[0032] <2. Processing of time series data in microcomputers> 2 is a flowchart showing a procedure for processing the time series data D1. Every time the microcomputer 20 receives the latest data v(n) from the sensor 10, the microcomputer 20 executes the process shown in FIG.
[0033] 2, the microcomputer 20 acquires the latest data v(n) from the sensor (step S1). Then, the microcomputer 20 executes a first feature amount calculation process S2 and a second feature amount calculation process S3 in parallel for the acquired data v(n). The first feature amount calculation process S2 is a process for acquiring a first feature amount F1, and the second feature amount calculation process S3 is a process for acquiring a second feature amount F2.
[0034] The microcomputer 20 predicts a future value, which is a predetermined physical quantity, by using the first and second characteristic amounts acquired by the first and second characteristic amount calculation processes S2, S3 (step S4).
[0035] <2.1 First feature calculation process> 3 is a flowchart showing details of the first feature amount calculation process S2. In the first feature amount calculation process S2, the microcomputer 20 stores the data acquired in step S1 in the memory 22 (step S11).
[0036] FIG. 4 is a diagram showing an example of time series data D1. As shown in FIG. 3, the time series data D1 is composed of a plurality of data v(m), v(m+1), v(m+2), . . . , v(n-2), v(n-1), and v(n) arranged at very short time intervals. Both m and n are natural numbers, and n>m. In the following description, the time at which each data v(m) to v(n) is obtained is referred to as t m , t m+1 , t n It is written as follows:
[0037] Fig. 5 is a diagram conceptually illustrating time-series data D1 stored in memory 22. In the example illustrated in Fig. 5, memory 22 stores a plurality (=n-m+1) of data v(m), v(m+1), ... v(n-1), v(n).
[0038] In step S11, the microcomputer 20 deletes the oldest data v(m-1) from the memory 22 in synchronization with storing the latest data v(n) in the memory 22. As a result, in the first feature amount calculation process S2, only a predetermined number of data v(m) to v(n) are stored in the memory 22 at any one time.
[0039] After the latest data v(n) is stored in the memory 22, the microcomputer 20 calculates a first feature amount F1(n) indicating the features of a predetermined number of data v(m) to v(n) stored in the memory 22 (step S12). The first feature amount F1(n) is, for example, the maximum value, minimum value, median value, average value, standard deviation, or regression coefficient of the data v(m) to v(n). However, the first feature amount F1(n) may also be kurtosis, gradient, quartile, outlier, or the like. The microcomputer 20 calculates at least one type of first feature amount F1(n).
[0040] The microcomputer 20 stores the calculated first feature amount F1(n) in the memory 22 (step S13). At this time, the microcomputer 20 overwrites and saves the first feature amount F1(n). That is, the microcomputer 20 deletes the past first feature amount F1 and stores the latest first feature amount F1(n) in the memory 22. In this way, by overwriting and saving the latest first feature amount F1(n), the amount of memory 22 used can be reduced. After executing step S13, the microcomputer 20 ends the first feature amount calculation process S2.
[0041] <2.2 Second feature calculation process> 6 is a flowchart showing details of the second feature calculation process S3. In the second feature calculation process S3, it is determined whether the current time is a timing that satisfies a predetermined condition (hereinafter, referred to as "calculation timing"), and if it is determined that the current time is the calculation timing, the second feature F2 is calculated.
[0042] The microcomputer 20 uses a calculation flag when determining whether the current time is the calculation timing. Specifically, the calculation flag includes "before start (0)" indicating the timing before the start of calculation of the second feature amount F2, "calculating (1)" indicating the period during which the second feature amount F2 is calculated, and "completed (2)" indicating the period during which the second feature amount F2 is calculated has ended.
[0043] First, the microcomputer 20 determines whether the calculation flag is set to "Complete (2)" (step S21). If the calculation flag is set to "Complete (2)" (Yes in step S21), the microcomputer 20 executes step S22, and if the calculation flag is not set to "Complete (2)" (No in step S21), the microcomputer 20 executes step S23.
