Prediction method of coating properties, prediction method of correcting formula composition, prediction system of coating properties, prediction system of correcting formula composition, and coating manufacturing method
Through machine learning models, the problems of subtle color differences and formula errors in paint manufacturing are solved, and high-precision and efficient paint manufacturing are achieved.
Patent Information
- Application Number
- CN202180084266.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-12-14
- Filing Date
- 2021-12-10
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-12-10
AI Technical Summary
The prior art is difficult to accurately predict and adjust subtle differences in color in coating manufacturing, especially in coatings for coil coatings, which lead to the problem of color inconsistency, and the formulation errors caused by different batches of primary color coatings are difficult to correct.
Machine learning methods are used to predict the composition of paint properties and correction formulas through artificial intelligence models, and computer input raw material correction formula composition data for coating properties adjustment, calculate the adjusted coating properties and change amount to achieve high-precision coating properties close to the expected value.
It improves the prediction accuracy and efficiency of coating manufacturing, can approach the target coating properties with high accuracy, reduces the number of manual adjustments and errors, and is suitable for paints of different uses, including automotive repair and coil coating.
Smart Images

Figure CN116583570B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for predicting coating properties, a method for predicting corrected formulation composition, a system for predicting coating properties, a system for predicting corrected formulation composition, and a method for manufacturing coating. Background Art
[0002] In the past, when manufacturing a coating, first, after preparing the coating before adjustment, it is repeatedly adjusted to approach the target coating properties. In the case of preparing the coating before adjustment, first, a plurality of primary color coatings are prepared by dispersing the pigment, resin, and solvent using an SG grinder or the like. Then, as a feeding process, primary color coatings, resins, solvents, and additives are added to the prepared coating before adjustment to prepare a coating in which they are dispersed and mixed. Here, the addition amount of primary color coatings, resins, solvents, and additives can be close to the target coating properties to a certain extent by pre-recording historical data such as an operation management table and referring to them.
[0003] Then, the viscosity of the prepared paint is measured using a viscometer, and the prepared paint is applied to the object, and its color difference, gloss, etc. are measured using a colorimeter and a gloss meter. Based on the color difference and gloss difference between each target coated board and historical data such as the work management table, the colorist manually adds the original color paint, solvent, and gloss adjuster for adjustment. By repeating such operations until the color difference and gloss difference between the target coated board are below a certain level, a paint with specified paint properties can be obtained.
[0004] In order to mechanize the repeated color matching operations, a computer color matching device used in the final color adjustment process is used. The computer color matching device generally used for automobile repair paint or architectural paint obtains the color expression mode when the basic primary colors are mixed according to the theoretical formulas of Kubelka-Munk and Duncan. Next, according to the color measurement values of the spectral reflection spectrum of the paint when each single primary color is actually used and the color measurement values when they are mixed at their respective predetermined ratios, corrections based on the measured theoretical calculations are performed, so that the color of the unknown ratio is calculated based on the small amount of measured data (original data), or the ratio of the target color is calculated. According to such a calculation method, the correction formula amount for changing from the current color (known formula) to the target color (formula calculation) (complementing the formula difference) is calculated.
[0005] Furthermore, a scheme has been proposed to achieve more efficient coating manufacturing by using a prediction method based on an artificial intelligence model using machine learning instead of calculation based on theoretical formulas, with the goal of predicting colors with higher accuracy and calculating the corrected formula quantities with high accuracy (for example, Patent Documents 1 and 2).
[0006] Prior art literature
[0007] Patent Literature
[0008] Patent document 1: Japanese Patent No. 3212101;
[0009] Patent document 2: Japanese Patent No. 6703639. Summary of the invention
[0010] Problem that the invention aims to solve
[0011] However, the tolerance of the slight difference in the color of each manufacturing of the coating varies according to the purpose. For example, in the coating for automobile repair, the boundary between the normal painted part of the faulty car (equivalent to the target painted plate) and the part of the paint adjusted for the new repair after the repair can be smudged (gradient), so the color consistency is about ΔE<0.5, which is not strictly required. On the other hand, the coil coating coating used as a pre-coat (painted before assembly) sometimes assembles the painted plates with different manufacturing batches adjacent to each other during product manufacturing, so there are also many products that allow the color difference caused by different manufacturing batches to be extremely small, such as ΔE<0.1. The color difference level is an area that varies according to the model and condition of the colorimeter, so the accuracy of the absolute value of the color is insufficient. Based on customer consultation, a standard plate (target painted plate) is prepared, and the color difference between the standard plate is measured according to each manufacturing and each color adjustment at the same time under the same conditions, and the color difference with the standard plate is controlled within the specified color difference. This level of color adjustment requires, at a minimum, adjustment of the amount of primary color added of about 0.01% of the total amount of paint.
[0012] When calculating the color according to the overall ratio and the theoretical formula, or making the artificial intelligence learn the relationship to make predictions, a slight deviation in the prediction accuracy can easily result in the wrong direction of the correction. For example, when the current formula is 10.0% black primary and 30.0% white primary, and the actual formula of the target color is 10.1% black primary and 29.9% white primary, the formula correction amount should be calculated based on the addition of about 0.1% black primary and the overall formula adjustment accompanying this. However, when the formula calculated based on the theoretical formula and actual correction of the target color or predicted by the artificial intelligence is 9.9% black primary and 30.1% white primary, the formula correction amount is calculated based on the addition of 0.1% white primary and the overall formula adjustment accompanying this, resulting in the color being moved away from the target color by the correction.
