Method and system for predicting variation in properties of coating film, method and system for predicting variation in manufacturing conditions of coated object, and manufacturing method of coated object

Abstract

In this prediction method and prediction system for the amount of change in coating film properties, a prescribed artificial intelligence model is obtained by computer machine learning, using the amount of change in production conditions as an input and the amount of change in coating film properties as an output. The computer predicts the amount of change in coating film properties by using the amount of change in the production conditions as an input in the artificial intelligence model. In this prediction method and prediction system for the amount of change in production conditions, the optimal amount of change in production conditions is determined as the optimal amount of change in production conditions for achieving a target, on the basis of the predicted amount of change in a plurality of coating film properties. In this coated object production method, the production conditions are adjusted by the predicted amount of change in production conditions, a target object is painted, a coating film is obtained, and the properties of the coating film on the obtained coated object are measured.

Description

[Technical Field]

[0001] The present invention relates to a method and system for predicting the amount of variation in the properties of a coating film, a method and system for predicting the amount of variation in the manufacturing conditions of a coated object, and a method for manufacturing a coated object. [Background technology]

[0002] Conventionally, when applying paint to an object, coating is first performed under standard or previous conditions, and then repeated adjustments are made to achieve the desired coating film properties. The standard or previous conditions are typically recorded using past data, such as a work management chart, and referenced to achieve the desired coating film properties. This involves a sequential, manual process. This process requires the coating manager to manually adjust production conditions such as deposition pressure, line speed, dispensing pressure, voltage, and temperature. This process is repeated until the target film thickness and the color and gloss differences between the target coated plate and the original are within a certain range, resulting in a coating film with the desired coating film properties.

[0003] The above-mentioned adjustment work is largely dependent on the experience of the paint manager and is a complicated process. Aiming to automate such repetitive adjustment work, it has been proposed to predict the numerical values ​​of the coating film properties using machine learning methods when evaluating the properties of the coating film obtained by applying a paint to an object (Patent Documents 1 to 8). This makes it possible to adjust the manufacturing conditions based on the predicted values. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2020-039992 [Patent Document 2] Japanese Patent Application Laid-Open No. 2001-078627 [Patent Document 3] Japanese Patent Application Laid-Open No. 2000-219097 [Patent Document 4] Japanese Patent Application Laid-Open No. 1999-149118 [Patent Document 5] Japanese Patent Application Laid-Open No. 1993-300386 [Patent Document 6] Japanese Patent Application Laid-Open No. 1991-070565 [Patent Document 7] Japanese Patent Application Laid-Open No. 2016-501119 [Patent Document 8] International Publication No. 2019 / 171498 Brochure Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the methods of Patent Documents 1 to 8, etc., if the calculated predicted value is within the range of the quality specification value, it is determined that there is no need to adjust the manufacturing conditions even if there is a discrepancy between the predicted value and the actual measured value. As a result, the manufacturing conditions are not appropriately adjusted, and as a result, it may not be possible to obtain a coated product with the desired coating film quality.

[0006] Therefore, an object of the present invention is to provide a method and system for predicting the amount of variation in the properties of a coating film, which can quickly predict the amount of variation in the properties of a coating film; a method and system for predicting the amount of variation in the production conditions of a coated product, which allows the production conditions to be adjusted regardless of the range of quality specification values; and a method for producing a coated product, which can produce a coated product having the desired coating film quality. [Means for solving the problem]

[0007] The gist and configuration of the present invention are as follows. (1) A method for predicting the amount of variation in the properties of a coating film obtained by applying a paint to an object, comprising: an artificial intelligence model acquisition step of acquiring a predetermined artificial intelligence model that includes a variation amount of a manufacturing condition as an input and has a variation amount of the properties of the coating film as an output; A method for predicting the amount of variation in the properties of a coating film, comprising: a property variation prediction step in which a computer predicts the amount of variation in the properties of the coating film by inputting the amount of variation in the manufacturing conditions into the artificial intelligence model.

[0008] (2) A method for predicting the optimal variation amount of the production conditions to achieve a target when a coated object is produced by applying a paint to an object and adjusting the production conditions to produce a coated object having target properties of a coating film of the coated object, comprising: an artificial intelligence model acquisition step of acquiring a predetermined artificial intelligence model that includes a variation amount of a manufacturing condition as an input and has a variation amount of the properties of the coating film as an output; a property variation prediction step of predicting a variation in the properties of the coating film by inputting the variation in the manufacturing conditions into the artificial intelligence model by a computer, In the property variation prediction step, a variation amount of each of the plurality of manufacturing conditions is input, and a variation amount of each of the plurality of coating film properties corresponding to the variation amount of each of the manufacturing conditions is predicted; A method for predicting a variation amount of a manufacturing condition, characterized by further comprising a manufacturing condition variation prediction step of determining an optimal variation amount of the manufacturing condition as the optimal variation amount of the manufacturing condition for achieving the target, based on the predicted variation amount of the properties of the plurality of coating films.

[0009] (3) further comprising a relational data input step of inputting relational data indicating the relationship between the amount of variation in the manufacturing conditions and the amount of variation in the properties of the coating film into the computer; The method described in (1) or (2) above, wherein the artificial intelligence model acquisition process is performed by the computer performing machine learning using the relational data input in the relational data input process as training data.

[0010] (4) The method according to any one of (1) to (3) above, wherein the properties of the coating film include one or more of the color, gloss, viscosity, film thickness, smoothness, visual appearance, flip-flop property, and coating film abnormality of the coating film.

[0011] (5) The manufacturing conditions are: (a) one or more of the viscosity and heating residue of the paint; (b) one or more of the roll peripheral speed, the coating pressure on the object, and the flow rate of the coating material in the process of applying the coating material to the object; (c) one or more of the baking temperature and baking time in the baking step; and (d) one or more of the temperature, humidity, and paint liquid temperature of the production line; The method according to any one of claims 1 to 4, comprising any one or more of (a) to (d).

[0012] (6) A method according to any one of (1) to (5) above, wherein the artificial intelligence model acquisition process is a process of creating the artificial intelligence model by machine learning, and the machine learning uses a neural network, an ensemble tree consisting of multiple decision trees, or a prediction algorithm using partial least squares regression.

[0013] (7) The method described in (6) above, wherein the machine learning uses a neural network, a random forest, or a gradient boosting method.

[0014] (8) The method according to any one of claims 1 to 7, further comprising the step of performing data cleansing on the relational data after the relational data input step.

[0015] (9) The method described in (2) above, wherein in the manufacturing condition variation prediction step, the manufacturing condition variation corresponding to the predicted variation in the properties of the coating film that has the smallest difference from the target property is determined as the manufacturing condition variation that is optimal for achieving the target.

[0016] (10) a post-conditions-adjustment manufacturing process, in which the manufacturing conditions are adjusted by the amount of fluctuation in the manufacturing conditions predicted by the method described in (2) above, and the coating material is applied to the object to obtain the coating film; and a property measuring step of measuring the properties of the coating film of the obtained coated object, A method for manufacturing a coated object, wherein, when the properties of the coating film measured in the property measurement process deviate from desired properties by a predetermined threshold value or more, the property variation prediction process, the manufacturing condition variation prediction process, the manufacturing process after condition adjustment, and the property measurement process are repeated.

[0017] (11) When the property of the coating film measured in the property measuring step deviates from the desired property by a predetermined threshold value or more a predetermined number of times, an artificial intelligence model acquisition sub-process of acquiring another artificial intelligence model that includes the amount of variation in the manufacturing conditions as an input and has the amount of variation in the properties of the coating film as an output; a property variation prediction sub-step of predicting a property variation amount of the coating film by inputting a variation amount of the manufacturing condition into the other artificial intelligence model by the computer, In the property variation prediction sub-step, variations of a plurality of the manufacturing conditions selected in advance as candidates are input, and variations of the properties of a plurality of the coating films corresponding to the variations of each of the manufacturing conditions are predicted; a manufacturing condition variation prediction sub-step of determining another optimum manufacturing condition variation amount as the other manufacturing condition variation amount optimum for achieving the target, based on the predicted variation amounts of the properties of the plurality of coating films; a post-condition-adjustment manufacturing sub-process of adjusting the manufacturing conditions by the predicted amount of variation in the other manufacturing conditions and applying the paint to the object to obtain the coated object; a property measurement sub-step of measuring the properties of the coating film of the obtained coated object, The method for manufacturing a coated object described in (10) above, wherein, when the properties of the coating film measured in the property measurement process deviate from the desired properties by a predetermined threshold or more, the artificial intelligence model acquisition sub-process, the property variation prediction sub-process, the manufacturing condition variation prediction sub-process, the manufacturing sub-process after condition adjustment, and the property measurement sub-process are repeated.

[0018] (12) The method for producing a coated object according to (10) above, further comprising a relational data update step of inputting relational data indicating the relationship between the amount of variation in the production conditions and the amount of variation in the properties of the coating film measured in the property measurement step into the computer, and updating the relational data.

[0019] (13) A system for predicting the amount of variation in the properties of a coating film obtained by applying a paint to an object, an artificial intelligence model acquisition unit that acquires a predetermined artificial intelligence model that includes a variation amount of a manufacturing condition as an input and has a variation amount of the properties of the coating film as an output; The system is characterized by comprising a computer having a property prediction unit that calculates and predicts the amount of variation in the properties of the coating film by inputting the amount of variation in the manufacturing conditions in the artificial intelligence model.

[0020] (14) A system for predicting the optimal variation amount of the manufacturing conditions to achieve a target when manufacturing a coated object by applying paint to an object and adjusting the manufacturing conditions to manufacture a coated object having target properties of a coating film of the coated object, the system comprising: an artificial intelligence model acquisition unit that acquires a predetermined artificial intelligence model that includes the amount of variation in the manufacturing conditions as an input and has the amount of variation in the properties of the coating film as an output; a property prediction unit that calculates and predicts variations in the properties of the coating films by inputting variations in the manufacturing conditions in the artificial intelligence model; A system characterized by comprising a computer having a manufacturing condition fluctuation prediction unit that determines optimal manufacturing condition fluctuation amounts as the optimal manufacturing condition fluctuation amounts for achieving the target based on the predicted fluctuation amounts of the properties of the plurality of coating films.

