Manufacturing methods for consumables

Abstract

To provide a manufacturing method for consumables that facilitates determination of whether prediction data is correct or not when degree of deterioration of consumables is predicted using a machine learning model.SOLUTION: A predictive data display device includes: a prediction unit that uses information about first characteristic data indicating degree of deterioration of consumables in a first period as input data, inputs, to a model where learning processing is performed using learning data whose correct data is information about second characteristic data indicating the degree of deterioration of the consumables in a second period after the first period, information about third characteristic data indicating degree of deterioration of consumables to be predicted during the first period, and calculates information about fourth characteristic data indicating degree of deterioration of the consumables to be predicted during the second period; and a display unit that, when displaying a graph of the fourth characteristic data, displays a graph of the second characteristic data together in a display mode according to degree of similarity of information about the first and third characteristic data.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] This disclosure is, Manufacturing methods for consumables Regarding. [Background technology]

[0002] A prediction technique is known that uses machine learning models to predict the degree of degradation of consumables such as batteries. According to this prediction technique, for example, characteristic data indicating the degree of degradation is acquired over a certain period, and the relationship between the characteristic data before degradation and the characteristic data after degradation is learned. This allows the characteristic data after degradation of a consumable to be predicted to be obtained from the characteristic data of the consumable to be predicted. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Special Publication No. 2010-539473 [Patent Document 2] Japanese Patent Publication No. 2013-217897 [Patent Document 3] Japanese Patent Publication No. 2019-113524 [Overview of the Initiative] [Problems that the invention aims to solve]

[0004] However, with prediction technologies that use machine learning models, it is difficult for users to judge whether the predicted data is accurate or not.

[0005] This disclosure makes it easier to determine the accuracy of prediction data when predicting the degree of deterioration of consumables using a machine learning model. [Means for solving the problem]

[0006] A predictive data display device according to the first aspect of this disclosure is: A prediction unit that takes information about first characteristic data indicating the degree of deterioration of a first consumable during a first period as input data, and information about second characteristic data indicating the degree of deterioration of the first consumable during a second period following the first period as ground truth data, as training data to train a trained model, inputs information about third characteristic data indicating the degree of deterioration of the second consumable to be predicted during the first period, and calculates information about fourth characteristic data indicating the degree of deterioration of the second consumable to be predicted during the second period. The system includes a display unit that, when displaying the graph of the fourth characteristic data, also displays the graph of the second characteristic data in a display manner corresponding to the similarity between the information regarding the first characteristic data and the information regarding the third characteristic data.

[0007] Furthermore, a second aspect of this disclosure is a predictive data display device described in the first aspect, When the display unit displays the graph of the second characteristic data, it displays it coupled with the graph of the third characteristic data.

[0008] Furthermore, a third aspect of this disclosure is a predictive data display device described in the second aspect, The information relating to the second characteristic data is the decay rate of the second characteristic data, with respect to characteristic data indicating the degree of deterioration of the first consumable at the end of the first period.

[0009] Furthermore, a fourth aspect of this disclosure is a predictive data display device described in the third aspect, The display unit displays a graph of the second characteristic data, which is generated by multiplying the characteristic data indicating the degree of deterioration of the second consumable item to be predicted at the end of the first period by the decay rate of the second characteristic data.

[0010] Furthermore, a fifth aspect of this disclosure is a predictive data display device as described in the second aspect, The information relating to the fourth characteristic data is the decay rate of the fourth characteristic data, based on characteristic data indicating the degree of deterioration of the second consumable item being predicted at the end of the first period.

[0011] Furthermore, a sixth aspect of this disclosure is a predictive data display device as described in the fifth aspect, The display unit displays a graph of the fourth characteristic data, which is generated by multiplying characteristic data indicating the degree of deterioration of the second consumable item to be predicted at the end of the first period by the decay rate of the fourth characteristic data.

[0012] Furthermore, a seventh aspect of this disclosure is a predictive data display device as described in the second aspect, The display unit displays the graph of the second characteristic data in a display color corresponding to the similarity.

[0013] Furthermore, an eighth aspect of this disclosure is a predictive data display device described in the first aspect, When the display unit displays the graph of the fourth characteristic data, it also displays information regarding the confidence interval calculated by the prediction unit.

[0014] Furthermore, a ninth aspect of this disclosure is a predictive data display device described in the first aspect, The display unit displays information indicating the type of the first consumable and information indicating the manufacturing conditions of the first consumable, along with the corresponding similarity score, as text.

[0015] Furthermore, a tenth aspect of this disclosure is a predictive data display device as described in the first aspect, The display unit calculates and displays the normality level of the fourth characteristic data by comparing the width of the confidence interval calculated by the prediction unit with the minimum and maximum widths of the confidence interval calculated when the model is trained using the training data.

[0016] Furthermore, an eleventh aspect of this disclosure is a predictive data display device described in the first aspect, The aforementioned model is a Gaussian process regression model.

[0017] Furthermore, a twelfth aspect of this disclosure is a predictive data display device described in the first aspect, The first and second consumables are batteries, and the information regarding the first and third characteristic data are characteristic quantities measured by repeated charging and discharging tests of the batteries.

[0018] Furthermore, a thirteenth aspect of this disclosure is a predictive data display device described in the twelfth aspect, The first and third characteristic data mentioned above represent the discharge capacity maintenance rate in each cycle.

[0019] Furthermore, the predictive data display method according to the 14th aspect of this disclosure is: A prediction step involves inputting information about first characteristic data indicating the degree of deterioration of a first consumable during a first period as input data, and information about second characteristic data indicating the degree of deterioration of the first consumable during a second period following the first period as ground truth data, into a trained model that has undergone training processing using this trained model, and inputting information about third characteristic data indicating the degree of deterioration of a second consumable to be predicted during the first period, and calculating information about fourth characteristic data indicating the degree of deterioration of a second consumable to be predicted during the second period; When displaying the graph of the fourth characteristic data, the computer performs a display step of displaying the graph of the second characteristic data together with the information about the first characteristic data in a display manner corresponding to the similarity between the information about the third characteristic data and the information about the first characteristic data.