[0044] In step S22, the microcomputer 20 determines whether or not to set the calculation flag to pre-start (0). Whether or not to set the calculation flag to pre-start (0) is determined based on the data acquired in step S1, for example. If the calculation flag is to be set to pre-start (0) (Yes in step S22), the microcomputer 20 changes the calculation flag to pre-start (0) (step S24) and proceeds to step S31. On the other hand, if the calculation flag is not to be set to pre-start (0) (No in step S22), the microcomputer 20 skips step S24 and proceeds to step S31.
[0045] In step S23, the microcomputer 20 determines whether the calculation flag is before start (0). If the calculation flag is before start (0) (Yes in step S23), the microcomputer 20 executes step S25, and if the calculation flag is not before start (0) (No in step S23), the microcomputer 20 executes step S26.
[0046] In step S25, the microcomputer 20 determines whether the current time is the timing to start calculating the second feature amount F2 (hereinafter referred to as the "calculation start timing"). Whether the current time is the calculation start timing is determined, for example, based on the data acquired in step S1. If the current time is the calculation start timing (Yes in step S25), the microcomputer 20 changes the calculation flag to calculation in progress (1) (step S27), and then executes step S26. If the current time is not the calculation start timing (No in step S25), the microcomputer 20 executes step S31.
[0047] In step S26, the microcomputer 20 determines whether the current time is the timing to end the calculation of the second feature amount (hereinafter referred to as the "calculation end timing") (step S25). Whether the current time is the calculation end timing is determined, for example, based on the data acquired in step S1. If the current time is the calculation end timing (Yes in step S26), the microcomputer 20 changes the calculation flag to completion (2) (step S28), and then executes step S29. On the other hand, if the microcomputer 20 does not change the calculation flag to completion (2) (No in step S26), it skips step S28 and executes step S29.
[0048] In step S29, the microcomputer 20 stores the data acquired in step S1 in memory. Also in step S29, the microcomputer 20 deletes the oldest data v(n-10) from the memory 22 in synchronization with storing the latest data v(n) in the memory 22, as described in Fig. 5. As a result, also in the second feature amount calculation process S3, only a predetermined number of data v(m) to v(n) are always stored in the memory 22.
[0049] After the latest data is stored in the memory 22, the microcomputer 20 calculates a second feature amount F2(n) indicating the features of a predetermined number of data v(m) to v(n) stored in the memory 22 (step S30). The second feature amount F2(n) is, for example, the maximum value, minimum value, median value, average value, standard deviation, or regression coefficient of the data v(n-9) to v(n). However, the second feature amount F2(n) may also be kurtosis, gradient, quartile, outlier, or the like. The microcomputer 20 calculates at least one type of second feature amount F2(n).
[0050] The microcomputer 20 stores the calculated second feature F2(n) in the memory 22 (step S31). At this time, the microcomputer 20 overwrites and saves the second feature F2(n). That is, the microcomputer 20 deletes the previous second feature F2 and data v, and stores the latest second feature F2(n) in the memory 22. In this way, by overwriting and saving the second feature F2, the amount of memory 22 used can be reduced. After executing step S31, the microcomputer 20 ends the second feature calculation process S3.
[0051] If the calculation flag is not cleared in step S22 and remains at completion (2), or if the calculation flag is changed to before start (0) in step S24, a default value (e.g., zero, -1, ∞, etc.) is stored in the memory 22 as the second feature F2 in step S31. As a result, during a period in which the second feature F2 is not calculated, the second feature F2 calculated in the past is updated to the default value. This makes it possible to prevent the past second feature F2 from affecting the prediction result of the future value.
[0052] As described above, in the first feature amount calculation process S2 and the second feature amount calculation process S3, the microcomputer 20 stores only a predetermined number of data v(m) to v(n) in the memory 22, rather than the entire time-series data D1. This makes it possible to save resources of the memory 22.
[0053] Furthermore, in the first feature amount calculation process S2 and the second feature amount calculation process S3, the microcomputer 20 does not calculate the feature amount based on the entire time series data D1, but calculates the feature amount based on only a predetermined number of data v(m) to v(n) stored in the memory 22. This can save resources of the processor 21. Therefore, even if the resources of the microcomputer 20 are small, future values can be appropriately predicted from the time series data D1.
[0054] Furthermore, in the second feature calculation process S3, the second feature F2 is not calculated every time data is acquired in step S1, but is calculated only when the current timing satisfies a predetermined condition, thereby saving resources of the processor 21.