[0021] (15) The manufacturing conditions include process manufacturing conditions for each of a plurality of processes, monitoring each of the production conditions when adjusting the production conditions by the amount of fluctuation of the production conditions predicted by the method described in (2) above; When any of the manufacturing conditions is observed to be deviated from the fluctuation amount by a predetermined value or more, a property re-prediction step in which the monitored process manufacturing conditions are fixed in the artificial intelligence model, and then fluctuation amounts of other multiple manufacturing conditions are input to calculate and predict fluctuation amounts of the properties of the multiple coating films; a manufacturing condition variation re-prediction step of re-determining optimal manufacturing condition variation amounts as optimal manufacturing condition variation amounts for achieving the target, based on the predicted variation amounts of properties of the plurality of coating films, The manufacturing method according to (11) above, wherein the fluctuation amounts of the manufacturing conditions in processes subsequent to the current process are readjusted based on the results of the manufacturing condition fluctuation amount re-prediction process.

[0022] Note that, during learning in the above-mentioned variation prediction method, etc., a mode in which a fixed, representative absolute value of the paint film properties before adjustment is added to the variation amount of the paint film properties before and after adjustment obtained by measurement, and the representative absolute value after variation is used for learning, can also be included as one aspect of the above-mentioned method.Furthermore, during output of the prediction result, a predicted value of the variation amount of the paint film properties obtained by the above-mentioned technique can be obtained, and then added to the fixed, representative absolute value of the paint film properties before adjustment, to indirectly obtain the absolute value of the paint film properties, can also be included as one aspect of the above-mentioned method. [Effects of the Invention]

[0023] According to the present invention, it is possible to provide a method and system for predicting the amount of variation in the properties of a coating film, which can quickly predict the amount of variation in the properties of a coating film; a method and system for predicting the amount of variation in the production conditions of a coated product, which allows the production conditions to be adjusted regardless of the range of quality specification values; and a method for producing a coated product, which can produce a coated product having the desired coating film quality. [Brief explanation of the drawings]

[0024] [Figure 1] 1 is a flowchart of a method for predicting the amount of variation in properties of a coating film according to one embodiment of the present invention. [Figure 2] 1 is a flowchart of a method for predicting the amount of fluctuation in manufacturing conditions according to one embodiment of the present invention. [Figure 3] 1 is a flowchart of a method for producing a coated object according to one embodiment of the present invention. [Figure 4] 10 is a flowchart showing a subflow of a method for producing a coated object according to one embodiment of the present invention. [Figure 5] Schematic diagram of a CCL. [Figure 6] 1 is a block diagram of a system for predicting the amount of variation in the properties of a coating film according to one embodiment of the present invention. [Figure 7] 1 is a block diagram of a system for predicting fluctuation amounts of manufacturing conditions according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0025] First, we will explain the technology related to the prediction method of the present invention. The tolerance for paint film properties varies depending on the application. For example, in automotive refinishing, the boundary between the normally painted area of ​​a damaged vehicle (corresponding to the target painted panel) and the area to be painted with a new paint prepared for the repair after repair can be blurred (graded). This means that color matching, such as ΔE<0.5, is not strictly required. On the other hand, in the case of coil coating, which is used for pre-coating (painting before assembly), painted panels from different production lots may be assembled side by side during product manufacturing. Therefore, the tolerance for color differences due to differences in production lots is extremely small, for example, ΔE<0.1. Because this color difference level varies depending on the colorimeter model and conditions, absolute color values ​​are insufficient for accuracy. Therefore, a standard panel (the target painted panel) must be prepared with the customer's consent. The color difference between the standard panel and the target painted panel must be measured at the same time under the same conditions for each production run and each color adjustment, and adjustments must be made to keep the color difference within the specified range.

[0026] The adjustment work described above is highly dependent on the experience of the paint manager and is a complicated process, so there is a demand for automation of the adjustment work. However, when using artificial intelligence to learn the relationship between the manufacturing conditions and the absolute values ​​of the coating film properties, even a slight deviation in the prediction accuracy tends to result in the wrong direction of correction. For example, in the case of color, if the b of the target color * Value 3.0, true b of the color of the paint before adjustment * If the b value of the paint color before adjustment predicted from the manufacturing conditions is 2.9, and the b value is 3.1 (i.e., the error is 0.2), then the actual b * To raise the value by 0.1, the temperature of the drying oven should be increased, but the absolute value prediction was over 0.1 (too yellow), so the temperature of the drying oven was lowered. With absolute value prediction, adjustments could actually move the color away from the target color, making fine adjustments impossible in some cases.

[0027] Even when the manufacturing conditions are the same, slight differences in paint properties, changes in the equipment used over time, and other factors that are difficult to quantify, difficult to measure, require a large amount of measurement time, and result in a small amount of data (high data costs) are all factors that machine learning cannot fully learn, resulting in slight variations in coating film properties. This is one of the causes of prediction errors using the aforementioned absolute value prediction artificial intelligence. Therefore, for example, methods such as those described in Patent Documents 1 to 8 may have difficulty making accurate predictions in cases such as those described above.

[0028] Therefore, an object of the present invention is to provide a method and system for predicting the amount of variation in the properties of a coating film, which achieves efficiency through machine learning techniques while providing high prediction accuracy, a method and system for predicting the amount of variation in the manufacturing conditions of a coated object, a method for manufacturing a coated object, and a method for manufacturing a coating film, which achieves efficiency through machine learning techniques while providing high prediction accuracy, and which can accurately bring the coating film properties closer to those desired.

[0029] By predicting the amount of variation, even if the manufacturing conditions at the time of prediction have changed from the manufacturing conditions when the artificial intelligence was created, resulting in an error in the absolute value prediction, if the amount of variation has been learned, such as in roll coating, increasing the coating pressure by 50 kgf under certain manufacturing conditions reduces the film thickness by 1.5 μm, the effects of unintentional changes in manufacturing conditions will be similar before and after intentional changes in the manufacturing conditions, so the effects will be offset and the artificial intelligence will be able to practically apply to fine adjustments.

[0030] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0031] <Method for predicting the amount of variation in coating properties> FIG. 1 is a flowchart of a method for predicting the amount of variation in the properties of a paint film according to one embodiment of the present invention. Hereinafter, with reference to FIG. 1, an exemplary embodiment of a method for predicting the amount of variation in the properties of a paint film obtained by applying a paint to an object will be described. Note that the method for predicting the amount of variation in the properties of a paint film according to this embodiment can be implemented, for example, using a system for predicting the amount of variation in the properties of a paint film according to one embodiment of the present invention, which will be described later. The predicted paint film properties preferably include at least the color of the paint, and also preferably include the gloss and film thickness of the paint in addition to the color. Note that the predicted paint film properties preferably include, in addition to color, gloss, and film thickness, one or more of the following: hiding power, flip-flop property, smoothness, visual appearance, electrical resistance, contact angle, staining resistance, solar reflectance, UV transmittance, weather resistance, viscoelasticity, and paint film abnormality. In particular, adjusting the coating pressure to change the film thickness changes both the color and gloss, and adding a gloss adjuster to adjust the gloss also changes the color. Therefore, it is preferable to simultaneously predict and adjust the color, gloss, and film thickness. The color is, for example, L * a * b * L in color space * value, a * value, b *The color can be measured using a known color measurement method, for example, by using a CM-512m3 commercially available from Konica Minolta, Inc., and measuring the L value by irradiating a light source at angles of 25°, 45°, and 75°, assuming that the light receiving part perpendicular to the coating is at 0°. * value, a * value, b * The value can be measured. Alternatively, it can be measured using an X-Rite MA68II (manufactured by X-Rite). The measurement angle can be adjusted appropriately depending on the purpose or the equipment used. Any other index can be used. Furthermore, for example, any index can be used, such as reflection spectrum data, such as an index in which the reflection spectrum intensity at 5 nm intervals from 380 nm to 780 nm is used as a color index. The gloss is not particularly limited, but gloss can be used as an index. Gloss can be measured using a known gloss measurement method. For example, the 60° gloss of a coating film formed on a test plate can be measured using a specular gloss meter (VG 7000 gloss meter (manufactured by Nippon Denshoku Industries Co., Ltd.)) in accordance with JIS K 5600-4-7 (specular gloss). Viscosity can be measured using a known viscosity measurement method. For example, it can be measured in accordance with JIS K 5600-2-2 (flow cup method). It is preferable to use the wave scan value as an index of smoothness. The wavescan value is preferably one or more of du (wavelength 0.1 mm or less), Wa (wavelength 0.1 to 0.3 mm), Wb (wavelength 0.3 to 1.0 mm), Wc (wavelength 1.0 to 3.0 mm), Wd (wavelength 3.0 to 10.0 mm), We (wavelength 10.0 to 30.0 mm), Lw (wavelength 1.2 to 12 mm), and Sw (wavelength 0.3 to 1.2 mm). The smaller the wavescan value, the less unevenness at that wavelength on the surface, indicating better appearance quality of the coating film. Flip-flop properties can be measured using a variable-angle colorimeter or the like. Specific examples of coating abnormalities include unevenness, blisters, cracks, sagging, pinholes, and frame defects. These can also be measured using known methods.

[0032] The paint used for painting can be one that is normally used for that type of painting, but when preparing a paint by preparation, for example, pigments, resins, additives, and solvents and / or water can be dispersed in an SG mill or the like, and resins, solvents, and additives can be added to primary color paint prepared by preparing a large number of types of pigments, etc., and dispersed, and then the preparation can be carried out by repeating the process.