[0020] Furthermore, the predictive data display program according to the 15th aspect of this disclosure is A prediction step involves inputting information about first characteristic data indicating the degree of deterioration of a first consumable during a first period as input data, and information about second characteristic data indicating the degree of deterioration of the first consumable during a second period following the first period as ground truth data, into a trained model that has undergone training processing using this trained model, and inputting information about third characteristic data indicating the degree of deterioration of a second consumable to be predicted during the first period, and calculating information about fourth characteristic data indicating the degree of deterioration of a second consumable to be predicted during the second period; When displaying the graph of the fourth characteristic data, the computer is instructed to perform a display step of displaying the graph of the second characteristic data together with the information about the first characteristic data in a display manner corresponding to the similarity between the information about the third characteristic data and the information about the first characteristic data. [Effects of the Invention]

[0021] According to this disclosure, when predicting the degree of deterioration of consumables using a machine learning model, it becomes easier to determine whether the predicted data is accurate or not. [Brief explanation of the drawing]

[0022] [Figure 1] Figure 1 shows an example of the system configuration of the predictive data display system and the functional configuration of the learning device during the learning phase. [Figure 2] Figure 2 shows a specific example of the processing performed by the training data generation device. [Figure 3] Figure 3 shows an example of the system configuration of a prediction data display system and the functional configuration of a prediction data display device during the prediction phase. [Figure 4] Figure 4 shows a specific example of the processing performed by the data generation device for prediction. [Figure 5] Figure 5 shows an example of the hardware configuration of a learning device and a predictive data display device. [Figure 6] Figure 6 shows a specific example of the processing of the first scale conversion unit of the predictive data display device. [Figure 7]Figure 7 shows a specific example of processing by a trained Gaussian process regression model in a predictive data display device. [Figure 8] Figure 8 shows a specific example of the processing of the second scale conversion unit of the predictive data display device. [Figure 9] Figure 9 shows an example of a graph of the predicted data and the 95% confidence interval. [Figure 10] Figure 10 is the first diagram showing a specific example of the processing of the display screen generation unit of the predictive data display device. [Figure 11] Figure 11 is a second diagram showing a specific example of the processing of the display screen generation unit of the predictive data display device. [Figure 12] Figure 12 is a flowchart showing the learning process flow. [Figure 13] Figure 13 is the first flowchart showing the flow of the predictive data display process. [Figure 14] Figure 14 is a third diagram showing a specific example of the processing of the display screen generation unit of the predictive data display device. [Figure 15] Figure 15 is a second flowchart showing the flow of the predictive data display process. [Figure 16] Figure 16 is the fourth figure, which shows a specific example of the processing of the display screen generation unit of the predictive data display device. [Modes for carrying out the invention]

[0023] Each embodiment will be described below with reference to the attached drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0024] [First Embodiment] <System configuration of the predictive data display system and functional configuration of the learning device in the learning phase> First, we will describe the system configuration of the predictive data display system in the learning phase and the functional configuration of the learning device that constitutes the predictive data display system.

[0025] Figure 1 shows an example of the system configuration of the predictive data display system and the functional configuration of the learning device during the learning phase. As shown in Figure 1, the predictive data display system 100 during the learning phase includes a characteristic measurement device 110, a learning data generation device 120, and a learning device 130.

[0026] The characteristic measuring device 110 measures feature quantities as information related to characteristic data indicating the degree of degradation of consumables, and generates characteristic data based on the measured feature quantities. Consumables refer to items whose performance deteriorates with repeated use, such as batteries. When the consumable is a battery, the characteristic data indicating the degree of degradation refers to, for example, the discharge capacity maintenance rate in each cycle of a battery charge-discharge test. When the characteristic data is the discharge capacity maintenance rate for each cycle, the feature quantities, which are information related to the characteristic data, refer to current data, voltage data, etc., measured in each cycle.

[0027] Furthermore, the term "battery" here encompasses a variety of types, with lithium-ion secondary batteries being one example. Lithium-ion secondary batteries also include those manufactured under different manufacturing conditions, from the perspective of material development and battery design. More specifically, this includes batteries manufactured by comparing and modifying various components such as positive electrode materials, negative electrode materials, and electrolyte materials, as well as batteries manufactured by comparing and modifying various activation and aging conditions after battery assembly.

[0028] The characteristic measurement device 110 acquires characteristic data for various types of consumables (an example of a first consumable, for example, a battery). The example in Figure 1 shows how feature quantities I, II, III, ... are measured for consumables I, II, III, ... which are of different types, and how characteristic data I, II, III, ... are generated.

[0029] The generated characteristic data I, II, III, ... are input to the training data generation device 120 along with the feature quantities I, II, III, ... and the corresponding manufacturing conditions i, ii, iii, ... for the consumables.

[0030] The training data generation device 120 generates training data 121 used in the training process by the training device 130. As shown in Figure 1, the training data 121 has the following information items: "ID", "reference data", "input data", and "correct answer data", and each information item contains, • "ID": Type of each consumable, • "Reference data": Manufacturing conditions for each consumable item, • "Input data": Feature quantities measured from the start of measurement to the reference time. • "Correct Data": The decay rate of characteristic data after the reference time, based on the characteristic data at the reference time. The following is input. The reference time refers to the time when the period from the start of measurement to the end of the period from the start of measurement to the end of the period from the start of measurement to the end of the period from the start of measurement to the end of the period from the start of measurement to the end of the period from the start of measurement to the end of the period from the start of measurement to the end of the measurement of the feature quantity is called the second period.

[0031] However, the term "period" as used herein is not limited to time, but also includes concepts equivalent to time. For example, in the case of a consumable item being a battery, the reference time refers to the number of cycles from the start of measurement until the number of cycles reaches a predetermined value.

[0032] The training data 121 generated by the training data generation device 120 is stored in the training data storage unit 133 of the training device 130.

[0033] The learning device 130 has a learning program installed, and when this program is executed, the learning device 130 functions as a Gaussian process regression model 131 and a comparison / modification unit 132.

[0034] The Gaussian process regression model 131 is a nonparametric probabilistic model capable of outputting predicted data (in this embodiment, the predicted decay rate from the baseline onward) along with the variance of the predicted data (in this embodiment, the width of the 95% confidence interval). In this embodiment, the Gaussian process regression model 131 receives input data (features measured from the start of measurement to the baseline) of the training data 121 and outputs output data (predicted decay rate from the baseline onward).

[0035] The comparison / modification unit 132 updates the model parameters of the Gaussian process regression model 131 so that the output data matches the ground truth data of the training data 121 (the decay rate of the characteristic data after the reference time, with the characteristic data at the reference time as the reference).

[0036] The updated model parameters are retained in the trained Gaussian process regression model 332 (details below) and used in the prediction phase.

[0037] <Specific example of processing by a training data generation device> Next, we will describe a specific example of the processing of the training data generation device 120, which constitutes the predictive data display system 100 during the learning phase. Figure 2 shows a specific example of the processing of the training data generation device.

[0038] In Figure 2, the symbols 210_1, 210_2, 210_3, ... represent specific examples of characteristic data I, II, III, ..., respectively. As shown in Figure 2, in characteristic data I, II, III, ..., the horizontal axis represents time t, and the vertical axis represents characteristic data y(t) at time t.

[0039] Furthermore, in reference numeral 210_1, y(t0)_I is the characteristic data of consumable I at the start of measurement (time = t0), y(t B )_I is the reference time (time = t) of consumable I. B The characteristic data of ) are shown separately. Furthermore, y(t T )_I is the time when the measurement of consumable I ends (time = t Trepresents the characteristic data of ().