[0055] Furthermore, in the mechanical device 1, the calculation of the feature amount is performed in the microcomputer 20. In this manner, by performing the feature amount calculation in a device separate from the central processing unit 30, the load on the processor 31 of the central processing unit 30 can be reduced.
[0056] Furthermore, the microcomputer 20 can instantly calculate the feature amount from the current data, thereby improving the immediacy of prediction of the future state of the mechanical device 1.
[0057] <3. Example of application to liquid ejection device> 7 is a diagram showing the configuration of a liquid discharge device 1a, which is an example of the mechanical device 1. The liquid discharge device 1a is a device that discharges a processing liquid L onto the surface of a processing object 9. The processing object 9 is, for example, a semiconductor wafer, a substrate for a flat panel display, a printed circuit board, an industrial part, a base material for printing, etc. The processing liquid L is, for example, a cleaning liquid, a resist liquid, a developing liquid, etc.
[0058] As shown in FIG. 7, the liquid discharge device 1a includes a tank 41, a pump 42, a flow rate sensor 10a, a flow rate control valve 44, a nozzle 45, a valve controller 46, a microcomputer 20a, and a central processing unit 30a.
[0059] The tank 41 and the nozzle 45 are connected by a pipe 48. The pump 42, the flow rate sensor 10a, and the flow rate control valve 44 are provided on the pipe 48. When the pump 42 is operated, the processing liquid L is supplied from the tank 41 through the pipe 48 to the nozzle 45. Then, the processing liquid L is discharged from the discharge port of the nozzle 45 onto the surface of the workpiece 9.
[0060] The flow rate sensor 10a is a sensor that measures the flow rate (flow velocity) of the processing liquid L flowing through the pipe 48. The flow rate sensor 10a corresponds to the sensor 10 of the mechanical device 1 shown in FIG. 1. The flow rate sensor 10a measures the flow rate of the processing liquid L at short time intervals. This provides time-series data D1 that indicates the discharge flow rate of the processing liquid L. The flow rate sensor 10a outputs the measured data to the valve controller 46 and the microcomputer 20a.
[0061] The flow control valve 44 is a valve for controlling the flow rate of the processing liquid L supplied to the nozzle 45. The valve controller 46 adjusts the opening of the flow control valve 44 (hereinafter referred to as the "valve opening") in multiple stages. The valve controller 46 is configured, for example, by a computer equipped with a processor such as a CPU and a memory. The valve controller 46 adjusts the opening of the flow control valve 44 based on data measured by the flow sensor 10a. In this way, the flow rate of the processing liquid L flowing through the pipe 48 is feedback-controlled. In addition, the valve controller 46 outputs data indicating the valve opening to the microcomputer 20a, for example, at minute time intervals. The valve opening corresponds to time-series data D1 that changes over time. In other words, the valve controller 46 corresponds to a "time-series data output unit."
[0062] The microcomputer 20a is electrically connected to the flow sensor 10a. The microcomputer 20a corresponds to the microcomputer 20 of the mechanical device 1 shown in Fig. 1. The central processing unit 30a is electrically connected to the microcomputer 20a and the valve controller 46. The central processing unit 30a corresponds to the central processing unit 30 of the mechanical device 1 shown in Fig. 1.
[0063] 8 is a diagram showing time series data D1 of the discharge flow rate output from the flow sensor 10a and a time change in the valve opening degree output from the valve controller 46. The microcomputer 20a calculates a feature amount from the time series data D1 of the discharge flow rate output from the flow sensor 10a. The feature amount includes, for example, a first increase rate m1 and a second increase rate m2. The first increase rate m1 and the second increase rate m2 will be described below.
[0064] <First growth rate m1> The first increase rate m1 is an increase rate of the discharge flow rate calculated from the discharge flow rate measured at multiple points in time. The first increase rate m1 is calculated by the first characteristic amount calculation process S2 shown in FIG. 3. That is, the microcomputer 20a calculates the first increase rate m1 every time the flow rate sensor 10a acquires the discharge flow rate data v(n). Specifically, the microcomputer 20a calculates the first increase rate m1 at the current point in time t n In the above, when data v(m) to v(n) representing the discharge flow rate at multiple points in time stored in memory 22 is approximated to a straight line L1, a regression coefficient β1, which is the slope of the straight line L1, is calculated as a first increase rate m1. By acquiring the first increase rate m1 as a feature amount, it is possible to improve the accuracy of predicting a future increase or decrease in the flow rate.