[0033] 1, in this embodiment, first, relational data showing the relationship between the amount of variation in the manufacturing conditions and the amount of variation in the properties of the coating film is input to a computer (step S101: relational data input step). The relational data can be prepared from past data, etc.

[0034] Here, the manufacturing conditions are those related to applying a coating to an object to obtain a coating film. For example, the manufacturing conditions can be one or more characteristics of the coating. Specifically, the manufacturing conditions are preferably one or more of the viscosity and heating residue of the coating. As another example, the manufacturing conditions can be the conditions for each step in the process of applying a coating to an object. Specifically, in the case of roll coating, the manufacturing conditions include the line speed, roll peripheral speed, coating pressure and coating flow rate, the condition, material, hardness, and type of applicator roll, and the type of pickup roll. Preferably, the manufacturing conditions are one or more of the roll peripheral speed, coating pressure on the object, and coating flow rate. Here, the roll in the roll peripheral speed refers to, but is not limited to, an application roll, a backup roll, a pickup roll, a metering roll, and the like, with the application roll being particularly preferred. As another example, the manufacturing conditions can be the conditions for a baking process or other process performed between steps. For example, the manufacturing conditions include the baking temperature, baking time, and maximum temperature reached in the baking process. Of these, the manufacturing conditions are preferably one or more of the baking temperature and baking time. As yet another example, the manufacturing conditions can be environmental conditions on the manufacturing line, such as one or more of the temperature, humidity, and paint liquid temperature of the manufacturing line. Other examples include plate thickness, painting line, primer and topcoat adhesion amounts, color difference meter, physical property data such as primer type, color, roughness, and type, and chemical property data. In the case of spray painting, examples include discharge pressure, discharge volume, gun type, baking temperature, baking time, temperature, and humidity in the baking process. In the case of electrocoating, examples include painting voltage, liquid temperature, current application time, and painting direction (horizontal or vertical surface).

[0035] In summary, for example, in the case of roll coating, the manufacturing conditions can include one or more of (a) the viscosity and heating residue of the paint, (b) one or more of the roll peripheral speed, the application pressure on the object, and the paint flow rate in the process of applying the paint to the object, (c) one or more of the baking temperature and the baking time in the baking process, and (d) one or more of the temperature, humidity, and paint liquid temperature of the production line.

[0036] It is preferable that the above-mentioned relationship data be updated constantly, in a timely manner, or periodically to enable more accurate predictions. Furthermore, it is preferable to further include a step of performing data cleansing on the relational data after the relational data input step (step S101). Here, the prepared relational data is subjected to processing such as data normalization / standardization and new data generation, as well as data filtering to remove inappropriate data. For example, data that exceeds the average value ±2σ (where σ is the standard deviation) of the difference between the actual and predicted values ​​of coating film quality can be carefully examined, and data suspected of input errors or typographical errors can be deleted. Such data cleansing and normalization can prevent overlearning in the machine learning described below, enabling more accurate prediction of paint properties. However, in the present disclosure, data cleansing is not required, and this step can be omitted. Data cleansing can be applied commonly to multiple or all algorithms (e.g., removing outliers that are deemed inappropriate by all machine learning algorithms), or it can be applied exclusively to the machine learning algorithm actually used (e.g., removing data that is likely to cause errors in a specific algorithm). The artificial intelligence model creation step described below can also be created by re-learning the data after this data cleansing.

[0037] Next, as shown in FIG. 1 , a predetermined artificial intelligence model is created by computer-based machine learning, using the variation in manufacturing conditions as input and the variation in the properties of the coating film as output (step S102: artificial intelligence model creation step). In this example, this artificial intelligence model creation step is performed by computer-based machine learning using the relational data input in the relational data input step (step S101) as training data. The input relational data may be only the variation in manufacturing conditions, or may be the numerical values ​​and variation in manufacturing conditions. These may be measured values ​​or set values. The output may be measured values ​​of the properties of the coating film, and the difference between the measured values ​​of the properties of the coating film before and after a change in manufacturing conditions may be calculated to determine the variation. While the artificial intelligence model is created in this step in this embodiment, it is not necessary to create the artificial intelligence model; the created artificial intelligence model can also be acquired externally. For example, the artificial intelligence model can be acquired via a computer's communication unit, or it can be acquired by a human being through transfer or the like (artificial intelligence model acquisition step).

[0038] The machine learning algorithm may be any known algorithm, such as a decision tree, linear regression, partial least squares regression, lasso regression, ridge regression, polynomial regression, Gaussian process regression, support vector machine, random forest, gradient boosting, K nearest neighbors, neural network, Bayesian estimation, or a prediction algorithm based on ensemble learning of these. It is particularly preferable that the machine learning uses a prediction algorithm based on a neural network. The neural network may be a convolutional neural network or a deep learning network that forms 3 to several hundred layers while performing dropout. Additionally, machine learning frameworks such as TensorFlow, Keras, Caffe, PyTorch, Chainer, and Scikit-learn may be used to create artificial intelligence.

[0039] Here, the machine learning can be supervised learning (including reinforcement learning). In this embodiment, the artificial intelligence model creation step (step S103) is performed by a computer by performing machine learning (supervised learning) using the relational data input in the relational data input step (step S102) as learning data (supervised data in this example).

[0040] Here, the algorithm is preferably determined by the following method. Specifically, multiple artificial intelligence models are created using various machine learning algorithms, and the accuracy of each artificial intelligence model is confirmed by comparing the predicted output values ​​with the answers using data for which the answers are known in advance. Then, for the algorithm with the highest accuracy among the multiple algorithms, optimal hyperparameters are determined, for example, by creating a model using cross-validation or by using Bayesian optimization. By relearning all data using the determined hyperparameters, an artificial intelligence model can be created that uses the amount of variation in manufacturing conditions as input and the amount of variation in the coating film properties after adjustment as output.

[0041] In this embodiment, the prepared relational data is preprocessed. Here, the prepared relational data is subjected to processing such as data normalization / standardization, taking logarithms, multiplying by powers (power processing), processing such as generating new data, and data filtering to remove inappropriate data. For example, data showing a large difference between the measured and predicted values ​​of the coating film properties can be closely examined, and data suspected of input errors or entry errors can be deleted. As described above, in this embodiment, various machine learning algorithms can be used, and the above preprocessing can be applied commonly to multiple or all algorithms (for example, removing data such as outliers that are deemed inappropriate for any machine learning algorithm), or can be applied exclusively to the machine learning algorithm actually used (for example, removing data that is likely to cause errors in a specific algorithm).

[0042] Furthermore, in the preprocessing, it is preferable to add zero-point relation data indicating that when the amount of variation in the manufacturing conditions is zero, the amount of change in the coating film properties due to the adjustment is zero (zero point addition). For example, if the data used for machine learning is data recording adjustments made by a line manager, it will not include data indicating that "manufacturing conditions are not varied." This raises the concern that the results obtained by machine learning will not pass through the point where the amount of change in the coating film properties is zero when the manufacturing conditions are not varied, thereby reducing the accuracy of the prediction. Therefore, by adding the above-mentioned zero-point relation data, it is possible to avoid such problems and further improve the accuracy of the prediction. Here, it is also possible to replace zero with a number close to zero.

[0043] Furthermore, when recording consecutive adjustments, the data can be amplified by adding them together into a single adjustment record (combination). For example, in the first and second adjustment records, the first record shows that the deposition pressure was increased by 50 kgf, resulting in a 1.2 μm decrease in film thickness, and the second record shows that the deposition pressure was increased by 100 kgf, resulting in a further 2.6 μm decrease in film thickness. These records are then added together to create data showing that the deposition pressure was increased by 150 kgf, resulting in a 3.8 μm decrease in film thickness. This can further improve prediction accuracy by both learning from significantly fluctuating data and by increasing the amount of data.

[0044] As a preprocessing step, estimated data can be added between data. For example, an approximate line or approximate curve can be calculated from the data, and the points on the approximate line or approximate curve can be used as estimated data. This increases the amount of data and can further improve the accuracy of prediction. Alternatively, past data can be connected to the current data by a line or curve, and any number of data can be added to the line.

[0045] Such preprocessing and normalization enable more accurate prediction of coating film properties in the machine learning described below. However, in the present disclosure, preprocessing is not essential, and this step can be omitted. A predetermined artificial intelligence model can be created using the preprocessed data in a similar manner to that described above.

[0046] Next, in this embodiment, the computer predicts the amount of variation in the properties of the coating film by inputting the amount of variation in the manufacturing conditions into the artificial intelligence model (step S103: property variation prediction step). Note that when the input of the relational data includes not only the amount of variation in the manufacturing conditions but also the numerical values ​​of the manufacturing conditions, it is preferable to predict the amount of variation in the properties of the coating film by inputting the numerical values ​​and the amount of variation in the manufacturing conditions into the artificial intelligence model.

[0047] In the property variation amount prediction step (step S103), it is preferable to narrow down the range of manufacturing conditions to be input in advance by referring to past data, for example.

[0048] In this way, by inputting the amount of variation in the manufacturing conditions into the above-mentioned artificial intelligence model, the amount of variation in the properties of the coating film can be calculated by a computer. The calculated predicted value of the amount of variation in the properties of the coating film may be a single value or may consist of a group of multiple candidates. When the calculated predicted value of the amount of variation in the properties of the coating film consists of a group of multiple candidates, it is preferable to further include a step of appropriately selecting a single value from the group using a predetermined criterion. The predetermined criterion can be various.

[0049] According to the method for predicting the amount of variation in the properties of a coating film of this embodiment, a predicted value of the amount of variation in the properties of a coating film relative to the amount of variation in the manufacturing conditions can be quickly obtained by computer calculation.