[0040] On the other hand, feature quantity 1 (t X ), feature quantity 2 (t X ), feature quantity 3 (t X ), ··· are each feature quantity used to generate characteristic data y(t X (where t0 ≦ t X ≦ t T ). X )_I, etc.

[0041] According to the specific example of FIG. 2, the learning data generation device 120 · characteristic data I, II, III, ··· indicated by symbols 210_1, 210_2, 210_3, ···, and · feature quantity 1 (t X ), feature quantity 2 (t X ), feature quantity 3 (t X ), ···, and generate learning data 121' based on them.

[0042] The learning data 121' is a specific example of the learning data 121 shown in FIG. 1, and is associated with "ID" = I, · "reference data": manufacturing condition i · "input data": feature quantity 1 (t X ), feature quantity 2 (t X ), ··· feature quantity n (t X ), · "correct answer data": attenuation rate r(t X )_I = y(t X )_I / y(t B )_I, are each input.

[0043] Similarly, associated with "ID" = II, · "reference data" = manufacturing condition ii · "input data": feature quantity 1 (t X ), feature quantity 2 (t X ), ··· feature quantity n (t X ), · "correct answer data": attenuation rate r(tX )_II=y(t X )_II / y(t B )_II, These will be entered respectively.

[0044] Similarly, by associating "ID" with III, • "Reference data" = Manufacturing conditions iii ·"Input data": Feature 1(t X )_III, Feature 2(t X )_III, ···Feature n(t X )_III, • "Correct data": Attenuation rate r(t X )_III=y(t X )_III / y(t B )_III, These will be entered respectively.

[0045] <System configuration of the forecast data display system in the forecasting phase and functional configuration of the forecast data display device> Next, we will describe the system configuration of the prediction data display system in the prediction phase and the functional configuration of the prediction data display device that constitutes the prediction data display system.

[0046] Figure 3 shows an example of the system configuration of a prediction data display system and the functional configuration of a prediction data display device in the prediction phase. As shown in Figure 3, the prediction data display system 300 in the prediction phase includes a characteristic measuring device 110, a prediction data generation device 320, and a prediction data display device 330.

[0047] Of these, the characteristic measurement device 110 has already been explained using Figure 1, so its explanation will be omitted here. In the prediction phase, the characteristic measurement device 110 measures feature quantities N as information about the characteristic data N of the consumable N to be predicted (an example of a second consumable; for example, a battery manufactured under manufacturing conditions m). The characteristic measurement device 110 also generates characteristic data N based on the measured feature quantities N. More specifically, an example of a consumable, a battery manufactured under manufacturing conditions m, is a lithium-ion secondary battery of the same type as the one used in the generation of the training data described above, and is manufactured under manufacturing conditions m.

[0048] The prediction data generation device 320 generates prediction data 321 used for prediction data display processing by the prediction data display device 330. As shown in Figure 3, the prediction data 321 has the following information items: "ID", "Reference data", "Scale conversion data", "Input data", and "Display data", and each information item contains, • "ID": Type of consumable to be predicted, • "Reference data": Manufacturing conditions of the consumables to be predicted, • "Scale conversion data": Characteristic data at the reference time, • "Input data": Feature quantities measured from the start of measurement to the reference time. • "Display data": Characteristic data from the start of measurement to the reference time. The following is entered.

[0049] The prediction data 321 generated by the prediction data generation device 320 is notified to the prediction data display device 330.

[0050] The predictive data display device 330 has a predictive data display program installed, and when this program is executed, the predictive data display device 330 will • First scale conversion unit 331, • Pre-trained Gaussian process regression model 332, • Second scale conversion unit 333, ·Display screen generation unit 334, It functions as such.

[0051] The first scale conversion unit 331 is • The characteristic data of the reference time included in the “scale conversion data” of the prediction data 321, • The decay rate included in the "ground truth data" of training data 121, Based on this, the first scale transformation unit 331 calculates the characteristic data (scale-transformed characteristic data) for each consumable (consumables I, II, III, ...) from the reference time onward. The first scale transformation unit 331 also determines the display format of the characteristic data for each consumable from the reference time onward, based on the similarity for each consumable notified by the trained Gaussian process regression model 332. Furthermore, the first scale transformation unit 331 notifies the display screen generation unit 334 of the characteristic data for each consumable from the reference time onward, for which the display format has been determined, along with the similarity.

[0052] The trained Gaussian process regression model 332 is an example of the prediction unit, and it retains the model parameters that have been updated by performing a training process on the Gaussian process regression model 131 during the training phase.

[0053] In the above explanation, a Gaussian process regression model was given as an example of a regression model, but this embodiment is not limited to this. For example, other regression models may be used, such as a Poisson process regression model, as long as they are capable of outputting the prediction data along with the 95% confidence interval of the prediction data.

[0054] The trained Gaussian process regression model 332 receives the features measured from the start of measurement to the baseline, which are included in the "input data" of the prediction data 321. The process involves calculating the similarity for each consumable item to the features measured from the start of measurement to the reference time, which are included in the "input data" of the training data 121, and notifying the first scale conversion unit 331 of this calculation. The process involves using the calculated similarity to each consumable to calculate the predicted decay rate of the target consumable N from the reference time onward, calculating the 95% confidence interval for the calculated predicted decay rate of the target consumable N from the reference time onward, and notifying the second scale conversion unit 333 of this calculation. Execute this.

[0055] The second scale conversion unit 333 is • The characteristic data of the reference time included in the “scale conversion data” of the prediction data 321, • Predicted decay rate after the reference time, as notified by the trained Gaussian process regression model 332, Based on this, characteristic data (predictive data) of the target consumables from the reference time onward is predicted.

[0056] Furthermore, the second scale conversion unit 333 calculates a 95% confidence interval converted to characteristic data based on the characteristic data at the reference time included in the "scale conversion data" of the prediction data 321 and the 95% confidence interval notified by the trained Gaussian process regression model 332.

[0057] Furthermore, the second scale conversion unit 333 notifies the display screen generation unit 334 of the predicted characteristic data (predicted data) of the target consumable from the reference time onward, and the calculated 95% confidence interval (converted to characteristic data).

[0058] The display screen generation unit 334 is an example of a display unit and generates a display screen. The display screen generated by the display screen generation unit 334 includes at least: • Characteristic data from the start of measurement to the reference period, included in the "display data" of prediction data 321. • Predicted data of consumable N from the reference time onward, notified by the second scale conversion unit 333. • The 95% confidence interval (converted to characteristic data) of consumable N, notified by the second scale conversion unit 333, • Characteristic data of each consumable (consumables I, II, III, ...) from the reference time onward, as notified by the first scale conversion unit 331. It includes.

[0059] At this time, the display screen generation unit 334 displays the characteristic data from the reference time onward, which has been notified by the first scale conversion unit 331, in the display mode determined by the first scale conversion unit 331.