[0065] <Second increase rate m2> The second increase rate m2 is the increase rate of the discharge flow rate with respect to the time from the rising point of the discharge flow rate to the present time. The rising point is specifically the point t when the discharge flow rate exceeds the rising flow rate (for example, 50 L / min) corresponding to the rising. n′ The previous time point t n′-1 The second increase rate m2 is determined as the rising point t n′-1 The data v(n′-1) and the current time t n The second increase rate m2 is expressed by the slope of a straight line L2 that connects the data v(n) and the data v(n). Specifically, the second increase rate m2 is calculated by the following formula (1).
[0066] (Equation 1) m2 = (v(n) - v(n' - 1)) / (t n -t n′-1 )
[0067] The second increase rate m2 is calculated by the second characteristic amount calculation process S3. At this time, the calculation start timing is set to the rising point. Specifically, in step S25 shown in FIG. 6, the microcomputer 20a judges whether or not the current discharge flow rate has exceeded the rising flow rate. If the discharge flow rate has exceeded the rising flow rate, the microcomputer 20a judges that the current time is the calculation start timing and executes step S27 (setting of the calculating (1) flag). In addition, the microcomputer 20a sets the calculation start timing to the time t immediately before the time when the discharge flow rate exceeded the rising flow rate. n′-1 , and at that time t n′-1 The discharge flow rate v(n'-1) is stored as a second increase rate calculation parameter in the memory 22. In step S30, the microcomputer 20a substitutes the second increase rate calculation parameter into the above formula (1) to calculate the second increase rate m2, which is a feature quantity.
[0068] 6, the microcomputer 20a determines whether the current time is the calculation end timing for ending the calculation of the second increase rate m2. The calculation end timing is set, for example, to the time when the discharge flow rate exceeds a flow rate that is a reference for ending the calculation (hereinafter referred to as a "calculation end reference flow rate"). As an example, the calculation end reference flow rate is set to 97.5% of a preset target discharge flow rate.
[0069] In step S22, the microcomputer 20a may determine whether the discharge flow rate acquired in step S1 is a predetermined value (for example, zero). If the discharge flow rate is the predetermined value, the microcomputer 20a may execute step S24.
[0070] In step S25, the microcomputer 20a may determine whether the discharge flow rate has exceeded the rising flow rate based on the discharge flow rates at a plurality of points in time, rather than based on the current discharge flow rate. For example, the microcomputer 20a may determine whether the discharge flow rate has exceeded the rising flow rate based on the current discharge flow rate. n′ A number of consecutive time points in the past (e.g., five time points t n′ ~t n′-4If the representative value exceeds the rising flow rate, the microcomputer 20a may determine a representative value (for example, a median value) of the discharge flow rate at any one of the multiple time points (for example, the earliest time point t n′-4 ) may be set as the rising point, and that point and the discharge flow rate at that point may be stored in the memory 22 as a parameter for calculating the second increase rate.
[0071] Furthermore, in step S26, the microcomputer 20a may determine whether the discharge flow rate has exceeded the calculation end reference flow rate based on discharge flow rates at a plurality of points in time, rather than based on only the current discharge flow rate.
[0072] The microcomputer 20a predicts a future value of the discharge flow rate in step S4 shown in Fig. 2. Specifically, the microcomputer 20a predicts the discharge flow rate several seconds later using a trained model that receives as input the first increase rate m1, the second increase rate m2, the current discharge flow rate, and the current valve opening.
[0073] When the predicted future value of the discharge flow rate exceeds a predetermined threshold value (e.g., a target discharge flow rate), the microcomputer 20a transmits the future value of the discharge flow rate to the central processing unit 30a. The central processing unit 30a then controls the valve controller 46 based on the transmitted future value so that the discharge flow rate does not exceed the target flow rate.