[0050] <Method for predicting fluctuations in manufacturing conditions> FIG. 2 is a flowchart of a method for predicting the amount of variation in manufacturing conditions according to one embodiment of the present invention. Hereinafter, with reference to FIG. 2, an embodiment of a method for predicting the amount of variation in manufacturing conditions that is optimal for achieving a target when manufacturing a coated object having target properties for the coating film of the coated object by varying and adjusting the manufacturing conditions when applying paint to the target object will be described. Note that the method for predicting the amount of variation in manufacturing conditions according to this embodiment can be implemented, for example, by using a system for predicting the amount of variation in manufacturing conditions according to one embodiment of the present invention, which will be described later. Here, as in the embodiment of FIG. 1, the properties of the coating film preferably include one or more of the color, gloss, viscosity, film thickness, smoothness, visual appearance, flip-flop property, and coating film abnormality of the coating film. The color may be, for example, L * a * b * L in color space * value, a * value, b * The color can be measured using a known color measurement method, for example, by using a CM-512m3 commercially available from Konica Minolta, Inc., and measuring the L value by irradiating a light source at angles of 25°, 45°, and 75°, assuming that the light receiving part perpendicular to the coating is at 0°. * value, a * value, b *The value can be measured. Alternatively, it can be measured using an X-Rite MA68II (manufactured by X-Rite). The measurement angle can be adjusted appropriately depending on the purpose or the equipment used. Any other index can be used. Furthermore, for example, any index can be used, such as reflection spectrum data, such as an index in which the reflection spectrum intensity at 5 nm intervals from 380 nm to 780 nm is used as a color index. The gloss is not particularly limited, but gloss can be used as an index. Gloss can be measured using a known gloss measurement method. For example, the 60° gloss of a coating film formed on a test plate can be measured using a specular gloss meter (VG 7000 gloss meter (manufactured by Nippon Denshoku Industries Co., Ltd.)) in accordance with JIS K 5600-4-7 (specular gloss). Viscosity can be measured using a known viscosity measurement method. For example, it can be measured in accordance with JIS K 5600-2-2 (flow cup method). It is preferable to use the wave scan value as an index of smoothness. The wavescan value is preferably one or more of du (wavelength 0.1 mm or less), Wa (wavelength 0.1 to 0.3 mm), Wb (wavelength 0.3 to 1.0 mm), Wc (wavelength 1.0 to 3.0 mm), Wd (wavelength 3.0 to 10.0 mm), We (wavelength 10.0 to 30.0 mm), Lw (wavelength 1.2 to 12 mm), and Sw (wavelength 0.3 to 1.2 mm). The smaller the wavescan value, the less unevenness at that wavelength on the surface, indicating better appearance quality of the coating film. Flip-flop properties can be measured using a variable-angle colorimeter or the like. Specific examples of coating abnormalities include unevenness, blisters, cracks, sagging, pinholes, and frame defects. These can also be measured using known methods. As with the embodiment of FIG. 1, the manufacturing conditions preferably include one or more of (a) the viscosity and heating residue of the paint, (b) one or more of the roll peripheral speed, the application pressure on the object, and the flow rate of the paint in the process of applying the paint to the object, (c) one or more of the baking temperature and the baking time in the baking process, and (d) one or more of the temperature and humidity of the manufacturing line. Further details of the above items are the same as those in the embodiment of FIG. 1, and therefore will not be described again.

[0051] As shown in Fig. 2, in this embodiment, first, relational data showing the relationship between the amount of variation in manufacturing conditions and the amount of variation in the properties of the coating film is input to a computer (step S201: relational data input step). Next, a predetermined artificial intelligence model is created by machine learning using the computer, which includes the amount of variation in manufacturing conditions as an input and uses the amount of variation in the properties of the coating film as an output (step S202: artificial intelligence model creation step). These steps S201 to S202 are the same as those in the embodiment of Fig. 1, so a detailed description will be omitted.

[0052] As shown in Fig. 2, in this embodiment, the computer then inputs the variations in the manufacturing conditions into the artificial intelligence model, thereby predicting the variations in the properties of the coating film (step S203: property variation prediction step). As in the embodiment of Fig. 1, if the input of relational data includes not only the variations in the manufacturing conditions but also the numerical values ​​of the manufacturing conditions, it is preferable to predict the variations in the properties of the coating film by inputting the numerical values ​​and variations of the manufacturing conditions into the artificial intelligence model. In the property variation prediction step (step S203), it is preferable to narrow down the range of manufacturing conditions to be input in advance, for example by referring to past data.

[0053] In the property variation prediction step (step S203) in the embodiment of Fig. 3, the variation amounts of a plurality of manufacturing conditions are input, and the variation amounts of the properties of a plurality of coating films corresponding to the variation amounts of each manufacturing condition are predicted. When selecting a plurality of manufacturing conditions in advance, for example, past data or the like can be referenced.

[0054] In this embodiment, next, based on the predicted variations in the properties of the plurality of coating films, the optimum variations in the manufacturing conditions are determined as the optimum variations in the manufacturing conditions for achieving the target (step S204: manufacturing condition variation prediction step). Specifically, as an example, in the manufacturing condition variation prediction step (step S204), the variation in the manufacturing conditions corresponding to the variation in the properties of the coating film that has the smallest difference from the target properties among the predicted variations in the properties of the plurality of coating films is determined as the optimum variation in the manufacturing conditions for achieving the target.

[0055] According to the method for predicting the amount of variation in the manufacturing conditions of this embodiment, when it is desired to vary the properties of a coating film by a predetermined amount, it is possible to obtain a predicted value of the amount of variation in the manufacturing conditions that is optimal. Therefore, it is sufficient to adjust the manufacturing conditions by the amount of the predicted value. Furthermore, because this adjustment can be performed regardless of the range of the quality specification value, it is possible to avoid the problem of not appropriately adjusting the manufacturing conditions despite a discrepancy between the predicted value and the actual measured value, as described above, resulting in an inability to obtain a coated product with the desired coating film quality.

[0056] Here, when preparing paint using artificial intelligence, if the amount of variation in the manufacturing conditions required is predicted based on the properties of the coating, such as color, there may not be a single solution, and it may be possible to present a variation in the manufacturing conditions that is far removed from the actual records (for example, it may be possible to obtain the desired color, but it may not be possible to adjust it in reality). Therefore, it is preferable to further include a step of setting a numerical value generation range of the variation for each manufacturing condition, and in the manufacturing condition variation prediction step (step S204), input a plurality of variations of the manufacturing conditions, which are formed by combining the numerical values ​​of the variation of each manufacturing condition generated within the set numerical value generation range. This allows for efficient prediction.

[0057] It is also preferable to determine the manufacturing condition variation amounts within an allowable range from the manufacturing condition variation amounts predicted in the manufacturing condition variation amount prediction step (step S204) as the manufacturing condition variation amounts for achieving the target. This allows the allowable range to be set in advance before predicting the manufacturing condition variation amounts, thereby reducing the possibility of generating unnecessary blends and efficiently obtaining more accurate prediction results for the manufacturing condition variation amounts.

[0058] The above-mentioned range of numerical values ​​can be determined based on a certain range from the standard values ​​or actual values ​​of the manufacturing conditions that have been set in advance, thereby making it possible to obtain predicted values ​​with small deviations from the standard values ​​or actual values.

[0059] Here, it is preferable that the variation amounts of the multiple manufacturing conditions input in the manufacturing condition variation amount prediction step (step S204) consist of candidates equal to or greater than the exponent of the number of manufacturing conditions. For example, when the number of manufacturing conditions is 4, 10 4 = 1,000 possible combinations. This 1,000 corresponds to the number of combinations possible when creating candidate manufacturing condition variations in 10 steps at 50 kgf intervals for a manufacturing condition with an allowable coating pressure variation of -200 kgf to +250 kgf, and then performing the same 10-step process for the other three manufacturing condition variations. When color differences are critical and even finer toning is required, the number of candidates for the 10 manufacturing condition variations is the number of raw materials + 1 (10,000 for four raw materials). Furthermore, in applications such as coil coating where color differences are particularly critical or the allowable range of manufacturing condition variations is large, predictions can be made that accommodate such fine toning by creating candidates for the number of manufacturing condition variations + 2 (100,000 for four variations).

[0060] It is also preferable to generate values ​​for the amount of variation for each manufacturing condition, randomly combine the generated values ​​to generate candidate variations for the manufacturing conditions, and predict the amount of variation for the manufacturing conditions required to achieve the target for the candidates. For example, when fine adjustment of coating film properties such as film thickness is required, if the range of variations in the manufacturing conditions is large, the number of brute-force combinations may reach one trillion (e.g., 10 levels for 12 materials, or 100 levels for 6 materials), making it difficult to obtain predicted values ​​and impractical. Therefore, by using random combinations such as those described above, the number of combinations can be reduced to 10 million, thereby shortening the calculation time required to obtain predicted values. The number of brute-force combinations may also be in the order of 100,000, 1 million, or 100 million. In the above embodiment, the numerical value itself may be randomly generated within the numerical value generation range.

[0061] Here, it is preferable to determine the amount of variation in the manufacturing conditions to achieve the target by adjusting the amount of variation in the manufacturing conditions to the standard conditions or previous conditions multiple times and performing a predetermined calculation on the coating film properties, or by performing a predetermined calculation on the amount of variation in the manufacturing conditions obtained using a gap target value with the target coating film properties as the reference value. When performing a predetermined calculation on the predicted amount of fluctuation in the manufacturing conditions, for example, the predicted value may be multiplied by a value such as 70%. As a first example of using a gap target value, it can be set at, for example, 70% of the target ΔL*. For example, if ΔL*=3 and the gap target value is set at 70% of the target ΔL*, three adjustments can be made, from 3 to 0.9 to 0.27 to 0.08, making the difference from the target value 0.1 or less. As a second example of using a gap target value, it can be set at 700% of the target ΔL*, and the resulting predicted value can be multiplied by 10%.