[0013] In the case where strict color matching is not required, depending on the application, the color difference caused by a 0.1% difference in the original formula composition is not a problem in many cases. According to the existing overall ratio, it is effective to calculate the color through theoretical formulas and actual measurement corrections, or to predict the color through artificial intelligence. However, for coatings such as coil coating coatings that adjust the amount of primary color coating by about 0.01% to a slight difference in color tone from a standard plate, this method cannot be used, and the adjustment has to be performed by a skilled colorist.
[0014] However, the color of each batch of primary color paint is slightly different due to the pigment batches used, manufacturing conditions, etc. For the slight difference in color, it is not necessarily the case that the same composition formula of the primary color paint will produce the same color. This is one of the reasons for the prediction error caused by the aforementioned calculation error / artificial intelligence. In addition, when manufacturing paint, it is not limited to starting according to the initial formula. For example, there may be a situation where inventory is adjusted, or the customer entrusts the paint to be re-prepared. However, there are also cases where part of the components of the paint are condensed during the period when it is kept as inventory; or after delivery to the customer, etc., it is transferred back again in a state where the customer is unclear about how to add primary color paint and solvents. In such a case, the earliest and initial formula information is useless. In the methods of the above-mentioned patent documents 1 and 2, it may be difficult to make a correct prediction in the above-mentioned situation.
[0015] Therefore, the object of the present invention is to provide a method for predicting coating properties or corrected formula composition that is efficient and has high prediction accuracy through a machine learning method, a prediction system for coating properties or corrected formula composition, and a method for manufacturing coating that is efficient and can make coating properties close to desired coating properties with high accuracy through a machine learning method.
[0016] Solutions for solving problems
[0017] The gist of the present invention is as follows.
[0018] (1) A method for predicting coating properties, characterized in that the coating properties after adjustment are predicted by a machine learning method when a raw material for adjusting coating properties is added to the coating before adjustment, the method for predicting coating properties comprising:
[0019] The paint property prediction step is to predict the paint property by inputting the corrected formula composition data of the raw materials used to adjust the paint property into a specified artificial intelligence model that takes the adjusted paint property as input and outputs the adjusted paint property, thereby calculating the adjusted paint property.
[0020] Here, the "data for correcting the formulation composition" can be set as data related to the formulation ingredients and their formulation amounts, and can also use data of the trade name (product number) instead of or on the basis of the data. Regarding the formulation ingredients, the data can also include the shape, chemical properties, etc., or color (L* value, a* value, b* value, spectral reflection spectrum) of each component such as colorants, adhesives, additives, etc. As the shape, the shape of the colorant, etc. (spherical, flaky, fibrous, etc.), the average primary particle size, the average secondary particle size, the average dispersed particle size, the particle size distribution, the aspect ratio, the thickness, etc. can be cited. As chemical properties, molecular weight, molecular weight distribution, color change temperature, reactivity, etc. can be cited.
[0021] (2) A method for predicting coating properties, characterized in that a machine learning method is used to predict the change in coating properties after adjustment when a raw material for adjusting coating properties is added to the coating before adjustment, the method for predicting coating properties comprising:
[0022] The paint property prediction process is to predict by using a computer to input the corrected formula composition data of the raw materials used to adjust the paint properties, in a prescribed artificial intelligence model that takes as input the data of the corrected formula composition of the raw materials used to adjust the paint properties and takes as output the change in the paint properties after adjustment, thereby calculating the change in the paint properties after adjustment.
[0023] (3) A method for predicting a modified formulation composition, characterized in that, when a raw material for adjusting the properties of a coating is added to a coating before adjustment to prepare a coating having a target coating property, the formulation composition of the raw material for adjusting the properties of the coating used to achieve the target coating property is predicted by a machine learning method,
[0024] The method for predicting a corrected recipe composition comprises: calculating the adjusted coating properties for prediction by the coating property prediction step of the coating property prediction method described in (1);
[0025] In the coating property prediction step, a plurality of the corrected recipe compositions are input, and a plurality of the adjusted coating properties corresponding to the respective corrected recipe compositions are predicted.
[0026] The method for predicting a corrected recipe composition further includes a corrected recipe prediction step of determining the corrected recipe composition obtained based on the predicted plurality of paint properties after adjustment as a recipe composition of raw materials for adjusting paint properties for achieving the target.
[0027] (4) A method for predicting a modified formulation composition, characterized in that, when a raw material for adjusting the properties of a coating is added to a coating before adjustment to prepare a coating having a target coating property, the formulation composition of the raw material for adjusting the properties of the coating used to achieve the target coating property is predicted by a machine learning method,
[0028] The method for predicting a corrected recipe composition comprises: performing a prediction by calculating a change in the adjusted paint property through the paint property prediction step of the paint property prediction method described in (2);
[0029] In the coating property prediction step, a plurality of the corrected recipe compositions are input, and changes in a plurality of the adjusted coating properties corresponding to the respective corrected recipe compositions are predicted.
[0030] The method for predicting a corrected recipe composition further includes a corrected recipe prediction step of determining the corrected recipe composition obtained based on the predicted changes in the plurality of adjusted paint properties as a recipe composition of raw materials for adjusting the paint properties to achieve the target.
[0031] (5) The method according to (2) or (4), further comprising:
[0032] a relational data input step of inputting into the computer relational data indicating the relation between the corrected formulation composition of the raw materials for adjusting the properties of the coating and the amount of change in the properties of the coating after adjustment; and
[0033] an artificial intelligence model acquisition step, acquiring the predetermined artificial intelligence model,
[0034] In the artificial intelligence model acquisition step, the relationship data input in the relationship data input step is used as learning data to perform machine learning using the computer to create the specified artificial intelligence model.