[0074] <4. Example of application to printing devices> 9 is a diagram showing the configuration of a printing device 1b, which is another example of the mechanical device 1. The printing device 1b is an inkjet printing device that forms an image by ejecting ink onto the surface of a printing medium 9a. The printing device 1b includes a tank 51, a pump 52, a flow sensor 10b, a heater 53, a pump control unit 54, temperature sensors 10c and 10d, an ejection head 55, a heater control unit 56, a microcomputer 20b, and a central processing unit 30b.
[0075] The tank 51 and the ejection head 55 are connected by a pipe 58. The pump 52, the flow rate sensor 10b, the heater 53, and the temperature sensor 10d are provided on the pipe 58. When the pump 52 is operated, the ink La is supplied from the tank 51 through the pipe 58 to the ejection head 55. The ink La is then ejected from the multiple ejection openings of the ejection head 55 onto the surface of the printing medium 9a. The temperature sensor 10c is provided in the tank 51.
[0076] The pump control unit 54 controls the pump 52. The pump control unit 54 is configured by a computer having a processor such as a CPU and a memory. The pump control unit 54 controls the on / off of liquid delivery by the pump 52. The pump control unit 54 outputs a liquid delivery on / off signal Sg1 indicating the state of liquid delivery of the ink La by the pump 52 to the microcomputer 20b. The liquid delivery on / off signal Sg1 corresponds to time-series data D1 that changes over time. In other words, the pump control unit 54 corresponds to a "time-series data output unit."
[0077] The flow rate sensor 10b is a sensor that measures the flow rate (flow velocity) of the ink La flowing through the pipe 58. The flow rate sensor 10b corresponds to the sensor 10 of the mechanical device 1 shown in FIG. 1. The flow rate sensor 10b measures the flow rate of the ink La at very short time intervals. This provides time-series data D1 indicating the flow rate of the ink La. The flow rate sensor 10b outputs data indicating the measured flow rate to the microcomputer 20b.
[0078] The temperature sensor 10c measures the temperature of the ink La stored in the tank 51 (hereinafter referred to as the "tank temperature"). The temperature sensor 10c corresponds to the sensor 10 of the machine 1 shown in FIG. 1. The temperature sensor 10c measures the tank temperature at short time intervals. This provides time-series data D1 indicating the tank temperature. The temperature sensor 10c outputs data indicating the measured tank temperature to the microcomputer 20b.
[0079] The heater 53 heats the ink La supplied to the ejection head 55. The heater control unit 56 controls the amount of heat applied to the ink La by the heater 53 by controlling the voltage value applied to the heater 53. The heater control unit 56 is configured by, for example, a computer including a processor such as a CPU and a memory. The heater control unit 56 outputs data indicating the voltage value applied to the heater 53 (hereinafter referred to as "heater voltage") to the microcomputer 20b.
[0080] The temperature sensor 10d measures the temperature of the ink La heated by the heater 53 on the outlet side of the heater 53 (hereinafter referred to as the "outlet temperature"). The temperature sensor 10d corresponds to the sensor 10 of the mechanical device 1 shown in FIG. 1. The temperature sensor 10d measures the outlet temperature at short time intervals. As a result, time-series data D1 indicating the outlet temperature is obtained. The temperature sensor 10d outputs data indicating the measured outlet temperature to the microcomputer 20a. The heater control unit 56 controls the heater 53 based on the outlet temperature measured by the temperature sensor 10d. As a result, the outlet temperature of the ink La flowing through the pipe 58 is feedback-controlled.
[0081] The microcomputer 20b is electrically connected to the flow rate sensor 10b, the temperature sensor 10c, and the temperature sensor 10d. The microcomputer 20b is also electrically connected to the pump control unit 54 and the heater control unit 56. The microcomputer 20b corresponds to the microcomputer 20 of the mechanical device 1 shown in Fig. 1. The central processing unit 30b is electrically connected to the microcomputer 20b and the heater control unit 56. The central processing unit 30b corresponds to the central processing unit 30 of the mechanical device 1 shown in Fig. 1.
[0082] The microcomputer 20b predicts the "outlet temperature in a few seconds" as a future value by executing the processing procedure shown in Fig. 2. The inputs (explanatory variables) used in this prediction are, for example, the "current flow rate", the "current temperature inside the tank", the "current heater voltage", the "current outlet temperature", the "integrated flow rate for the past few seconds", the "moving average of the temperature inside the tank for the past few seconds", the "number of seconds from the start of liquid transfer", and the "number of seconds from the stop of liquid transfer".