[0062] Furthermore, when using a coating material as a manufacturing condition, it is preferable to use a blending composition in an amount that is 3 to 99% of the amount of variation in the optimal manufacturing conditions to achieve the target, assuming that adjustments will be made multiple times by adding the coating material to adjust the coating film properties. This method also makes it possible to prevent over-addition due to variations in the coating material between lots, prediction errors, etc.

[0063] Furthermore, if it is known in advance that the prediction accuracy of large fluctuations is higher than that of small fluctuations, the target fluctuation value can be set to be larger than the difference from the target.

[0064] <Method of manufacturing the coated product> Fig. 3 is a flow chart of a method for producing a coated object according to one embodiment of the present invention. Hereinafter, referring to Fig. 3, an embodiment of a method for producing a coated object having target properties of the coating film of the coated object will be described by varying and adjusting the production conditions when producing the coated object by applying paint to the object. Here, as in the embodiment of Figs. 1 and 2, the properties of the coating film preferably include one or more of the color, gloss, viscosity, film thickness, smoothness, visual appearance, flip-flop property, and coating film abnormality of the coating film. The color can be, for example, L * a * b * L in color space * value, a * value, b * The color can be measured using a known color measurement method, for example, by using a CM-512m3 commercially available from Konica Minolta, Inc., and measuring the L value by irradiating a light source at angles of 25°, 45°, and 75°, assuming that the light receiving part perpendicular to the coating is at 0°. * value, a * value, b *The value can be measured. Alternatively, it can be measured using an X-Rite MA68II (manufactured by X-Rite). The measurement angle can be adjusted appropriately depending on the purpose or the equipment used. Any other index can be used. Furthermore, for example, any index can be used, such as reflection spectrum data, such as an index in which the reflection spectrum intensity at 5 nm intervals from 380 nm to 780 nm is used as a color index. The gloss is not particularly limited, but gloss can be used as an index. Gloss can be measured using a known gloss measurement method. For example, the 60° gloss of a coating film formed on a test plate can be measured using a specular gloss meter (VG 7000 gloss meter (manufactured by Nippon Denshoku Industries Co., Ltd.)) in accordance with JIS K 5600-4-7 (specular gloss). Viscosity can be measured using a known viscosity measurement method. For example, it can be measured in accordance with JIS K 5600-2-2 (flow cup method). It is preferable to use the wave scan value as an index of smoothness. The wavescan value is preferably one or more of du (wavelength 0.1 mm or less), Wa (wavelength 0.1 to 0.3 mm), Wb (wavelength 0.3 to 1.0 mm), Wc (wavelength 1.0 to 3.0 mm), Wd (wavelength 3.0 to 10.0 mm), We (wavelength 10.0 to 30.0 mm), Lw (wavelength 1.2 to 12 mm), and Sw (wavelength 0.3 to 1.2 mm). The smaller the wavescan value, the less unevenness at that wavelength on the surface, indicating better appearance quality of the coating film. Flip-flop properties can be measured using a variable-angle colorimeter or the like. Specific examples of coating abnormalities include unevenness, blisters, cracks, sagging, pinholes, and frame defects. These can also be measured using known methods. As with the embodiment of FIG. 1, the manufacturing conditions preferably include one or more of (a) the viscosity and heating residue of the paint, (b) one or more of the roll peripheral speed, the application pressure on the object, and the flow rate of the paint in the process of applying the paint to the object, (c) one or more of the baking temperature and the baking time in the baking process, and (d) one or more of the temperature and humidity of the manufacturing line. Further details of the above matters are the same as those in the embodiment shown in FIGS. 1 and 2, and therefore will not be described again.

[0065] As shown in FIG. 3 , in this embodiment, first, relational data showing the relationship between the amount of variation in the manufacturing conditions and the amount of variation in the properties of the coating film is input to a computer (step S301: relational data input step). Next, a predetermined artificial intelligence model is created using computer machine learning, which includes the amount of variation in the manufacturing conditions as an input and outputs the amount of variation in the properties of the coating film (step S302: artificial intelligence model creation step). Next, the amount of variation in the properties of the coating film is predicted by inputting the amount of variation in the manufacturing conditions into the artificial intelligence model using a computer (step S303: property variation prediction step). Next, based on the predicted amount of variation in the properties of the multiple coating films, optimal amounts of variation in the manufacturing conditions are determined as the amount of variation in the manufacturing conditions optimal for achieving the target (step S304: manufacturing condition variation prediction step). These steps S301 to S304 are similar to those in the embodiment of FIG. 2 (steps S301 and S302 are also similar to those in the embodiment of FIG. 1 ), so detailed description will be omitted.

[0066] 3, in this embodiment, the manufacturing conditions are then adjusted by the predicted amount of variation in the manufacturing conditions, and the paint is applied to the object to obtain a coating film (step S305: manufacturing process after adjusting conditions). Next, the properties of the coating film on the obtained coated object are measured (step S306: property measurement process). The measurement can be performed using any measuring means corresponding to the property to be measured.

[0067] According to the method for manufacturing a coating material of this embodiment, it is possible to predict the amount of variation in manufacturing conditions corresponding to the amount of variation in properties required to obtain the desired coating film properties, adjust the manufacturing conditions based on the prediction results, and then form a coating film again, thereby making it possible to manufacture a coating material with the desired coating film quality.

[0068] Here, if the properties of the coating film measured in the property measurement step deviate from the desired properties by a predetermined threshold or more, it is preferable to repeat the property variation prediction step, the production condition variation prediction step, the production step after condition adjustment, and the property measurement step. In this way, even if the prediction in one step is off, a coated product having the desired coating film quality can be produced by multiple adjustments.

[0069] In this embodiment, it is preferable to further include a relational data updating step of inputting relational data indicating the relationship between the amount of variation in the manufacturing conditions adjusted in the post-condition-adjustment manufacturing step and the amount of variation in the properties of the coating film measured in the property measuring step into a computer and updating the relational data. This allows the relational data to be updated, making it possible to create an artificial intelligence model based on more data. The update can be performed immediately after new data is obtained, or can be performed periodically or as needed. However, this step is not essential and can be omitted because it may lead to overlearning.

[0070] FIG. 4 is a flowchart showing a subflow of the method for producing a coated object according to one embodiment of the present invention. In this embodiment, if the properties of the coating film measured in the property measurement process deviate from the desired properties by a predetermined threshold or more a predetermined number of times, it is preferable to perform the following sub-flow. As shown in FIG. 4 , in this subflow, first, a separate artificial intelligence model is created using computer-based machine learning, which includes variations in manufacturing conditions as input and variations in coating film properties as output (Step S307: AI model creation substep). Next, the computer inputs the variations in manufacturing conditions into the separate artificial intelligence model, predicting variations in the coating film properties (Step S308: property variation prediction substep). In the property variation prediction substep, variations in multiple manufacturing conditions selected in advance as candidates are input, and variations in the properties of multiple coating films corresponding to each of the variations in manufacturing conditions are predicted. Next, based on the predicted variations in the properties of the multiple coating films, another optimal variation in manufacturing conditions is determined as the variation in another manufacturing condition that is optimal for achieving the target (Step S309: manufacturing condition variation prediction substep). Next, the manufacturing conditions are adjusted by the predicted variations in the other manufacturing conditions, and a coating is applied to the target object to obtain a coated object (Step S310: condition-adjusted manufacturing substep). Next, the properties of the coating film of the obtained coated object are measured (step S311: property measurement sub-step). Then, as shown in Fig. 4, if the properties of the coating film measured in the property measurement step deviate from the desired properties by a predetermined threshold or more, the AI ​​model creation sub-step, property variation amount prediction sub-step, manufacturing condition variation amount prediction sub-step, condition-adjusted manufacturing sub-step, and property measurement sub-step are repeated.

[0071] Each sub-step can be performed in the same manner as the corresponding step in the main step shown in FIG. 3, except that a different artificial intelligence model is used to obtain a different prediction result. This subflow makes it possible to produce a coated object having a coating film with the desired properties, even if the accuracy of the initially created artificial intelligence model is low. In the main process shown in Fig. 3 or the sub-process shown in Fig. 4, this flow can be ended when the properties of the coating film measured in the property measurement process deviate from the desired properties by less than a predetermined threshold. Note that, in the present disclosure, it is not necessarily required to perform the sub-flow.