[0035] (6) The method described in any one of (1) to (5), wherein the predetermined artificial intelligence model further includes data on manufacturing conditions of the coating as input,
[0036] Contains any one or more of the following (a) to (d):
[0037] (a) any one or more of the viscosity and non-volatile content of the coating;
[0038] (b) any one or more of a roller peripheral speed in the step of applying the coating to an object, a coating pressure applied to the object, and a flow rate of the coating;
[0039] (c) one or more of the sintering temperature and the sintering time in the sintering step; and
[0040] (d) One or more of the temperature and humidity of the production line.
[0041] (7) The method described in any one of (1) to (6), characterized in that the input to the prescribed artificial intelligence model also includes data on the paint properties of the paint before the adjustment.
[0042] (8) The method according to any one of (1) to (7), wherein the properties of the coating material include at least the color of the coating material.
[0043] (9) The method described in (8), characterized in that the coating properties include gloss and / or viscosity of the coating in addition to the color.
[0044] (10) The method described in any one of (1) to (7), 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 machines, random forests, gradient boosting, K-nearest neighbor method, neural networks, Bayesian estimation, or their integrated learning.
[0045] (11) The method described in (8), characterized in that the machine learning uses a prediction algorithm based on a neural network.
[0046] (12) The method described in (4) is characterized in that, in the corrected recipe prediction step, the corrected recipe composition corresponding to the predicted change in the paint property after adjustment whose difference from the target paint property is less than a prescribed reference value among the multiple predicted changes in the paint property after adjustment is determined as the recipe composition of the raw materials for adjusting the paint property to achieve the target.
[0047] (13) The method according to (2) or (4), further comprising the step of setting a numerical value generation range of the recipe amount for each raw material in the modified recipe composition.
[0048] In the coating material property prediction step, the plurality of corrected recipe compositions are inputted by combining numerical values of recipe quantities of respective raw materials generated within the set numerical value generation range.
[0049] (14) The method described in (13), further comprising the step of determining an allowable range relative to the predicted modified recipe composition,
[0050] In the correction recipe prediction step, a recipe composition within the allowable range among the predicted correction recipe compositions is determined as a recipe composition of a raw material for adjusting coating properties for achieving the target.
[0051] (15) The method described in (4) is characterized in that the raw materials for adjusting the properties of the coating material are added to the coating material before adjustment a plurality of times, and the formulation composition of the raw materials for adjusting the properties of the coating material for achieving the target is predicted a plurality of times.
[0052] The formula composition of the raw materials for adjusting the paint properties to achieve the target is determined by performing prescribed operations on the predicted formula composition of the raw materials for adjusting the paint properties, or by performing prescribed operations on the difference formula composition of the raw materials for adjusting the paint properties obtained by using the difference target value with the target paint property as a baseline value.
[0053] (16) A method for producing a coating, characterized in that it further comprises:
[0054] an adjustment step of adding a raw material having the corrected formula composition predicted by the method described in (4) to the paint before adjustment, and adjusting the paint; and
[0055] The pass / fail judgment process measures the properties of the adjusted coating and judges whether it is pass / fail.
[0056] The modified recipe composition prediction step, the adjustment step, and the pass / fail judgment step are repeatedly performed until the pass / fail judgment step results in a pass / fail judgment result.
[0057] (17) The method for manufacturing the coating described in (16) is characterized in that it also comprises: a relationship data updating step, inputting into the computer relationship data representing the relationship between the predicted modified formula composition and the data on the amount of change in the coating property after adjustment measured in the pass / fail judgment step, and updating the relationship data.
[0058] (18) The method according to any one of (1) to (17), wherein the coating material is a coating material for coil coating.
[0059] (19) A system characterized in that a machine learning method is used to predict the coating properties of a coating after adjustment when raw materials for adjusting coating properties are added to a coating before adjustment, the system comprising a computer, the computer comprising a coating property prediction unit, the coating property prediction unit predicting the coating properties by inputting the corrected formula composition into a prescribed artificial intelligence model that takes data of the corrected formula composition of the raw materials for adjusting coating properties as input and takes the coating properties after adjustment as output, thereby calculating the coating properties after adjustment.
[0060] (20) A system characterized in that a machine learning method is used to predict the change in coating properties of a coating after adjustment when raw materials for adjusting coating properties are added to a coating before adjustment, the system comprising a computer, the computer comprising a coating property prediction unit, the coating property prediction unit performing prediction by inputting the corrected formula composition data including the raw materials for adjusting coating properties as input and taking the change in coating properties after adjustment as output into a prescribed artificial intelligence model, thereby calculating the change in coating properties after adjustment.
[0061] (21) A system characterized by predicting, by a machine learning method, a formula composition of a raw material for adjusting the properties of a coating material for achieving a target coating property when a raw material for adjusting the properties of a coating material is added to a coating material before adjustment to prepare a coating material having a target coating property,
[0062] The system is configured to predict the paint property after adjustment by the paint property prediction unit of the system described in (19),
[0063] The paint property prediction unit is configured to receive a plurality of the corrected recipe compositions as input, and predict a plurality of the adjusted paint properties corresponding to the respective corrected recipe compositions.
[0064] The system includes a computer, and the computer also includes a modified recipe composition prediction unit, which determines the modified recipe composition obtained based on the predicted plurality of adjusted paint properties as the recipe composition of raw materials for adjusting the paint properties to achieve the target.
[0065] (22) A system characterized by predicting, by a machine learning method, a formula composition of a raw material for adjusting the properties of a coating material for achieving a target coating property when a raw material for adjusting the properties of a coating material is added to a coating material before adjustment to prepare a coating material having a target coating property,
[0066] The system is configured such that the paint property prediction unit of the system described in (20) calculates the change amount of the paint property after adjustment to make a prediction,
[0067] The paint property prediction unit is configured to receive a plurality of the corrected recipe compositions as input, and to predict a plurality of changes in the adjusted paint properties corresponding to the respective corrected recipe compositions.