[0060] In this way, when the predictive data display device 330 displays predictive data for the consumables to be predicted using a machine learning model, it also displays characteristic data generated based on the training data used in the training process of the machine learning model. At this time, the predictive data display device 330 displays the characteristic data generated based on the training data in a display mode according to the similarity of the features. This makes it possible for the user to compare the predictive data for the consumables to be predicted with past characteristic data whose display mode has been changed according to the similarity. As a result, it becomes easier for the user to judge whether the predictive data is accurate or not.

[0061] <Specific examples of processing by a data generation device for prediction> Next, a specific example of the processing of the prediction data generation device 320, which constitutes the prediction data display system 300 in the prediction phase, will be described. Figure 4 is a diagram showing a specific example of the processing of the prediction data generation device.

[0062] In Figure 4, the symbol 410 indicates a specific example of characteristic data N. As shown in Figure 4, in characteristic data N, the horizontal axis represents time t, and the vertical axis represents the characteristic data y(t) of the consumable product N to be predicted at time t.

[0063] Furthermore, in reference numeral 410, y(t0)_N is the characteristic data of consumable N at the start of measurement (time = t0), y(t B )_N is the reference time (time = t) of consumable N. B These represent the characteristic data of each of the following:

[0064] On the other hand, feature 1(t X ), Feature 2(t X ), Feature 3(t X ), ... is at time t X (However, t0 ≤ t X≤t B ) characteristic data y(t X This represents each feature used to generate )_N.

[0065] According to the specific example in Figure 4, the prediction data generation device 320 uses characteristic data N shown by reference numeral 410 and feature quantity 1(t X ), Feature 2(t X ), Feature 3(t X Based on the above, predictive data 321' is generated.

[0066] Prediction data 321' is a specific example of prediction data 321 shown in Figure 3, and is associated with "ID"=N. • "Reference data": Manufacturing conditions ε, • Data for scaling conversion: Characteristic data y(t) B )_N, ·"Input data": Feature 1(t X )_N, Feature 2(t X )_N, ···feature n(t X )_N, • "Display data": Characteristic data y(t X )_N, These will be entered respectively.

[0067] <Description of the learning device and predictive data display device> Next, we will describe in detail the learning device 130 that constitutes the predictive data display system 100 in the learning phase and the predictive data display device 330 that constitutes the predictive data display system 300 in the prediction phase.

[0068] (1) Hardware configuration of the learning device and predictive data display device 330 The hardware configuration of the learning device 130 and the predictive data display device 330 will now be described. Figure 5 shows an example of the hardware configuration of the learning device and the predictive data display device. Since the learning device 130 and the predictive data display device 330 have similar hardware configurations, they will be described together using Figure 5.

[0069] As shown in Figure 5, the learning device 130 and the predictive data display device 330 each include a processor 501, memory 502, auxiliary storage device 503, I / F (Interface) device 504, communication device 505, and drive device 506. The hardware of the learning device 130 and the predictive data display device 330 are interconnected via a bus 507.

[0070] The processor 501 has various computing devices such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 501 reads various programs (for example, a learning program, a prediction data display program, etc.) into the memory 502 and executes them.

[0071] Memory 502 has main memory devices such as ROM (Read Only Memory) and RAM (Random Access Memory). The processor 501 and memory 502 form a so-called computer, and the computer realizes various functions by having the processor 501 execute various programs read from memory 502.

[0072] The auxiliary storage device 503 stores various programs and various data used when those programs are executed by the processor 501. For example, the learning data storage unit 133 mentioned above is implemented in the auxiliary storage device 503.

[0073] The I / F device 504 is a connection device that connects to external devices, namely the display device 510 and the operating device 520. The communication device 505 is a communication device for communicating with external devices (e.g., the learning data generation device 120 and the prediction data generation device 320) via a network.

[0074] The drive device 506 is a device for setting the recording medium 530. The recording medium 530 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 530 may also include semiconductor memory that records information electrically, such as ROMs and flash memory.

[0075] The various programs to be installed in the auxiliary storage device 503 are installed, for example, when the distributed recording medium 530 is set in the drive device 506 and the various programs recorded on the recording medium 530 are read by the drive device 506. Alternatively, the various programs to be installed in the auxiliary storage device 503 may be installed by downloading them from the network via the communication device 505.

[0076] (2) Specific example of processing of the first scale conversion unit of the predictive data display device Next, a specific example of the processing of the first scale conversion unit 331 of the predictive data display device 330 will be described. Figure 6 is a diagram showing a specific example of the processing of the first scale conversion unit of the predictive data display device.

[0077] As shown in Figure 6, the first scale conversion unit 331 includes a damping rate calculation unit 601 and a display mode changing unit 602.

[0078] The damping rate calculation unit 601 uses characteristic data y(t) as characteristic data for the reference time. B The characteristic data for each consumable (consumables I, II, III, ...) after the reference time is calculated by obtaining )_N and multiplying it by the decay rate after the reference time for each consumable (consumables I, II, III, ...). In the example in Figure 6, the decay rates for consumables I, II, III, ... are as follows: ·r(t x )_I=y(t X )_I / y(t B )_I(however, t B ≤t X ≤t T ), ·r(tx )_II=y(t X )_II / y(t B )_II(however, t B ≤t X ≤t T ), ·r(t x )_III=y(t X )_III / y(t B )_III(however, t B ≤t X ≤t T ), ... By multiplying these by the same factor, the characteristic data from the reference time onward will be obtained, · Characteristic data of consumables I from the reference time onward {y(t X )_I / y(t B )_I}×y(t B )_N(however, t B ≤t X ≤t T ), • Characteristic data of consumables II from the reference time onward {y(t X )_II / y(t B )_II}×y(t B )_N(however, t B ≤t X ≤t T ), • Characteristic data of consumable item III from the reference time onward {y(t X )_III / y(t B )_III}×y(t B )_N(however, t B ≤t X ≤t T ), ... This shows how it was calculated.

[0079] The display mode changing unit 602 determines the display mode of the characteristic data for each consumable (consumables I, II, III, ...) from the reference time based on the similarity for each consumable notified by the trained Gaussian process regression model 332. The example in Figure 6 shows how the similarity for consumables I, II, III, ... is notified by the trained Gaussian process regression model 332, with similarity scores of I, II, III, ... respectively.

[0080] Furthermore, the example in Figure 6 shows that when the similarity score for each consumable, as notified by the trained Gaussian process regression model 332, is "1", the display mode change unit 602 determines that the display color of the graph of characteristic data from the corresponding reference time onward is blue.

[0081] Furthermore, the example in Figure 6 shows that when the similarity score for each consumable, as notified by the trained Gaussian process regression model 332, is "0", the display mode change unit 602 determines that the display color of the graph of characteristic data from the corresponding reference time onward is gray.

[0082] Furthermore, if the similarity score for each consumable, as notified by the trained Gaussian process regression model 332, is greater than "0" and less than "1", the display mode change unit 602 determines the display color of the graph of the characteristic data from the corresponding reference time onward to an intermediate color between gray and blue.