[0072] Incidentally, the manufacturing method of a coated object often includes a plurality of steps, and Fig. 5 is a schematic diagram of a CCL (Color Coating Line) as an example. In the example shown schematically in FIG. 5, the process comprises a plurality of steps, including a step of preparing the paint (step (a)), a coating step using a roll (step (b)), a baking step using an oven (step (c)), and a winding step (step (d)). Each step has a process manufacturing condition. In step (a), the process manufacturing condition is, for example, one or more of the viscosity and heating residue of the paint; in step (b), it is, for example, one or more of the roll peripheral speed, the application pressure on the object, and the flow rate of the paint; in step (c), it is, for example, one or more of the baking temperature and the baking time; and in step (d), it is, for example, one or more of the temperature, humidity, and the temperature of the paint liquid on the production line. Here, the manufacturing conditions for each process are monitored when the manufacturing conditions are adjusted by the amount of fluctuation in the manufacturing conditions predicted in the manufacturing condition fluctuation amount prediction step in the flow of Fig. 3. For example, for paint, the above-mentioned properties and characteristics are measured, for manufacturing conditions, measured values ​​or (adjusted) set values ​​are used, for baking temperature and baking time, measured values ​​or (adjusted) set values ​​are used, and for environmental conditions of the production line, temperature and humidity can be measured. If any of the process manufacturing conditions is observed to have been adjusted to deviate from the expected fluctuation amount by a predetermined value or more, the following flow is carried out. First, in the created artificial intelligence model, the process manufacturing conditions that have been observed to have been adjusted to deviate from the planned variation by more than a predetermined value are fixed, and then the variation amounts of multiple other manufacturing conditions are input to calculate and predict the variation amounts of the properties of multiple coating films (property re-prediction process). Next, based on the predicted variations in the properties of the multiple coating films, the variations in the optimal manufacturing conditions are redetermined as the optimal variations in the manufacturing conditions for achieving the target (manufacturing condition variation re-prediction step). The amount of variation in the manufacturing conditions in the processes subsequent to the current process is adjusted based on the result of the manufacturing condition variation re-prediction process. To explain the above using the example of Figure 5, as an example, if a coating process (process (b)) using a roll is currently being carried out and it is observed that the adjustment of the paint properties and characteristics in process (a) has shifted, then, based on this assumption, it may be possible to re-predict the optimal amount of fluctuation in the conditions for process (c) and / or process (d), which are processes after the current process, and make appropriate adjustments that take into account the shift in adjustment in process (a). In this example, the artificial intelligence model is used to input multiple variations in the manufacturing conditions for steps (B) through (D), assuming that the paint properties and characteristics in step (A), where an adjustment error has already been observed, are substituted as fixed values. The system then calculates and predicts the variations in the properties of multiple coating films. Based on the predicted variations in the properties of multiple coating films, the optimal variations in the manufacturing conditions for steps (C) and / or (D) are re-determined as the optimal variations in the manufacturing conditions for achieving the target. This makes it possible to re-predict the optimal variations in the manufacturing conditions for steps (C) and / or (D) for achieving the desired properties, assuming an adjustment error in step (A). The manufacturing conditions for steps (C) and / or (D) are then adjusted based on the re-prediction results. Since the properties of the coating film must be evaluated after the series of processes are completed and the paint has dried, if the adjustment is off, it takes time to determine that readjustment is necessary, and during that time there is a risk that a coated product with a coating film that does not meet quality standards will be manufactured. In contrast, this method makes it possible to re-predict manufacturing conditions at any step in the process without waiting for the completion of a series of steps, thereby making it possible to avoid as much as possible the production of coated products with coating films that do not meet quality standards.

[0073] <Prediction system for fluctuations in coating properties> FIG. 6 is a block diagram of a system for predicting the amount of variation in the properties of a paint film according to one embodiment of the present invention. This system predicts the amount of variation in the properties of a paint film obtained by applying a paint to an object. As shown in FIG. 6, a system 10 for predicting the amount of variation in the properties of a paint film according to this embodiment includes a computer 11. The computer 11 has a machine learning function. The computer 11 also includes an artificial intelligence model creation unit 12 and a property prediction unit 13. The artificial intelligence model creation unit 12 has a machine learning function and creates a predetermined artificial intelligence model by machine learning, which includes variations in manufacturing conditions as input and outputs variations in the properties of the paint film. The property prediction unit 13 calculates and predicts the amount of variation in the properties of the paint film by inputting variations in the manufacturing conditions into the artificial intelligence model. The artificial intelligence model creation unit 12 and the property prediction unit 13 can be processors.

[0074] The above-mentioned relational data is input to the computer 11. In this embodiment, the computer 11 has a storage unit 14 (memory) for storing the relational data and a communication unit 15 for transmitting and receiving the relational data. The communication unit 15 can transmit and receive not only the relational data but also other data. The artificial intelligence model creation unit 12 preferably has a function for performing machine learning using the input relational data as training data. Furthermore, the computer 11 preferably further has a functional unit for performing data cleansing on the relational data. Furthermore, the present prediction system 10 preferably has a display unit (display) for displaying the prediction results. Details of the manufacturing conditions, properties of the coating film, machine learning, etc. are the same as those of the embodiment of the method shown in FIG. 1, so repeated explanations will be omitted.

[0075] According to the present prediction system 10, a predicted value of the amount of variation in the properties of the coating film in response to the amount of variation in the manufacturing conditions can be quickly obtained by computer calculation.

[0076] <Production condition fluctuation prediction system> FIG. 7 is a block diagram of a system for predicting the amount of variation in manufacturing conditions according to one embodiment of the present invention. This system predicts the amount of variation in manufacturing conditions that are optimal for achieving a target when manufacturing a coated object by applying paint to a target object, by varying and adjusting the manufacturing conditions to produce a coated object having target properties for the coating film of the coated object. As shown in FIG. 7, a manufacturing condition prediction system 20 according to this embodiment is configured with a computer 21. The computer 21 has a machine learning function. The computer 21 also has an artificial intelligence model creation unit 22, a property prediction unit 23, a memory unit 24, and a communication unit 25. These are similar to those described for the artificial intelligence model creation unit 12, the property prediction unit 13, the memory unit 14, and the communication unit 15 in the embodiment shown in FIG. 6, and therefore detailed description thereof will be omitted.

[0077] The property prediction unit 23 is configured to input variations in multiple manufacturing conditions and predict variations in the properties of multiple coating films corresponding to each variation in the manufacturing conditions. The computer 21 further includes a manufacturing condition prediction unit 26 that determines optimal variations in manufacturing conditions as optimal manufacturing conditions for achieving a target based on the predicted variations in the properties of the multiple coating films. The manufacturing condition prediction unit 26 can be a processor. The manufacturing condition prediction unit 26 is configured to determine, from the predicted variations in the properties of the multiple coating films, variations in manufacturing conditions corresponding to the variation in the properties of the coating film that has the smallest difference from the variation in the properties of the target coating film as the optimal variation in manufacturing conditions for achieving a target. The computer 21 preferably further includes a functional unit that performs data cleansing on the relational data. The prediction system 20 preferably includes a display unit 27 that displays the prediction results. Details of the manufacturing conditions, properties of the coating film, machine learning, etc. are the same as those of the embodiment of the method shown in Figure 2, so repeated explanations will be omitted.

[0078] The present prediction system 20 can obtain a predicted value of the optimal amount of variation in the manufacturing conditions when it is desired to vary the properties of a coating film by a predetermined amount. Therefore, it is sufficient to adjust the manufacturing conditions by the amount of the predicted value. Furthermore, because this adjustment can be performed regardless of the range of quality specifications, it is possible to avoid the problem described above of determining that no adjustment of the manufacturing conditions is necessary despite a discrepancy between the predicted value and the actual measured value, resulting in the failure to appropriately adjust the manufacturing conditions and thus to obtain a coated product with the desired coating film quality.

[0079] Examples of the present invention will be described below, but the present invention is not limited to the following examples. [Example]

[0080] The present invention will be explained in more detail with reference to the following examples, but the present invention is not limited to these examples.

[0081] First, the method for preparing the primary color paints used in preparing the paints of Examples 1 to 3 and Comparative Examples 1 and 2 will be described. <Preparation example of white primary color paint> 20 parts by mass of acrylic resin as the resin and 35 parts by mass of isophorone as the organic solvent were added, and the resin was uniformly dissolved using a disperser. Then, 46 parts by mass of titanium oxide as the white pigment was mixed in, and the mixture was dispersed using a sand mill (dispersion medium: glass beads) until the maximum particle size of the pigment coarse particles was 10 μm or less, thereby preparing a white primary color paint. <Preparation examples of other primary color paints> Black primary color paint 1, black primary color paint 2, yellow primary color paint, and red primary color paint were prepared in the same manner as in the preparation example of the white primary color paint, except that the type and amount of each material was changed as shown in Table 1 below. The formulations of each primary color paint are shown in Table 1.

[0082] [Table 1]

[0083] <Gloss adjuster> The gloss adjuster used was commercially available silica (silicon dioxide) used as a matting agent. <Viscosity modifier> The viscosity modifier used was isophorone, which is also used as a solvent.

[0084] <Preparation example of coating composition 1> Five parts by mass of acrylic resin, 25 parts by mass of fluororesin, 35 parts by mass of isophorone, and 35 parts by mass of cyclohexanone were added, and the resins were uniformly dissolved using a disper, followed by adding 25 parts by mass of white primary color paint, 2 parts by mass of black primary color paint 1, 69 parts by mass of yellow primary color paint, 4 parts by mass of red primary color paint, and 2 parts by mass of gloss adjuster 1, and mixing uniformly using a disper to prepare coating composition 1 (solids concentration: 47% by mass).

[0085] <Preparation Examples of Coating Compositions 2 to 4> Coating compositions 2 to 4 were prepared in the same manner as in the preparation example of coating composition 1, except that the type and amount of each material was changed as shown in Table 2 below. The formulation of each coating composition is shown in Table 2.

[0086] [Table 2]

[0087] Details of the materials used in preparing the coating compositions are as follows: Acrylic resin: Paraloid B44 (manufactured by Rohm & Haas), solid content: 100% by mass Fluorine resin: KYNAR500 (manufactured by ARKEMA), solid content: 100% by mass White pigment: TI-PURE R-706 (titanium dioxide, manufactured by DuPont) Black pigment 1: Mitsubishi Carbon Black MA-100 (carbon black, manufactured by Mitsubishi Chemical Corporation) Black pigment 2: SUNBLACK X15 (carbon black, manufactured by Shiraishi Calcium Co., Ltd.) Yellow pigment: TAROX synthetic iron oxide LL-XLO (yellow iron oxide, manufactured by Titan Kogyo Co., Ltd.) Red pigment: TODA COLOR 140ED (iron oxide, manufactured by Toda Kogyo Co., Ltd.) Gloss adjuster 1: GASIL HP395 (synthetic silica, manufactured by INEOS SILICAS) Gloss adjuster 2: Silysia 435 (silicon dioxide, manufactured by Fuji Silysia Chemical) Organic solvent: Isophorone (manufactured by ARKEMA) Organic solvent: Cyclohexanone (Ube Industries)

[0088] Coating film preparation method The following steps are carried out in one continuous process on a single coating line. Fine Tough G Primer (Primer 1) was applied as a primer to a substrate (zinc-aluminum alloy-plated steel sheet) using a roll coater (standard film thickness: 5 μm), and then baked for 60 seconds at a maximum temperature of 210°C to form a primer coating. Next, Coating Composition 1 was applied using a roll coater (standard film thickness: 15 μm), and then baked for 60 seconds at a maximum temperature of 250°C, followed by immediate cooling to form a coating of the unprepared coating composition.