[0068] The computer has a modified recipe composition prediction unit, which determines the modified recipe composition obtained based on the predicted changes in the plurality of adjusted paint properties as the recipe composition of the raw materials for adjusting the paint properties to achieve the target.
[0069] (23) The system described in (22) is characterized in that the computer also has a pass / fail judgment unit, which judges whether the paint is pass / fail based on the result of adding raw materials having the predicted corrected formula composition to the paint before adjustment and measuring the properties of the paint after adjustment.
[0070] (24) The system described in any one of (19) to (23) further includes: an artificial intelligence model acquisition unit that acquires the specified artificial intelligence model.
[0071] Effects of the Invention
[0072] According to the present invention, a method for predicting coating properties or corrected formulation compositions that is efficient and has high prediction accuracy through a machine learning method, a system for predicting coating properties or corrected formulation compositions, and a method for manufacturing coatings that is efficient and can make coating properties approach desired coating properties with high accuracy through a machine learning method can be provided. BRIEF DESCRIPTION OF THE DRAWINGS
[0073] Figure 1 This is a flowchart of a method for predicting coating properties according to one embodiment of the present invention.
[0074] Figure 2 This is a flowchart of a method for predicting a modified recipe composition according to one embodiment of the present invention.
[0075] Figure 3 This is a flow chart of a method for producing a coating material according to one embodiment of the present invention.
[0076] Figure 4 This is a flowchart showing a sub-flow of a method for producing a coating material according to one embodiment of the present invention.
[0077] Figure 5 This is a block diagram of a coating property prediction system according to one embodiment of the present invention.
[0078] Figure 6 It is a block diagram of a prediction system for correcting recipe composition according to one embodiment of the present invention. DETAILED DESCRIPTION
[0079] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
[0080] <Prediction method of paint properties>
[0081] Figure 1 1 is a flow chart of a method for predicting coating properties according to one embodiment of the present invention. Figure 1 , an embodiment of a method for predicting the properties of a coating (or the amount of change in the properties of a coating) after adjustment when a raw material for adjusting the properties of the coating is added to the coating before adjustment is illustrated. In addition, as an example, the method for predicting the properties of the coating of the present embodiment can be performed using a prediction system for the properties of a coating of an embodiment of the present invention described later. Here, the predicted coating properties preferably include at least the color of the coating, and in addition, preferably also include the gloss and / or viscosity of the coating in addition to the color. In addition, the predicted coating properties can include, for example, film thickness, contrast, flip flop properties, smoothness, visual appearance, resistance, contact angle, contamination, solar reflectance, ultraviolet transmittance, weather resistance, viscoelasticity, coating abnormality, wire drawing, flash point, foaming, etc. In particular, when the gloss is changed, the color also changes, and the gloss changes when the primary color coating is added. If a solvent, water or other diluent or viscosity adjuster is added to adjust the viscosity, the color will change in the same way. If a primary color paint is added, the viscosity will change. Since they affect each other in this way, it is preferred to adjust the color by simultaneously predicting the gloss and viscosity.
[0082] For example, if a small amount of primary color paint is added to the paint before adjustment, the color of the coating film changes. In addition, if isophorone is added, the non-volatile content changes, so the film thickness changes. It is assumed that the "paint properties" mentioned here include such coating film properties such as the color and film thickness of the coating film.
[0083] Colors can be measured using color systems such as L* value, a* value, b* value in L*a*b* color space (JISZ8781-4 (2013)), XYZ color system, RGB color system, Yxy color system, Hunter Lab color system, L*C*h* color system, Munsell color system, etc. Colors can be measured using known color measurement methods. As an example, the commercially available CM-512m3 manufactured by Konica Minolta, Inc. can be used to irradiate light sources at angles of 25°, 45°, and 75° when the light receiving portion perpendicular to the coating is set to 0°. Alternatively, X-Rite MA68II (manufactured by X-Rite Inc.) can be used for measurement. The measurement angle can be appropriately adjusted according to the purpose or the equipment used. Any other index can be used. Furthermore, any index such as the reflection spectrum intensity of every 5 nm from 380 nm to 780 nm as the reflection spectrum data can also be used as an index of color. The gloss is not particularly limited, and gloss can be used as an index. The gloss can be measured using a known gloss measurement method. As an example, a mirror gloss meter (gloss meter VG 7000 (manufactured by Nippon Denshoku Industries Co., Ltd.)) can be used to measure the 60° gloss of the coating formed on the test plate according to JIS K 5600-4-7 (mirror gloss). The viscosity can be measured using a known viscosity measurement method. As an example, it can be measured according to JIS K 5600-2-2 (flow cup method). The smoothness preferably uses a waveform scan value (wave-scan) as an index. The waveform scanning value is preferably du (wavelength 0.1mm or less), Wa (wavelength 0.1-0.3mm), Wb (wavelength 0.3-1.0mm), Wc (wavelength 1.0-3.0mm), Wd (wavelength 3.0-10.0mm), We (wavelength 10.0-30.0mm), Lw (wavelength 1.2-12mm), and Sw (wavelength 0.3-1.2mm). In addition, the waveform scanning value means that the smaller the value, the less unevenness of the wavelength on the surface, and the better the appearance quality of the coating. The angular color effect can be measured using a variable color difference meter. The coating abnormality is specifically uneven, bulging, cracks, sagging, pinholes, thick edges, etc. For these, they can also be measured separately by known methods.
[0084] The inventors have discovered the following: instead of using a method to calculate (the absolute value of) paint properties (e.g., color) based on a recipe composition (the recipe composition of all constituent materials such as primary color paint, additives or main resin, or the primary color paint alone), a large amount of data on "how much and in which direction the paint properties (e.g., color) change after adding a few percent of a certain raw material to the paint before adjustment" is obtained from the work results of skilled colorists, and then learned through machine learning, which makes it easy to predict the direction and distance of correction, thereby making it possible to create a highly accurate artificial intelligence model.