[0083] For example, in the display mode changing unit 602, if the similarity is close to "0", the display color of the graph of the characteristic data from the corresponding reference time onward is determined to be a color close to gray from the gray-to-blue gradient. Also, in the display mode changing unit 602, if the similarity is close to "1", the display color of the graph of the characteristic data from the corresponding reference time onward is determined to be a color close to blue from the gray-to-blue gradient.

[0084] However, the method for determining the display color by the display mode changing unit 602 is not limited to this. For example, the color may be set to gray if the similarity is less than a predetermined threshold, and to blue if the similarity is equal to or greater than a predetermined threshold.

[0085] (3) Specific Example of the Process of the Learned Gaussian Process Regression Model of the Prediction Data Display Device Next, a specific example of the process of the learned Gaussian process regression model 332 of the prediction data display device 330 will be described. FIG. 7 is a diagram showing a specific example of the process of the learned Gaussian process regression model of the prediction data display device.

[0086] As shown in FIG. 7, the learned Gaussian process regression model 332 includes a similarity calculation unit 701, a confidence interval calculation unit 702, a decay rate prediction unit 703, and a model parameter holding unit 704.

[0087] The similarity calculation unit 701 is input with the feature amounts measured from the start time of measurement to the reference time included in the "input data" of the prediction data 321 and the feature amounts measured from the start time of measurement to the reference time included in the "input data" of the learning data 121.

[0088] In the example of FIG. 7, as the feature amounts measured from the start time of measurement to the reference time included in the "input data" of the prediction data 321, · Feature amount F y (Feature amount 1(t X )_N, Feature amount 2(t X )_N, ··· Feature amount n(t X )_N, provided that t0 ≦ t X ≦ t B ), shows the state of being input. Also, in the example of FIG. 7, as the feature amounts measured from the start time of measurement to the reference time included in the "input data" of the learning data 121, · Feature amount F y (Feature amount 1(t X )_I, Feature amount 2(t X )_I, ··· Feature amount n(t X )_I, provided that t0 ≦ t X ≦ t B ), · Feature amount F y (Feature amount 1(t X )_II, Feature amount 2(t X )_II, ··· Feature amount n(tX )_II, where t0≦t X ≤t B ), Feature F y (Feature 1(t) X )_III, Feature 2(t X )_III, ···Feature n(t X )_III, where t0≦t X ≤t B ), ... This shows how the input was received.

[0089] Furthermore, the similarity calculation unit 701 calculates the input feature quantity F x , Feature F y Based on this, the similarity of each consumable (Consumable I, II, III, ...) is calculated using the following formula 1.

[0090]

number

[0091] Furthermore, the similarity calculation unit 701 notifies the first scale conversion unit 331 of the similarity for each consumable that it has calculated. The example in Figure 7 shows how the similarity for consumable I is notified to the first scale conversion unit 331, the similarity for consumable II is notified to the first scale conversion unit 331, and the similarity for consumable III is notified to the first scale conversion unit 331.

[0092] The confidence interval calculation unit 702 uses the similarity to each consumable (consumable I, II, III, ...) calculated by the similarity calculation unit 701 to calculate a 95% confidence interval for the predicted attenuation rate calculated by the attenuation rate prediction unit 703. In the example in Figure 7, the 95% confidence interval is (Max_r'(t X )_N、Min_r'(t X This shows how )_N) was calculated.

[0093] The attenuation rate prediction unit 703 calculates the predicted attenuation rate of the consumable N to be predicted from the reference time onward, using the similarity to each consumable (consumable I, II, III, ...) calculated by the similarity calculation unit 701. In the example in Figure 7, the predicted attenuation rate of the consumable N to be predicted from the reference time onward is calculated as the predicted attenuation rate r'(t X This shows how )_N was calculated. The attenuation rate prediction unit 703 calculates the predicted attenuation rate r'(t) by, for example, calculating the kernel shown in equation 2 below. X Calculate )_N.

[0094]

number

[0095] The model parameter holding unit 704 holds the model parameters calculated when the learning process was performed during the learning phase. In the example in Figure 7, the model parameter holding unit 704 holds at least C as model parameters. var γ, C bias This indicates that α is retained.

[0096] Furthermore, in the similarity calculation unit 701, the input feature quantity F x , Feature F y The input features may be transformed and used; for example, values ​​obtained by performing dimensionality reduction processing such as principal component analysis, independent component analysis, or kernel PCA may be used.

[0097] (4) Specific example of processing of the second scale conversion unit of the predictive data display device Next, a specific example of the processing of the second scale conversion unit 333 of the predictive data display device will be described. Figure 8 is a diagram showing a specific example of the processing of the second scale conversion unit of the predictive data display device.

[0098] As shown in Figure 8, the second scale conversion unit 333 includes a prediction data calculation unit 801 and a confidence interval calculation unit 802.

[0099] The prediction data calculation unit 801 uses characteristic data y(t) as characteristic data for the reference time. B By obtaining )_N and multiplying it by the predicted decay rate for the consumable N to be predicted, the predicted data for the consumable N from the reference time onward is predicted. In the example in Figure 8, the predicted decay rate for the consumable N to be predicted is r'(t X By multiplying by )_N, the predicted data from the reference time onward is obtained. · Predicted data y'(t X )_N=y(t B )_N×r'(t X )_N This shows how it was predicted.

[0100] The confidence interval calculation unit 802 uses characteristic data y(t) as characteristic data for the reference time. B The 95% confidence interval for characteristic data is calculated by obtaining )_N and multiplying it by the 95% confidence interval for the predicted attenuation rate. In the example in Figure 8, the 95% confidence interval for the predicted attenuation rate is as follows: 95% confidence interval (Max_r'(t X )_N,Min_r'(t X )_N) By multiplying by this, the 95% confidence interval converted to characteristic data is obtained. 95% confidence interval (Max_r'(t X )_N×y(t B )_N,Min_r'(t X )_N×y(t B )_N) This shows how it was calculated.

[0101] Furthermore, there are several methods for graphing the predicted data predicted by the prediction data calculation unit 801 and the 95% confidence interval (converted to characteristic data) calculated by the confidence interval calculation unit 802.

[0102] Figure 9 shows an example of a graph of the predicted data and 95% confidence interval. Note that in the example in Figure 9, for ease of explanation, the characteristic data y(t) is used as the display data. X )_N(where t0≦t X ≤t B ) is also shown (see the red line in the colorized Figure 9).

[0103] In Figure 9, Graph 910 shows only three predicted time points as forecast data after the reference time, plotted and connected by dotted lines as a line graph.

[0104] Furthermore, in Figure 9, Graph 920 shows only three time points as predicted data after the reference time, along with a 95% confidence interval, as a box plot (see the green box plot in the colorized Figure 9).

[0105] On the other hand, in Figure 9, graph 930 shows the predicted data from the reference time onward, predicted hourly, along with the 95% confidence interval, as a curve graph (see the green dotted line in the colorized Figure 9).