[0089] Film thickness measurement method The film thickness of the primer paint or paint composition was calculated from the weight and specific gravity of the paint film before and after the paint film of the cut-out coated plate was removed by sandblasting. How to measure color difference The color of the coating was measured using Lab Scan XE (manufactured by HunterLab) and the difference from the color of the target coated plate was recorded as the color difference.

[0090] Artificial intelligence creation example - Data input for creating artificial intelligence models As coating parameters for adjusting the thickness, color, and gloss of the coating film on the coating line, data on the application roll (AP roll) speed fluctuation, coating pressure (coating pressure between the backup roll and AP roll) fluctuation, and primary color correction additive amount were input into a computer. For film thickness, color, and gloss, the film thickness, color, gloss, and paint viscosity before adjustment of the coating parameters and the film thickness variation, color variation, and gloss variation after adjustment of the coating parameters were used. - Creating artificial intelligence models Preprocessing such as data cleansing, zero point introduction, combination, power processing, and normalization was carried out in advance. A specified artificial intelligence model was created on a computer, with the pre-adjustment film thickness, color, gloss, paint viscosity, AP roll speed variation, coating pressure variation, and primary color paint additive amount as explanatory variables, and the film thickness variation, color variation, and gloss variation after painting parameter adjustment as objective variables. The optimal hyperparameters were determined using cross-validation for gradient boosting, random forest, support vector machine, and neural network as machine learning algorithms, and all data was retrained using the determined hyperparameters to create a specified artificial intelligence model with the data on manufacturing condition variation as explanatory variables and the variation in post-adjustment paint film properties as objective variables. Similarly, the color L * , a * , b * An artificial intelligence model was created to predict gloss fluctuations, resulting in a total of five artificial intelligence models. Comparative AI creation example The explanatory variables used in inputting data and creating the artificial intelligence model were AP roll speed fluctuation, coating pressure fluctuation, and primary color paint additive amount, but the AP roll speed absolute value, coating pressure absolute value, and primary color mixing ratio were used.The artificial intelligence model was created in the same way as the artificial intelligence creation example, except that the unadjusted film thickness, color, gloss, and paint viscosity were not used as explanatory variables, and comparative artificial intelligence model 1 was obtained. Similarly, the color L * , a * , b * A comparative artificial intelligence model was created to predict the absolute gloss value, resulting in a total of five artificial intelligence models. Table 3 below shows an example of artificial intelligence creation.

[0091] Next, the criteria for determining whether the prediction is acceptable for each of the examples and comparative examples will be described. Paint film property acceptance criteria The criteria for passing were to meet the following criteria. The film thickness of the standard plate and the adjusted paint plate, L * value, a * value, b * The difference between the values ​​is ΔFT and ΔL. * , Δa * , Δb * It was decided. The criteria for passing the test were that ΔFT was 1 μm or less and the color satisfied all three of the following conditions: ΔE=√(ΔL *2 +Δa *2 +Δb *2 ) is 0.1 or less, and ΔL * , Δa * , Δb * Each must be less than 0.1. The Δgloss value was considered to pass if the difference from the standard plate was 0.5 or less for gloss values ​​of 10 or less, 1 or less for gloss values ​​of 10 to 20, and 3 or less for gloss values ​​of 30 or more. Criteria for passing the number of adjustments If automation is possible, and the difference between the film thickness of the coating film before adjustment and the target film thickness is within 5 μm and ΔE2, and the number of adjustments of the coating film properties is within three times to meet the pass criteria, it is considered to have passed.

[0092] The evaluation results for each example and comparative example are as follows: The evaluation results are also shown in Table 4. Example 1 For coating composition 1 and manufacturing conditions with coating film properties that differ from the target film thickness by approximately 3 μm and have a ΔE of approximately 1, tolerance ranges were set for viscosity variation, coating pressure variation, and AP roll peripheral speed variation, and 10 million numbers were generated within each range. These were then randomly combined and, together with the film thickness before adjustment, entered into five artificial intelligence models that had been trained using gradient boosting. 10 million candidate data sets were obtained for film thickness variation, L* value variation, a* value variation, b* value variation, and gloss variation after coating parameter adjustment. 100% of ΔFT with the standard plate, ΔL * value, Δa * value, Δb *The 90% point of the value and 80% of the Δ gloss are set as the gap target value of the fluctuation amount, and the ΔL * value, Δa * value, Δb * One manufacturing condition that simultaneously satisfied the fluctuations in value and gloss was selected from 10 million options and adjusted. Since it did not pass, a second adjustment was performed in the same way with the gap target value set to 100%, and it passed, so it took two adjustments to pass. This type of adjustment was also carried out for coating compositions 2 to 4, with an average number of adjustments being 2.0 times.

[0093] <Example 2> For coating composition 1 and manufacturing conditions with coating film properties that differ from the target film thickness by approximately 2 μm and have a ΔE of approximately 2.5, tolerance ranges were set for viscosity variation, coating pressure variation, amount of primary color paint added, baking temperature variation, and production line temperature variation. 10 million numbers were generated within each range, and then these were randomly combined and input into five artificial intelligence models that had been trained using neural networks, along with the film thickness before adjustment, color, gloss, and paint viscosity. 10 million candidate data sets were obtained for film thickness variation, L* value variation, a* value variation, b* value variation, and gloss variation after adjusting the coating parameters. 100% of ΔFT with the standard plate, ΔL * value, Δa * value, Δb * The 90% point of the value and 90% of the Δgloss are set as the gap target value of the fluctuation amount, and the ΔL * value, Δa * value, Δb * One manufacturing condition that simultaneously satisfied the fluctuations in value and gloss was selected from 10 million combinations, and after adjustments, it was approved. This type of adjustment was also carried out for coating compositions 2 to 4, with the average number of adjustments being 1.5 times.

[0094] <Comparative Example 1> Using paint composition 1 and manufacturing conditions with a coating film property that differs from the target film thickness by approximately 2 μm and has a ΔE of approximately 2.5, a paint manager with over three years of experience adjusted the various coating conditions without using an artificial intelligence model until the film thickness, color, and gloss adjustments were successful.The test was successful after three adjustments. This type of adjustment was also carried out for coating compositions 2 to 4, with an average number of adjustments being 3.5 times.

[0095] <Comparative Example 2> For coating composition 1 and manufacturing conditions with a coating film property where the difference from the target film thickness is about 3 μm and about ΔE = 1, tolerance ranges are set for the absolute viscosity value, absolute coating pressure value, and absolute AP roll speed value. After generating 10 million numbers within each range, these are randomly combined and input into five pre-trained comparative artificial intelligence models. The absolute film thickness value after adjusting the coating parameters, L * Value absolute value, a * Value absolute value, b * We obtained 10 million candidate data sets for absolute values ​​of color and gloss. 100% of ΔFT with the standard plate, ΔL * value, Δa * value, Δb * The fluctuation amount at the point of 90% of the value and 80% of the Δgloss is added to each absolute value to obtain the gap target value, and the gap target value is simultaneously met. * Absolute value, a * Absolute value, b * One combination of the absolute value and gloss absolute value, the absolute value of the AP roll peripheral speed, the absolute value of the coating pressure, and the primary color mixing ratio was selected from 10 million combinations and adjusted. Since it did not pass, a second adjustment was performed in the same way with the gap target value of 100%, but it did not pass, and like the second time, three subsequent adjustments (a total of five) did not pass either. This same adjustment was also carried out for the four coating compositions 2 to 4, but the adjustment could not be made within five tries in any case.

[0096] Examples 1 and 2 passed the test with an average of 1.5 and 2.0 adjustments, respectively, meeting the above-mentioned pass criteria. In contrast, in Comparative Example 1, even an experienced paint supervisor required more than three adjustments, which increased the labor costs and was not automated, leaving the process dependent on the skill of the paint supervisor. In Comparative Example 2, when film thickness adjustment was performed using an AI model created by a comparative AI model, the paint parameters could not be adjusted. This was because, even when the predicted value and the actual measured value differed, if the predicted value was within the target film thickness range, it was determined that no adjustment was necessary, making it impossible to change the paint parameters. Furthermore, when absolute value predictions approached the target, adjustments often resulted in the properties moving away from the target properties.

[0097] [Table 3]

[0098] [Table 4] [Explanation of symbols]

[0099] 10: Prediction system, 11: Computer, 12: Artificial Intelligence Model Creation Department, 13: Property prediction unit, 14: Storage part, 15: Communications Department, 20: Prediction system, 21: Computer, 22: Artificial Intelligence Model Creation Department, 23: Property prediction unit, 24: Memory section, 25: Communications Department, 26: Manufacturing condition prediction unit, 27: Display section

Claims

1. A method for predicting the amount of variation from a reference value of the properties of a coating film obtained by applying a paint to an object, comprising: an artificial intelligence model acquisition step of acquiring an artificial intelligence model that includes, as an input, a variation amount of a manufacturing condition from a reference value and outputs a variation amount of a property of the coating film from the reference value; a property variation prediction step of predicting, by a computer, the amount of variation of the properties of the coating film from a reference value by inputting the amount of variation of the manufacturing conditions from a reference value into the artificial intelligence model, The manufacturing conditions are: (a) one or more of the viscosity and heating residue of the paint; (b) one or more of a roll peripheral speed, a coating pressure, and a coating flow rate in the step of applying a coating material to the object; (c) one or more of the baking temperature and the baking time in the baking step; and (d) one or more of the temperature, humidity, and paint liquid temperature of the production line; (a) to (d) above, the artificial intelligence model acquisition step is a step of creating the artificial intelligence model by machine learning, The method for predicting the amount of variation from a reference value of the properties of a coating film, characterized in that the machine learning uses a prediction algorithm based on decision trees, linear regression, partial least squares regression, lasso regression, ridge regression, polynomial regression, Gaussian process regression, support vector machine, random forest, gradient boosting, K-nearest neighbor method, neural network, Bayesian estimation, or an ensemble learning of these.