[0085] Furthermore, as an example, this method can use the original color batch by performing a color measurement once at the initial charging stage when 90 to 98% of the required amount of paint is added. As a result, the current color value can be obtained, so even if there is a color difference of the original color batch, only 2 to 10% remains, so the influence can be reduced. Furthermore, based on the information of the color of the initial charging, the remaining 2 to 10% of the charging amount can be adjusted to reflect the coloring power of the original color batch.
[0086] Here, in the case of preparing a coating, for example, the coating before adjustment can be a coating prepared by performing the following operations: using an SG grinder, for example, to disperse a pigment, a resin, and a solvent, adding a resin, a solvent, and an additive to a primary color coating prepared by preparing a plurality of pigments, and dispersing them, and then repeatedly adjusting. In addition, if it is just after adjustment, the raw material can be in a clear state. In addition, if a long time has passed after adjustment, sometimes part of the composition condenses, etc., and the information of the raw material becomes incorrect information for predicting the properties of the coating. Alternatively, it is also conceivable that the information of the raw material is unclear when, for example, a customer entrusts the coating to be re-prepared, etc. In addition, it is also conceivable that after delivery to a customer, the raw material is returned again in a state where the customer is unclear about how to add the raw material, etc.
[0087] In addition, the coating before adjustment is not particularly limited, and as its method, for example, water-based coating, solvent-based coating, powder coating, solvent-free coating, etc. can be used. In addition, as the use, for example, coil coating, general industrial coating, automotive coating, automotive repair coating, architectural coating, heavy anti-corrosion coating, marine coating can be used, as the coating method, spray coating, drum coating, brush coating, roller coating (including forward and reverse), curtain flow coating, die coating, electrophoretic coating, powder coating, electrostatic coating can be used, as the drying method, sintering drying, forced drying, natural drying, ultraviolet curing can be used, and as the formulation composition, resin raw materials, pigments, design raw materials, solvents (including water), additives, etc. can be mentioned. As resin raw materials, acrylic resins, polyester resins, epoxy resins, alkyd resins, fluororesins, polyurethane resins, aminomelamine resins, isocyanate resins, blocked isocyanate resins, polyurethane-modified polyester resins and the like, modified resins thereof can be used; as pigments, inorganic pigments, organic pigments, coloring pigments and body pigments can be cited; as design raw materials, bright materials such as metallic colors and pearl colors, aggregates, silica, resin beads, waxes and the like can be cited; as additives, viscosity regulators, silicon-based additives, rust inhibitors, catalysts, defoaming agents and the like can be cited.
[0088] The data on the coating properties of the coating before adjustment can be set as data on coating property values, formula components and their formula correction amounts, or images of the coating film after coating. The coating property values can include non-volatile content, pH, acid value, amine value, conductivity, pigment concentration, SP, coating specific gravity, coating film specific gravity, pigment concentration, etc. Regarding the formula components, the shape, chemical properties, etc., or color (L*, a*, b* values, spectral reflection spectrum) of each component such as colorants, adhesives, additives, etc. can be cited. Data of the trade name (product number) can be used instead of or on the basis of the data. As the shape, the shape of the colorant, etc. (spherical, flaky, fibrous, etc.), average primary particle size, average secondary particle size, average dispersed particle size, particle size distribution, aspect ratio, thickness, etc. can be cited. As chemical properties, molecular weight, molecular weight distribution, color change temperature, reactivity, etc. can be cited. The image includes "static images and dynamic images".
[0089] The raw material for adjusting properties can use the raw material of the paint before correction, or an intermediate material manufactured as the raw material for adjusting properties can be prepared. Examples of the raw material for adjusting properties before correction include solvents or water for viscosity adjustment, viscosity regulators, gloss regulators such as silica and acrylic beads for gloss adjustment, primary color paints, dyes, pearlescent and metallic glitter materials for color adjustment, etc. Regarding the intermediate material manufactured as a raw material for adjusting properties or purchased as a commodity, as a color adjustment raw material, there can be mentioned: a paint that is substantially the same as the paint before adjustment made from a single coloring pigment; a primary color paint with a high concentration of a pigment made from a single coloring pigment; a glitter material dispersion in which a glitter material is dispersed in a resin solution. As the raw material for adjusting gloss, there can be mentioned a raw material obtained by dispersing the previous gloss adjustment raw material in a solvent or a small amount of a resin solution having a dispersing function, etc.
[0090] The data of the corrected formula composition of the raw materials used for adjusting the properties of the coating can be set as data related to the formula components and their formula correction amounts, and the types of raw materials used for correction and the combination of their respective amounts can be cited. Specifically, when the black primary color, yellow primary color, gloss regulator and solvent are used for correction, the respective correction amounts are expressed by a combination of formula correction amounts such as [0.5%, 0.02%, 0.6%, 3.5%]. It is also possible to use the data of the trade name (product number) instead of this data or on the basis of this data. Regarding the formula components, the shape, chemical properties, etc., or color (L*, a*, b* value, spectral reflection spectrum) of each component such as colorant, pigment, adhesive, additive, etc. can also be included in the data. As the shape, the shape of the colorant (spherical, flaky, fibrous, etc.), average primary particle size, average secondary particle size, average dispersed particle size, particle size distribution, aspect ratio, thickness, etc. can be cited. As chemical properties, molecular weight, molecular weight distribution, color change temperature, reactivity, etc. can be cited. The data of the corrected formula composition of the raw materials used to adjust the properties of the paint can also be expressed as a ratio, for example, it can be set as "amount of a single raw material added / total amount of paint before adjustment" or "amount of a single raw material added / total amount of original colors of the paint before adjustment".