[0106] Note that graphs 910-930 are just examples of graphing methods, and graphs may be created using methods other than those shown.

[0107] (5) Specific example of processing of the display screen generation unit of the predictive data display device Next, a specific example of the processing of the display screen generation unit 334 of the predictive data display device 330 will be described. Figure 10 is the first figure showing a specific example of the processing of the display screen generation unit of the predictive data display device. In Figure 10, reference numerals 1010_1, 1010_2, 1010_3, ... are notified by the first scale conversion unit 331. · Characteristic data of consumables I from the reference time onward {y(t X )_I / y(t B )_I}×y(t B )_N(however, t B ≤t X ≤t T ), • Characteristic data of consumables II from the reference time onward {y(t X )_II / y(t B )_II}×y(t B )_N(however, t B ≤t X ≤t T ), • Characteristic data of consumable item III from the reference time onward {y(t X )_III / y(t B )_III}×y(t B )_N(however, t B ≤t X ≤t T ), ... This is a graph of the data, displayed using the chosen color scheme.

[0108] Furthermore, in Figure 10, reference numeral 920 denotes the consumable N to be predicted. • Predicted data y'(t) from the reference time onward X )_N=y(t B )_N×r'(t X )_N and, • 95% confidence interval converted to characteristic data (Max_r'(t X )_N×y(t B )_N,Min_r'(t X )_N×y(t B )_N) and, This is a graph illustrating that data.

[0109] Furthermore, as shown in Figure 10, the display screen generation unit 334 includes a text data generation unit 1001 and a graph merging unit 1002.

[0110] The text data generation unit 1001 reads the "ID" and "reference data" of the training data 121 stored in the training data storage unit 133. The text data generation unit 1001 also obtains the similarity (similarity I, II, III, ...) for each consumable (consumable I, II, III, ...) from the first scale conversion unit 331.

[0111] Furthermore, the text data generation unit 1001 generates a text table 1021 by associating the read "ID" and "reference data" with the acquired similarity score, and displays it on the display screen 1020.

[0112] The graph merging unit 1002 obtains a graph of characteristic data for the consumable N to be predicted, from the "display data" of the prediction data 321, from the start of measurement to the reference time. The graph merging unit 1002 also obtains a graph of the predicted data and a graph of the 95% confidence interval for the consumable N to be predicted from the second scale conversion unit 333 (indicated by 920). The graph merging unit 1002 also obtains graphs of characteristic data for each consumable (consumable I, II, III, ...) from the reference time onward from the first scale conversion unit 331 (indicated by 1010_1, 1010_2, 1010_3, ...).

[0113] Furthermore, the graph merging unit 1002 generates a merged graph 1022 by merging the acquired graphs and displays it on the display screen 1020.

[0114] In the combined graph 1022, the red line represents the characteristic data for the consumable N being predicted, from the start of measurement to the baseline. Also in the combined graph 1022, the green box plot represents the predicted data and 95% confidence interval for the consumable N being predicted (symbol 920). Furthermore, in the combined graph 1022, the blue to gray lines represent the characteristic data for each consumable (consumable I, II, III, ...) from the baseline onward (symbols 1010_1, 1010_2, 1010_3, ...). Finally, the color bars 1023 represent the relationship between the display color of the characteristic data for each consumable from the baseline onward and the similarity for each consumable (see the colorized Figure 10 for each color).

[0115] In this embodiment, when displaying the graph of predicted data (green box plot), a graph of characteristic data for each consumable (blue to gray), generated based on the training data used in the training process of the Gaussian process regression model, is also displayed. At this time, the graph from the baseline onward is scaled according to the characteristic data at the baseline before being displayed. This allows the user to compare the data for the consumable being predicted with past characteristic data color-coded according to similarity. As a result, it becomes easier for the user to judge the accuracy of the predicted data.

[0116] In the example shown in Figure 10, the graph of the predicted data (green box plot) largely overlaps with the graph of the characteristic data for each consumable, which is blue. Therefore, the user can determine that the predicted data is correct.

[0117] On the other hand, Figure 11 is a second figure showing a specific example of the processing of the display screen generation unit of the predictive data display device. In the example of the display screen 1110 in Figure 11, the similarity between the feature quantities for the consumable N to be predicted and the feature quantities for each consumable (consumables I, II, III, ...) is low. Therefore, in the combined graph 1122, the graph of the predicted data (green box plot) is within the range of the gray characteristic data graph. As a result, the user can determine that there is a high possibility that the predicted data is incorrect.

[0118] Judging the accuracy of such predictive data is particularly useful when new consumables are included in the prediction.

[0119] <Processing flow in the predictive data display system> Next, we will explain the processing flow in the predictive data display systems 100 and 300.

[0120] (1) Flow of the learning process in the predictive data display system during the learning phase First, we will explain the learning process flow in the predictive data display system 100 during the learning phase. Figure 12 is a flowchart showing the learning process flow.

[0121] In step S1201, the characteristic measuring device 110 acquires characteristic data generated for each consumable.

[0122] In step S1202, the learning data generation device 120 acquires the feature quantities used to generate characteristic data for each consumable from the start of measurement to the reference period.

[0123] In step S1203, the learning data generation device 120 calculates the decay rate for each consumable item based on the characteristic data from the reference time onward.

[0124] In step S1204, the training data generation device 120 generates training data.

[0125] In step S1205, the learning device 130 performs a learning process on the Gaussian process regression model using the training data.

[0126] In step S1206, the learning device 130 notifies the prediction data display device 330 of the trained Gaussian process regression model (and its model parameters) generated by the learning process.

[0127] (2) Flow of predictive data display processing in the predictive data display system during the prediction phase Next, we will explain the flow of the prediction data display process in the prediction data display system 300 during the prediction phase. Figure 13 is the first flowchart showing the flow of the prediction data display process.

[0128] In step S1301, the prediction data generation device 320 acquires characteristic data for the consumable to be predicted, generated by the characteristic measurement device 110, from the start of measurement to the reference period.

[0129] In step S1302, the prediction data generation device 320 acquires the feature quantities used by the characteristic measurement device 110 when it generated characteristic data for the consumables to be predicted from the start of measurement to the reference period.

[0130] In step S1303, the prediction data generation device 320 generates prediction data.

[0131] In step S1304, the prediction data display device 330 calculates the similarity between the feature quantities of each consumable stored as training data and the feature quantities of the consumable to be predicted stored as prediction data.

[0132] In step S1305, the predictive data display device 330 scales the characteristic data of each consumable stored as training data from a reference time onward.

[0133] In step S1306, the predictive data display device 330 determines the display mode of the characteristic data for each consumable after scale conversion from the reference time onward, based on the corresponding similarity.

[0134] In step S1307, the prediction data display device 330 inputs the feature quantities of the consumable to be predicted from the start of measurement to the reference point into the trained Gaussian process regression model.