2. A method for predicting the amount of variation from a reference value of optimal production conditions for achieving a target when a coated object is produced by applying a paint to an object and adjusting the production conditions to vary them from a reference value, the method comprising: an artificial intelligence model acquisition step of acquiring an artificial intelligence model that includes, as an input, a variation amount of a manufacturing condition from a reference value and outputs a variation amount of a property of the coating film from the reference value; a property variation prediction step of predicting, by a computer, the amount of variation of the properties of the coating film from a reference value by inputting the amount of variation of the manufacturing conditions from a reference value into the artificial intelligence model, In the property variation prediction step, variations from reference values ​​of a plurality of the manufacturing conditions are input, and variations from reference values ​​of a plurality of the coating film properties corresponding to the variations from the reference values ​​of each of the manufacturing conditions are predicted; The method further includes a manufacturing condition variation prediction step of determining, based on the predicted variations from the reference values ​​of the properties of the plurality of coating films, variations from the reference values ​​of the optimal manufacturing conditions as variations from the reference values ​​of the optimal manufacturing conditions for achieving the target; The manufacturing conditions are: (a) one or more of the viscosity and heating residue of the paint; (b) one or more of a roll peripheral speed, a coating pressure, and a coating flow rate in the step of applying a coating material to the object; (c) one or more of the baking temperature and the baking time in the baking step; and (d) one or more of the temperature, humidity, and paint liquid temperature of the production line; (a) to (d) above, the artificial intelligence model acquisition step is a step of creating the artificial intelligence model by machine learning, The method for predicting the amount of variation from a reference value of a manufacturing condition, characterized in that the machine learning uses a prediction algorithm such as a decision tree, linear regression, partial least squares regression, lasso regression, ridge regression, polynomial regression, Gaussian process regression, support vector machine, random forest, gradient boosting, K-nearest neighbor method, neural network, Bayesian estimation, or an ensemble learning of these.

3. a relational data input step of inputting relational data indicating a relationship between a variation amount of the production conditions from a reference value and a variation amount of the properties of the coating film from a reference value into the computer, 3. The method according to claim 1, wherein the artificial intelligence model acquisition step is performed by the computer performing machine learning using the relational data input in the relational data input step as training data.

4. The method according to claim 1 or 2, wherein the properties of the coating film include one or more of the color, gloss, film thickness, smoothness, visual appearance, flip-flop property, and coating film abnormality of the coating film.

5. The method according to claim 1 or 2, wherein the machine learning uses a neural network, an ensemble tree composed of multiple decision trees, or a prediction algorithm based on partial least squares regression.

6. The method of claim 5 , wherein the machine learning uses a neural network, a random forest, or a gradient boosting method.

7. The method of claim 3 , further comprising the step of performing data cleansing on the relational data after the step of inputting the relational data.

8. 3. The method according to claim 2, wherein in the manufacturing condition variation prediction step, the variation amount from the standard value of the manufacturing condition corresponding to the variation amount from the standard value of the coating film property that has the smallest difference from the target property among the predicted variation amounts from the standard value of the properties of the coating film is determined as the variation amount from the standard value of the manufacturing condition that is optimal for achieving the target.

9. a post-conditions-adjusted manufacturing process, in which the manufacturing conditions are adjusted by an amount corresponding to a variation from a reference value of the manufacturing conditions predicted by the method of claim 2, and the coating material is applied to the object to obtain the coating film; and a property measuring step of measuring the properties of the coating film of the obtained coated object, A method for manufacturing a coated object, wherein, when the properties of the coating film measured in the property measurement process deviate from desired properties by a predetermined threshold value or more, the property variation prediction process, the manufacturing condition variation prediction process, the manufacturing process after condition adjustment, and the property measurement process are repeated.

10. When the property of the coating film measured in the property measurement step deviates from the desired property by a predetermined threshold value or more a predetermined number of times, an artificial intelligence model acquisition sub-process for acquiring another artificial intelligence model that includes, as an input, a variation amount of the manufacturing condition from a reference value and outputs a variation amount of the coating film property from a reference value; a property variation prediction sub-step of predicting, by the computer, a variation of the properties of the coating film from a reference value by inputting a variation of the manufacturing conditions from a reference value into the other artificial intelligence model; In the property variation prediction sub-step, variations from reference values ​​of a plurality of the manufacturing conditions selected in advance as candidates are input, and variations from reference values ​​of a plurality of the coating film properties corresponding to the variations from the reference values ​​of each of the manufacturing conditions are predicted; a manufacturing condition variation prediction sub-step of determining, based on the predicted variations from the reference values ​​of the properties of the plurality of coating films, a variation from the reference value of another optimal manufacturing condition as a variation from the reference value of another optimal manufacturing condition for achieving the target; a post-condition-adjustment manufacturing sub-process of adjusting the manufacturing conditions by an amount corresponding to the predicted variation from the reference value of the other manufacturing condition, and applying the paint to the object to obtain the coated object; a property measurement sub-step of measuring the properties of the coating film of the obtained coated object, 10. The method for manufacturing a coated object according to claim 9, wherein, when the properties of the coating film measured in the property measurement process deviate from the desired properties by a predetermined threshold or more, the artificial intelligence model acquisition sub-process, the property variation prediction sub-process, the manufacturing condition variation prediction sub-process, the manufacturing sub-process after condition adjustment, and the property measurement sub-process are repeated.

11. 10. The method for manufacturing a coated object according to claim 9, further comprising a relational data update step of inputting relational data indicating the relationship between the amount of variation of the manufacturing conditions from their reference values ​​and the amount of variation of the properties of the coating film measured in the property measurement step from their reference values ​​into the computer, thereby updating the relational data.

12. A system for predicting the amount of variation from a reference value of the properties of a coating film obtained by applying a paint to an object, an artificial intelligence model acquisition unit that acquires an artificial intelligence model that includes, as an input, a variation amount of a manufacturing condition from a reference value and outputs a variation amount of a property of the coating film from the reference value; a property prediction unit that calculates and predicts the amount of variation from a reference value of the property of the coating film by inputting the amount of variation from a reference value of the manufacturing condition in the artificial intelligence model; The manufacturing conditions are: (a) one or more of the viscosity and heating residue of the paint; (b) one or more of a roll peripheral speed, a coating pressure, and a coating flow rate in the step of applying a coating material to the object; (c) one or more of the baking temperature and the baking time in the baking step; and (d) one or more of the temperature, humidity, and paint liquid temperature of the production line; (a) to (d) above, the artificial intelligence model acquisition unit is configured to create the artificial intelligence model by machine learning; The system is characterized in that the machine learning uses a prediction algorithm such as a decision tree, linear regression, partial least squares regression, lasso regression, ridge regression, polynomial regression, Gaussian process regression, support vector machine, random forest, gradient boosting, K-nearest neighbor method, neural network, Bayesian estimation, or an ensemble learning of these.

13. A system for predicting the amount of variation from a reference value of optimal manufacturing conditions for achieving a target when manufacturing a coated object by applying paint to an object and adjusting the manufacturing conditions to vary them from a reference value, the system comprising: an artificial intelligence model acquisition unit that acquires an artificial intelligence model that includes, as an input, a variation amount of the manufacturing condition from a reference value and outputs a variation amount of the coating film property from the reference value; a property prediction unit that calculates and predicts deviations from reference values ​​of a plurality of properties of the coating film by inputting deviations from reference values ​​of a plurality of the manufacturing conditions in the artificial intelligence model; a manufacturing condition variation prediction unit that determines, based on the predicted variations from the reference values ​​of the properties of the plurality of coating films, variations from the reference values ​​of the optimal manufacturing conditions as variations from the reference values ​​of the optimal manufacturing conditions for achieving the target; The manufacturing conditions are: (a) one or more of the viscosity and heating residue of the paint; (b) one or more of a roll peripheral speed, a coating pressure, and a coating flow rate in the step of applying a coating material to the object; (c) one or more of the baking temperature and the baking time in the baking step; and (d) one or more of the temperature, humidity, and paint liquid temperature of the production line; (a) to (d) above, the artificial intelligence model acquisition unit is configured to create the artificial intelligence model by machine learning; The system is characterized in that the machine learning uses a prediction algorithm such as a decision tree, linear regression, partial least squares regression, lasso regression, ridge regression, polynomial regression, Gaussian process regression, support vector machine, random forest, gradient boosting, K-nearest neighbor method, neural network, Bayesian estimation, or an ensemble learning of these.

14. the manufacturing conditions include process manufacturing conditions for each of a plurality of processes, monitoring the manufacturing conditions for each step when the manufacturing conditions are adjusted by the amount of variation from the reference value of the manufacturing conditions predicted by the method of claim 2; When any of the process manufacturing conditions is observed to be deviated from the fluctuation amount by a predetermined value or more, a property re-prediction step in which, after fixing the monitored process manufacturing conditions in the artificial intelligence model, amounts of variation from reference values ​​of a plurality of other manufacturing conditions are input, thereby calculating and predicting amounts of variation from reference values ​​of a plurality of properties of the coating film; and a manufacturing condition variation re-prediction step of re-determining, based on the predicted variations from the reference values ​​of the properties of the plurality of coating films, variations from the reference values ​​of the optimal manufacturing conditions as variations from the reference values ​​of the optimal manufacturing conditions for achieving the target, The manufacturing method according to claim 10 , further comprising the step of readjusting the amount of variation from the reference value of the manufacturing condition in a process subsequent to the current process based on the result of the manufacturing condition variation re-prediction step.

Images