[0091] The associated condition data can also be used to predict the properties of the coating. The associated condition data can include substrate data, coating condition data, measurement condition data, etc. The coating condition data can also be used as data for prediction. The coating condition data can include the type of substrate or the color, surface tension, and surface roughness of the substrate, and a simple trade name or substance can be used as data. As coating condition data, if it is roller coating, the linear speed, roller circumferential speed, clamping pressure, coating pressure and flow rate of the coating, the state of the coating roller, material, hardness, type of pick-up roller, sintering temperature in the sintering process, sintering time, temperature and humidity, plate thickness, coating line, maximum reaching temperature, bottom coating / surface coating adhesion, furnace temperature, colorimeter, type or color of the bottom coating, roughness, type and other physical property data, chemical property data, etc. can be cited. If it is spray coating, the ejection pressure, ejection amount, type of spray gun, sintering temperature in the sintering process, sintering time, temperature and humidity, etc. can be cited. In the case of electrophoretic coating, coating voltage, liquid temperature, power-on time, coating direction (horizontal or vertical), etc. can be cited. Measurement condition data can also be used as data for prediction. The measurement condition data can include the model of the colorimeter, the measurement temperature, the time from the power-on of the machine, and the name of the individual measuring device.
[0092] The predetermined artificial intelligence model further includes data on the manufacturing conditions of the coating as input, preferably including any one or more of the following (a) to (d):
[0093] (a) Any one or more of the viscosity and non-volatile content of the coating;
[0094] (b) any one or more of the peripheral speed of a roller in the step of applying the coating to an object, the coating pressure applied to the object, and the flow rate of the coating;
[0095] (c) one or more of the sintering temperature and the sintering time in the sintering step; and
[0096] (d) One or more of the temperature and humidity of the production line.
[0097] The properties of the paint after adjustment may also include color, gloss, viscosity, film thickness, contrast, angular color variation, micro-gloss, smoothness, visual appearance, resistance, contact angle, contamination, solar reflectance, UV transmittance, weather resistance, viscoelasticity, wire drawing, flash point, foaming, etc. In particular, when the gloss is changed, the color also changes, and the gloss changes when the primary color paint is added. If a diluent such as solvent or water or a viscosity regulator is added to adjust the viscosity, the color also changes, and if the primary color paint is added, the viscosity changes, and they affect each other in this way, so it is preferred to predict the gloss and viscosity at the same time when adjusting the color to adjust it.
[0098] In addition, when the film thickness may vary due to coating conditions, especially in the case of incompletely covering coatings, the film thickness will also affect the color and gloss. Therefore, the film thickness can be predicted at the same time, or the film thickness can be predicted once and the color and gloss can be predicted again using the film thickness data.
[0099] When preparing the paint before adjustment, it is preferred to prepare the paint to be close to the desired paint properties to some extent. As an example, it is preferred to prepare the paint with a ratio of raw materials that is considered appropriate using historical data such as a work management table. At this time, similar adjustment records can also be searched by computer.
[0100] Here, the "amount of change in the coating property" in this specification may be the numerical value (value) of the amount of change in the coating property after the above adjustment, or may be the ratio of the change in the coating property of the above adjustment. Here, as the "ratio of change in the coating property", for example, "the numerical value (value) of the amount of change in the coating property after adjustment / the amount of raw materials added for the preparation of the coating property", "the numerical value (value) of the coating property after adjustment / the ratio of raw materials added for the preparation of the coating property", "the sum of the absolute values of the numerical values (values) of each adjusted coating property / the numerical values (values) of the adjusted coating property", etc. may be cited.
[0101] like Figure 1As shown, in this embodiment, first, relationship data indicating the relationship between the coating properties of the coating before adjustment, the modified formula composition of the raw materials for adjusting the coating properties, and the coating properties after adjustment (or the amount of change in the coating properties) are prepared (step S101). The relationship data can be prepared using, for example, manufacturing performance data, data obtained by experiments, data calculated by a commercially available computer color matching device, data calculated based on a theoretical formula such as Kubelka-Munk or Duncan's formula, etc. (in a form that can be input to a computer). In addition, as described above, the predicted paint properties preferably include at least the color of the paint. Therefore, the relationship data is preferably a large data set that associates at least the color of the paint before adjustment (indicators such as other color spaces such as L* value, a* value, and b* value in the L*a*b* color space can also be used, and further, arbitrary indicators such as the reflection spectrum intensity every 5nm from 380nm to 780nm can also be used as reflection spectrum data as an indicator of color), the amount of each primary color paint (black primary paint, white primary paint, blue primary paint, yellow primary paint, red primary paint, etc.) of the raw materials added to the paint before adjustment, and the color of the paint after adjustment. The value is sufficient for machine learning described later. In addition, the predicted coating properties preferably include the gloss and / or viscosity of the coating in addition to the color, so the relationship data is preferably set to a large number of data sets obtained by associating the following values: the color value, and the gloss and / or viscosity value of the coating (other indicators can also be used respectively); the value of the index of the amount of each primary color (black primary color coating, white primary color coating, blue primary color coating, yellow primary color coating, red primary color coating, etc.) of the raw material coating added to the coating before adjustment, and the value of the index of the amount of gloss modifier and / or viscosity modifier or the color value; and the color value of the coating after adjustment, and the gloss and / or viscosity value. Here, the data is not limited to historical data, and data obtained by experiments, data calculated by a computer color matching device, data calculated according to theoretical formulas such as the Kubelka-Munk or Duncan formula, etc. can also be used.