[0083] The "integrated flow rate for the past few seconds" is a feature calculated by integrating the flow rate measured by the flow sensor 10b for the past few seconds. The "moving average of the temperature inside the tank for the past few seconds" is a feature obtained by calculating the moving average of the temperature inside the tank measured by the temperature sensor 10c for the past few seconds. The "integrated flow rate for the past few seconds" and the "moving average of the temperature inside the tank for the past few seconds" are calculated by the first feature calculation process S2 shown in FIG. 3. That is, these feature amounts are calculated each time a new flow rate and a new temperature inside the tank are acquired in step S1.
[0084] 10 is a diagram showing time-series data D1 of the outlet temperature output from the temperature sensor 10d, and the liquid feed on / off signal Sg1 output from the pump control unit 54. "Number of seconds from start of liquid feed" and "number of seconds from stop of liquid feed" are feature amounts calculated from the liquid feed on / off signal Sg1 output from the pump control unit 54. "Number of seconds from start of liquid feed" and "number of seconds from stop of liquid feed" are calculated by the second feature amount calculation process S3 shown in FIG. 6.
[0085] The "number of seconds from the start of liquid delivery" is the elapsed time from the start of liquid delivery to the current time. The start of liquid delivery is, specifically, the point at which the liquid delivery on / off signal Sg1 changes from off (0) to on (1). The "number of seconds from the start of liquid delivery" is calculated only from the start of liquid delivery to the stop of liquid delivery. The stop of liquid delivery is the point at which the liquid delivery on / off signal Sg1 changes from on (1) to off (0). In other words, the "number of seconds from the start of liquid delivery" is calculated from the time when the liquid delivery on / off signal Sg1 changes from off to on until the next time it turns off.
[0086] The "number of seconds since the liquid delivery stopped" is the time elapsed from the time when the liquid delivery stopped to the current time. The "number of seconds since the liquid delivery stopped" is calculated only from the time when the liquid delivery stopped to the time when the liquid delivery started. In other words, the "number of seconds since the liquid delivery stopped" is calculated from the time when the liquid delivery on / off signal Sg1 goes from on to off until it goes on again.
[0087] For example, when calculating the "number of seconds from the start of liquid transfer", the liquid transfer start time is used as the calculation start time in step S25 shown in FIG. 6. That is, in step S25, it is determined whether the current time is the liquid transfer start time. If the current time is the liquid transfer start time, the calculation flag is changed to "calculating (1)" in step S27. This starts the calculation of the "number of seconds from the start of liquid transfer". Also, the microcomputer 20b sets the current time to the liquid transfer start time t n′ Then, in step S30, the microcomputer 20b stores the current time t n From the start of liquid transfer t n′ By subtracting this, the "number of seconds from the start of liquid delivery" is calculated (see Figure 10).
[0088] In addition, the time when the liquid transfer is stopped is used as the calculation end time in step S26 shown in Fig. 6. That is, in step S26, it is determined whether the current time is the time when the liquid transfer is stopped. If the current time is the time when the liquid transfer is stopped, the calculation flag is changed to completion (2) in step S28. This stops the calculation of the "number of seconds from the start of the liquid transfer."
[0089] When the predicted "outlet temperature in a few seconds" is outside the target temperature range (hereinafter, "target temperature range"), the microcomputer 20b outputs the predicted value to the central processing unit 30b. The central processing unit 30b controls the heater control unit 56 based on the future value output from the microcomputer 20b so that the outlet temperature in a few seconds falls within the target temperature range.
[0090] <5. Modifications> Although the embodiment has been described above, the present invention is not limited to the above, and various modifications are possible.
[0091] For example, in the above description, the microcomputer 20 predicts the future value, but the central processing unit 30 may predict the future value. In this case, the microcomputer 20 outputs the feature amounts calculated by the first feature amount calculation process S2 and the second feature amount calculation process S3 to the central processing unit 30. Then, the central processing unit 30 may input the feature amounts output from the microcomputer 20 into the trained model to predict the future value.
[0092] The microcomputer 20b of the printer 1b may predict the "heater voltage to be set" as a future value. In this case, the "target value of the discharge liquid temperature in a few seconds" may be used as an input (explanatory variable) in addition to the above-mentioned inputs used to predict the "outlet temperature in a few seconds".