[0135] In step S1308, the predictive data display device 330 acquires the predictive decay rate and the 95% confidence interval.

[0136] In step S1309, the prediction data display device 330 scales the acquired prediction attenuation rate and 95% confidence interval to obtain prediction data and a 95% confidence interval (converted to characteristic data).

[0137] In step S1310, the predictive data display device 330, • A graph of characteristic data for the consumables being predicted, from the start of measurement to the reference period, • Graphs of predicted data and 95% confidence intervals (converted to characteristic data) from the reference time onward, • Graphs of characteristic data for each consumable product from the reference time onward, with the display color determined according to similarity. Generate a joined graph by combining the elements.

[0138] In step S1311, the predictive data display device 330 generates a text table that associates the similarity with the type of consumable and the manufacturing conditions.

[0139] In step S1312, the predictive data display device 330 generates and displays a display screen that includes a combined graph and a text table.

[0140] <Summary> As is clear from the above description, the predictive data display system according to the first embodiment is The system generates training data using information (features) related to the first characteristic data, which shows the degree of deterioration of each consumable from the start of measurement to the reference point, as input data, and information (attenuation rate) related to the second characteristic data, which shows the degree of deterioration of each consumable after the reference point, as ground truth data. • The generated training data is used to train the Gaussian process regression model, generating a trained Gaussian process regression model. • Information (features) regarding a third characteristic data point, which indicates the degree of deterioration of the consumables being predicted from the start of measurement to the reference time, is input into a pre-trained Gaussian process regression model. This allows for the calculation of information (predicted decay rate) regarding a fourth characteristic data point, which indicates the degree of deterioration of the consumables being predicted after the reference time. When displaying the fourth characteristic data, which shows the degree of deterioration of the consumables being predicted since the reference time, the second characteristic data is displayed in a manner that corresponds to the similarity between the information (features) related to the first characteristic data and the information (features) related to the third characteristic data.

[0141] This allows users to compare predicted data for the consumables they are predicting with historical characteristic data, whose display format has been modified according to similarity. As a result, users can more easily determine the accuracy of the predicted data.

[0142] In other words, according to this embodiment, when the degree of deterioration of consumables is predicted using a machine learning model, it becomes easier to determine whether the predicted data is correct or incorrect.

[0143] [Second Embodiment] In the first embodiment described above, the user can easily determine the accuracy of the prediction data by graphing and comparing the prediction data for the consumables to be predicted with the characteristic data for each consumable included in the training data. In contrast, in the second embodiment, the prediction data for the consumables to be predicted and the characteristic data for each consumable included in the training data are quantified and compared using a 95% confidence interval. The second embodiment will be described below, focusing on the differences from the first embodiment.

[0144] <Specific example of processing in the display screen generation unit of the predictive data display device> First, a specific example of the processing of the display screen generation unit of the predictive data display device 330 will be described. Figure 14 is a third figure showing a specific example of the processing of the display screen generation unit of the predictive data display device. The difference from the specific example described using Figure 10 is that the display screen generation unit 1410 in Figure 14 has a level calculation unit 1411.

[0145] The level calculation unit 1411 obtains the 95% confidence interval (converted to characteristic data) for the prediction data from the second scale conversion unit 333 and calculates the width of the 95% confidence interval (converted to characteristic data) for the prediction data.

[0146] Furthermore, the level calculation unit 1411 obtains the maximum and minimum widths (converted to characteristic data) of the 95% confidence interval calculated when the Gaussian process regression model 131 was trained using the training data 121 from the learning device 130 as the worst and best values.

[0147] Furthermore, the level calculation unit 1411 compares the width of the 95% confidence interval (converted to characteristic data) for the predicted data with the worst and best values ​​obtained from the learning device 130. Based on this, the level calculation unit 1411 calculates the normality level of the width of the 95% confidence interval (converted to characteristic data) for the predicted data.

[0148] In Figure 14, data table 1421 is an example of a table containing levels calculated by the level calculation unit 1411.

[0149] In data table 1421, the "best value" is the minimum width of the 95% confidence interval (converted to characteristic data) obtained from the learning device 130, which in the example in Figure 14 is "0.016".

[0150] Furthermore, in data table 1421, the "worst-case value" is the maximum width of the 95% confidence interval (converted to characteristic data) obtained from the learning device 130, which in the example in Figure 14 is "0.298".

[0151] Furthermore, in data table 1421, the "predicted value" is the width of the 95% confidence interval (converted to characteristic data) for the predicted data, which in the example in Figure 14 is "0.083".

[0152] Furthermore, in data table 1421, "Level" represents the normality level of the 95% confidence interval width (converted to characteristic data) for the predicted data. In the example in Figure 14, the normality level is {(Predicted value ("0.083") - Best value ("0.016")) / (Worst value ("0.298") - Best value ("0.016"))} × 100 = 24.

[0153] <Flowchart of the prediction data display process in the prediction data display system during the prediction phase> Next, we will explain the flow of the prediction data display process in the prediction data display system 300 during the prediction phase. Figure 15 is a second flowchart showing the flow of the prediction data display process. The difference from the flowchart explained using Figure 13 is in steps S1501, S1502-S1504.

[0154] In step S1501, the prediction data display device 330 obtains the maximum and minimum widths (converted to characteristic data) of the 95% confidence interval calculated when the Gaussian process regression model was trained during the training phase, as the worst and best values.

[0155] In step S1502, the predictive data display device 330 compares the 95% confidence interval (converted to characteristic data) obtained in step S1309 with the worst and best values ​​obtained in step S1501. Based on this, the predictive data display device 330 calculates the level of normality of the width of the 95% confidence interval (converted to characteristic data) for the predictive data.

[0156] In step S1503, the predictive data display device 330 generates a text table that associates the similarity with the type of consumable and the manufacturing conditions. The predictive data display device 330 also generates a data table that includes the calculated levels.

[0157] In step S1504, the predictive data display device 330 generates and displays a display screen that includes a combined graph, a text table, and a data table.

[0158] <Summary> As is clear from the above description, the predictive data display system according to the second embodiment is - The width of the 95% confidence interval when predicting data for the consumables to be predicted is compared with the maximum and minimum widths of the 95% confidence interval when training is performed using the training data, and the level of accuracy of the width of the 95% confidence interval when predicting data is calculated.

[0159] This makes it possible to quantify the level of accuracy of the calculated prediction data according to the second embodiment. As a result, users can more easily determine whether the prediction data is correct or incorrect.

[0160] In other words, according to this embodiment, when the degree of deterioration of consumables is predicted using a machine learning model, it becomes easier to determine whether the prediction data is accurate or not.

[0161] [Third Embodiment] In the second embodiment described above, the display screen showed a combined graph, a text table, and a data table. In contrast, in the third embodiment, the data in the data table is displayed as a box plot on the display screen. The third embodiment will now be described, focusing on the differences from the second embodiment.