[0102] The corrected recipe composition may be, for example, a value obtained by dividing each corrected recipe amount by the total amount of the paint composition before adjustment, or a value obtained by dividing each corrected recipe amount by the total amount of the primary colors.
[0103] In addition, in addition to including the paint properties of the paint before adjustment, the corrected formula composition of the raw materials used to adjust the paint properties, and the paint properties after adjustment, the relationship data can also include, for example, the relationship with the primer information during coating film production and / or the measuring instrument information during color and gloss measurement.
[0104] Furthermore, instead of the numerical value of the color, the shape of the reflection spectrum itself may be used as image data as an index to serve as color data.
[0105] In order to enable more accurate prediction, it is preferred that the above-mentioned relationship data is always updated timely or regularly.
[0106] like Figure 1 As shown, in the present embodiment, next, relational data representing the relationship between the paint properties of the paint before adjustment, the revised formula composition of the raw materials used for adjusting the paint properties, and the paint properties after adjustment (or the change in the paint properties) are input into the computer (relational data input process: step S102).
[0107] Next, if Figure 1 As shown, through computer-based machine learning, a prescribed artificial intelligence model is produced that takes the data of the coating properties of the coating before adjustment and the data of the modified formula composition of the raw materials used for adjusting the coating properties as input, and takes the coating properties after adjustment (or the amount of change in the coating properties) as output (artificial intelligence model production process: step S103). In addition, in the present embodiment, an artificial intelligence model is produced in this step, but it is not necessarily necessary to produce an artificial intelligence model, and an artificial intelligence model that has been produced can also be obtained from the outside. For example, the artificial intelligence model can be obtained by the communication unit of the computer, or it can also be obtained by being transferred by someone else.
[0108] The machine learning algorithm can use any known algorithm, such as 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 their integrated learning.
[0109] It is particularly preferable to use a prediction algorithm based on a neural network for machine learning. Deep learning in which 3 to several hundred layers are formed while performing convolutional neural networks and dropout can also be used for the neural network.
[0110] In addition, when making artificial intelligence, you can use machine learning frameworks such as TensorFlow, Keras, Caffe, PyTorch, Chainer, Scikit-learn, etc.
[0111] Here, machine learning can be set as supervised learning (including reinforcement learning). In this embodiment, the computer uses the above-mentioned relationship data input in the relationship data input process (step S102) as learning data (supervised data in this example) to perform machine learning (supervised learning), thereby performing an artificial intelligence model production process (step S103).
[0112] Here, the preferred algorithm is determined by the following method. That is, various machine learning algorithms are used to create multiple artificial intelligence models, and data with known answers in advance are used to compare the predicted values of the output with the answers, thereby confirming the accuracy of each artificial intelligence model. Then, for the algorithm with high accuracy among the multiple algorithms, a model is created by, for example, cross-validation, or the optimal hyperparameters are determined by using Bayesian optimization, etc. By re-learning all the data with the determined hyperparameters, it is possible to create an artificial intelligence model that uses the data of the coating properties of the coating before adjustment and the data of the modified formula composition of the raw materials used for adjusting the coating properties as input, and uses the adjusted coating properties (or the change in coating properties) as output.
[0113] Here, in this embodiment, the prepared relational data is preprocessed. Here, the prepared relational data is processed as follows: for example, data normalization / standardization, logarithm taking, exponentiation (exponentiation) and other prescribed operations, processing such as generating new data, or data filtering processing to remove inappropriate data. For example, data with a large difference between the measured value and the predicted value of the coating property can be carefully checked, and data suspected of input errors or record errors can be deleted.
[0114] As described above, in this embodiment, various machine learning algorithms can be used, but the above-mentioned preprocessing can be set to be commonly applied to multiple or even all algorithms (for example, removing data such as outliers that are considered to be inappropriate regardless of which machine learning algorithm is used), or can also be set to be specifically applied to the machine learning algorithm actually used (for example, removing data that is prone to errors in a specific algorithm).
[0115] In addition, it is preferred to add zero-point relationship data (zero point addition) in the pre-processing, which shows the relationship that the change in the paint property obtained by adjustment is 0 when the addition amount of the correction formula composition of the raw materials used for adjusting the paint property is 0. The data used for machine learning does not include data such as "no correction formula is added (the addition amount is 0)" in the case of, for example, the colorist's color adjustment record data. The result obtained by machine learning becomes a calibration curve that does not pass through the origin (the point where the addition amount of the correction formula is 0 and the change in the paint property is 0), which may lead to a decrease in the accuracy of the prediction. Therefore, by adding the above-mentioned zero-point relationship data, such problems can be avoided and the accuracy of the prediction can be further improved. Here, a value close to zero can also be used instead of zero.
[0116] Furthermore, when the records of continuous color adjustment operations are used as data, the results of adding them together can be used as a color adjustment record to increase the data (combination). For example, in the first and second color adjustment records, the first record is that 0.1% of the black primary color is added, and the L* value decreases by 0.5, and the second record is that 0.1% of the black primary color and the red primary color are added, and the L* value further decreases by 1.1. Adding these records becomes the following operation: the color adjustment record in which 0.2% of the black primary color and 0.1% of the red primary color are added initially to decrease the L* value by 1.6 is added as data. In this way, the accuracy of prediction can be further improved from the two aspects of the effect of learning the data with large changes and the effect of increasing the data.
[0117] As a preprocessing, it is also possible to add inferred data between data. For example, an approximate straight line or an approximate curve can be obtained from the data, and the points on the approximate straight line or the approximate curve can be used as inferred data. In this way, the number of data can be increased to further improve the accuracy of the prediction. In addition, historical data can be connected with data by a straight line or a curve, and any number of data can be added to the line.