[0093] For example, the mechanical device 1 is not limited to the liquid ejection device 1a or the printing device 1b, but may be, for example, a transport device that transports an object.
[0094] Although the present invention has been described in detail, the above description is merely illustrative in all respects, and the present invention is not limited thereto. It is understood that countless variations not illustrated can be envisioned without departing from the scope of the present invention. The configurations described in the above embodiments and variations can be appropriately combined or omitted as long as they are not mutually contradictory. [Explanation of symbols]
[0095] 1 Mechanical equipment 10 Sensor (time-dependent data output section) 20 Microcomputer 21 Processors 22 Memory 30 Central Processing Unit
Claims
1. A data processing method for processing time series data using a device having a processor and a memory, comprising: a) storing the latest data of the time series data in the memory and deleting a part of past data from the memory; b) calculating feature quantities of the plurality of data stored in the memory; c) inputting the feature amount into a trained model to predict a future value; (c) a method for processing data,
2. 2. A data processing method according to claim 1, comprising: The step b) comprises: b-1) determining whether a timing that satisfies a predetermined condition has arrived; b-2) calculating the feature amount when it is determined in the step b-1) that the timing satisfies the predetermined condition; (c) a method for processing data,
3. 3. A data processing method according to claim 1, further comprising: A data processing method, wherein the time series data includes data indicating a discharge flow rate of a treatment liquid discharged from a nozzle.
4. 4. A data processing method according to claim 3, comprising: A data processing method, wherein the characteristic amount includes an increase rate of the discharge flow rate.
5. 5. A data processing method according to claim 3, further comprising: A data processing method, wherein the characteristic amount includes a rate of increase of the discharge flow rate with respect to an elapsed time after the discharge flow rate exceeds a first flow rate.
6. 3. A data processing method according to claim 1, further comprising: A data processing method, wherein the time series data is data measured by a sensor installed in a device that sends a treatment liquid in a tank to a heater.
7. 7. A data processing method according to claim 6, comprising: the time series data includes data indicating a flow rate of the treatment liquid sent from the tank to the heater, A data processing method, wherein the characteristic amount includes an integrated value of the flow rate over a predetermined period of time.
8. A data processing method according to claim 6 or 7, comprising: the time series data includes a temperature of the treatment liquid in the tank; A data processing method, wherein the characteristic amount includes a moving average of the temperature over a predetermined period of time.
9. A data processing method according to claim 6 or 7, comprising: the time series data includes data indicating start and stop of liquid transfer from the tank to the heater; A data processing method, wherein the feature amount includes a time elapsed since the start of liquid transfer or a time elapsed since the stop of liquid transfer.
10. A device for processing time series data, comprising: A processor; Memory, Equipped with The processor, a) storing the latest data of the time series data in the memory and deleting a part of past data from the memory; b) calculating feature quantities of the plurality of data stored in the memory; c) inputting the feature amount into a trained model to predict a future value; The device that runs it.
11. A mechanical device, a time series data output unit that outputs time series data; A device electrically connected to the time series data output unit, the device having a processor and a memory; a central processing unit electrically connected to the device; Equipped with The device comprises: a) storing the latest data of the time series data in the memory and deleting a part of past data from the memory; b) calculating feature quantities of the plurality of data stored in the memory; c) inputting the feature amount into a trained model to predict a future value; A mechanical device that performs the following:
12. A mechanical device, a time series data output unit that outputs time series data; A device electrically connected to the time series data output unit, the device having a processor and a memory; a central processing unit electrically connected to the device; Equipped with The device comprises: A) storing the latest data of the time series data in the memory and deleting a part of past data from the memory; B) calculating feature quantities of the plurality of data stored in the memory; C) outputting the feature amount to the central processing unit; Run The central processing unit is D) inputting the feature amount into a trained model to predict future values; A mechanical device that performs the following:
Citation Information
Patent Citations
Substrate processing apparatus and substrate processing method
JP2018163978A
Data processing device, data processing system, data processing method and program
JP2018190050A
Control apparatus, control method, and control program
JP2020135757A
Information processing device and information processing method
JP6935046B1