[0162] <Specific example of processing in the display screen generation unit of the predictive data display device> First, a specific example of the processing of the display screen generation unit of the predictive data display device 330 will be described. Figure 16 is the fourth figure showing a specific example of the processing of the display screen generation unit of the predictive data display device. As shown in Figure 16, the display screen generation unit 1610 has a level calculation unit 1411.

[0163] The function of the level calculation unit 1411 has already been explained using Figure 14 in the second embodiment described above, so its explanation will be omitted here. However, in the case of the level calculation unit 1411 shown in Figure 16, there are multiple consumables to be predicted (five in the example in Figure 16), and multiple 95% confidence intervals (converted to characteristic data) are obtained from the second scale conversion unit 333. In addition, in the case of the level calculation unit 1411 shown in Figure 16, the width of the 95% confidence interval (converted to characteristic data) is calculated for each of the multiple 95% confidence intervals (converted to characteristic data) obtained.

[0164] Furthermore, in the case of the level calculation unit 1411 shown in Figure 16, the normality level is calculated by comparing each of the calculated widths of the multiple 95% confidence intervals (converted to characteristic data) with the worst-case and best-case values.

[0165] Furthermore, in the case of the level calculation unit 1411 shown in Figure 16, a box plot is generated based on each of the calculated levels, as well as the worst and best values, and displayed on the display screen 1620.

[0166] In the display screen 1620, the box plot 1621 represents the level calculated based on the width of the 95% confidence interval (converted to characteristic data) when predicting data for the consumable product N1 that is the target of prediction.

[0167] Similarly, in display screen 1620, box plots 1622-1625 represent the levels calculated based on the width of the 95% confidence interval (converted to characteristic data) when predicting data for consumables N2-N5.

[0168] In this way, when prediction data is made for multiple consumable items, displaying box plots showing the levels on a single screen allows the user to easily determine the accuracy of the prediction data for those multiple consumable items.

[0169] <Summary> As is clear from the above explanation, the predictive data display system according to the third embodiment is This method compares the width of the 95% confidence interval (CVT) when predicting data for multiple consumable items with the maximum and minimum CVT widths when training data is used. This allows for the calculation of the accuracy level of the CVT widths when predicting multiple data points. • The multiple levels calculated are displayed as a box plot, arranged on a single screen.

[0170] As a result, according to the third embodiment, the user can view a list of the accuracy levels of the calculated prediction data, and can easily determine whether the prediction data is correct or incorrect.

[0171] In other words, according to this embodiment, when the degree of deterioration of multiple consumables is predicted using a machine learning model, it becomes easier to determine whether the multiple prediction data are correct or incorrect.

[0172] [Other embodiments] In the embodiments described above, the learning device and the prediction device were described as separate devices. However, the learning device and the prediction device may be configured as a single integrated device. Similarly, in the embodiments described above, the learning data generation device and the learning device were described as separate devices. However, the learning data generation device and the learning device may be configured as a single integrated device.

[0173] Furthermore, although the similarity calculation unit was described as a part of the trained Gaussian process regression model in each of the above embodiments, it may be implemented as a separate functional unit from the trained Gaussian process regression model. Also, the method of calculating similarity by the similarity calculation unit is arbitrary, and any function that increases similarity when the difference value of the features is small can be used for calculation.

[0174] Furthermore, in each of the above embodiments, the battery discharge capacity retention rate was used as an example of characteristic data indicating the degree of deterioration of consumables, but the characteristic data indicating the degree of deterioration of consumables may be other characteristic data besides the battery discharge capacity retention rate.

[0175] Furthermore, while batteries were used as examples of consumables in the above embodiments, consumables may be items other than batteries.

[0176] Furthermore, in each of the above embodiments, the damping ratio is r X = y(t X ) / y(t B Although the attenuation rate was calculated, the method for calculating the attenuation rate is not limited to this. For example, r X The attenuation rate obtained by converting using the following equation 3 may also be used.

[0177]

Number

[0178]

Number

[0179] Also, in each of the above embodiments, the usage scenario of the prediction data for the consumable to be predicted was not mentioned. However, for example, the prediction data for the consumable to be predicted may be reflected in the manufacturing conditions of the consumable to be predicted. Specifically, when the consumable to be predicted is a battery (for example, a lithium-ion secondary battery, etc.), the conditions such as activation and aging after the battery assembly is completed may be determined based on the prediction data.

[0180] Note that the present invention is not limited to the configurations and the like described in the above embodiments, such as combinations with other elements. Regarding these points, it is possible to make changes without departing from the spirit of the present invention, and it can be appropriately determined according to the application form.

[0181] This application claims priority based on Japanese Patent Application No. 2021-150217 filed on September 15, 2021, and incorporates the entire contents of the Japanese patent application by reference.

Explanation of Signs

[0182] 100: Prediction data display system 110: Characteristic measurement device 120: Learning data generation device 121: Training data 130: Learning device 131: Gaussian process regression model 300: Predictive Data Display System 320: Predictive data generation device 321: Predictive data 330: Predictive data display device 331: First scale conversion unit 332: Pre-trained Gaussian process regression model 333: Second scale conversion section 334:Display screen generation section 601: Damping rate calculation unit 602: Display Mode Change Section 701: Similarity calculation unit 702: Confidence interval calculation unit 703: Attenuation rate prediction unit 704: Model parameter storage unit 801: Prediction Data Calculation Unit 802: Confidence interval calculation unit 1001: Text data generation unit 1002: Graph connection section 1020:Display screen 1021: Text Table 1022: Joined Graph 1110:Display screen 1410:Display screen generation section 1411: Level calculation unit 1420 :Display screen 1421: Data Table 1610:Display screen generation section 1620:Display screen

Claims

1. A prediction step involves inputting information on a trained model, which has been trained using training data that takes information on first characteristic data indicating the degree of deterioration of a first consumable during a first period as input data, and information on second characteristic data indicating the degree of deterioration of the first consumable during a second period following the first period as ground truth data, information on third characteristic data indicating the degree of deterioration of the second consumable to be predicted during the first period, and calculating information on fourth characteristic data indicating the degree of deterioration of the second consumable to be predicted during the second period. A display step in which, when displaying the graph of the fourth characteristic data, the graph of the second characteristic data is displayed together in a display manner corresponding to the similarity between the information regarding the first characteristic data and the information regarding the third characteristic data, Based on the information regarding the fourth characteristic data calculated in the prediction step, a determination step is made to determine the manufacturing conditions of the second consumable product to be predicted. A method for manufacturing consumables, which is performed by a computer.

2. The method for manufacturing a consumable according to claim 1, wherein the second consumable to be predicted is a battery.

3. The method for manufacturing a consumable product according to claim 2, wherein the battery is a lithium-ion secondary battery.

4. The method for manufacturing a consumable product according to claim 3, wherein the manufacturing conditions include activation or aging conditions after the completion of assembly of the lithium-ion secondary battery.

Images