[0118] Such preprocessing or normalization can make it possible to predict the properties of the coating with higher accuracy in the machine learning described later. However, in the present disclosure, preprocessing is not essential and this step can be omitted.
[0119] The preprocessed data can be used to create a specified artificial intelligence model using the same method as described above.
[0120] Next, in this embodiment, the data of the paint properties of the paint before adjustment and the corrected formula composition are input into the above-mentioned artificial intelligence model by a computer, so as to calculate the paint properties after adjustment (or the change in the paint properties) for prediction (paint property prediction process: step S104).
[0121] First, in the coating property prediction process (step S104), the data of the coating property of the input coating before adjustment can be prepared by measuring the coating property of the coating before adjustment. The color can be measured using any known colorimeter, etc., and the gloss can be measured using any known gloss meter, and the viscosity can be measured using any known viscometer. Other coating properties can also be measured by any known measuring means. The measuring method of each coating property is not particularly limited. For example, the object after the coating is actually applied to the predetermined object to be coated with the coating can be measured, or the coating before coating can be directly measured (liquid colorimetry). In the case of coating (applying), various coating methods such as rod coating, roller coating, spray coating, dip coating, electrophoretic coating, printing coating, curtain flow coating, brush coating, roller coater coating, airless spray coating, powder coating, offset printing, flexographic printing, electrostatic coating, air spray coating, smear coating, spin coating, etc. can be used.
[0122] In addition, in this embodiment, in steps S101 to 104, the data of the paint properties of the paint before adjustment is used as an explanatory variable in the artificial intelligence model, but the data of the paint properties of the paint before adjustment may not necessarily be included in the explanatory variable. For example, in the color adjustment of the paint for automobile repair, the vehicle type information and the paint color number may also be used as explanatory variables.
[0123] On the other hand, when using data on the coating properties of the coating before adjustment, especially when obtaining data by measuring the object to be coated, it is possible to include data such as plate thickness, film thickness, PMT (maximum temperature reached), sintering temperature, and base coating conditions in the data on the coating properties of the coating before adjustment of the explanatory variables.
[0124] In addition, in the paint property prediction process (step S104), the input correction formula composition becomes a numerical value of an index of the amount of each primary color paint (black primary color paint, white primary color paint, blue primary color paint, yellow primary color paint, red primary color paint, etc.) of the raw material when the predicted paint property is color. When gloss and viscosity are added thereto, it can be set as a numerical value of an index of the amount of a gloss regulator and / or a viscosity regulator.
[0125] In order to improve the accuracy of prediction, the input corrected recipe composition is preferably narrowed down in advance as a candidate by referring to data (such as actual production performance data) that is the basis of the above-mentioned relationship data input for machine learning.
[0126] In this way, by inputting the data of the paint properties of the paint before adjustment and the correction formula composition into the above-mentioned artificial intelligence model, the paint properties after adjustment (or the change in the paint properties) can be calculated by a computer. In addition, the calculated predicted value of the paint property can be a single value, or it can be composed of multiple candidate groups. In the case where the calculated predicted value of the paint property is composed of multiple candidate groups, it is preferably also included. A process of appropriately selecting a single value from them using a prescribed benchmark. The above-mentioned prescribed benchmark can be set in various ways. For example, in the case of color, it can be an E value (ΔE=(ΔL *2 +Δa *2 +Δb *2 ) 0.5 ) is selected so as to be within a predetermined range (for example, ΔE is less than 0.1, gloss is less than 0.2, etc.). As an example, it is possible to select so as to minimize ΔE.
[0127] In addition, the "calculating the adjusted coating properties to predict by using a computer to input the corrected formula composition data containing raw materials for adjusting coating properties as input and the adjusted coating properties as output" in "calculating the adjusted coating properties to predict by using a computer to input the corrected formula composition in a prescribed artificial intelligence model that takes data of the corrected formula composition of raw materials for adjusting coating properties as input and takes the adjusted coating properties as output" can include the following method: predicting the difference between the expected properties of the adjusted coating properties and the predicted coating properties when the raw materials for preparing the coating properties are added to the coating before adjustment by a machine learning method, thereby calculating the adjusted coating properties to predict. In this case, the above difference is directly predicted by machine learning, thereby calculating the adjusted coating properties to predict.
[0128] According to the coating property prediction method of the present embodiment, the predicted value of the coating property can be obtained by using the artificial intelligence model input data, so that high efficiency can be achieved. In addition, the artificial intelligence model takes the data of the coating property of the coating before adjustment and the data of the modified formula composition of the raw materials used for adjusting the coating property as input, and takes the adjusted coating property (or the change in the coating property) as output, so regardless of whether the information of the raw materials of the coating before adjustment is useful for predicting the coating property at the current moment, the data of the coating property of the coating before adjustment that is actually measured can be used as input to obtain the above-mentioned predicted value, even in the case where the coating property at the current moment changes from the coating property obtained from the initial raw materials, this will not affect the calculation of the predicted value, and the adjusted coating property can be predicted with high accuracy.
[0129] By simultaneously predicting gloss and / or viscosity as paint properties in addition to color as in this example, it is possible to perform prediction with higher accuracy taking into account their mutual influence.
[0130] In addition, by using the change in paint properties as the output of a specified artificial intelligence model, errors or deviations caused by measuring instruments can be offset, further improving the accuracy of predictions.
[0131] <Prediction method for corrected recipe composition>
[0132] Figure 2 1 is a flowchart of a method for predicting a modified recipe composition according to an embodiment of the present invention. Figure 2 , an embodiment of a method for predicting the composition of raw materials for adjusting the properties of a coating material to achieve the target coating material properties when raw materials for adjusting the properties of the coating material are added to the coating material before adjustment to prepare the coating material having the target coating material properties is illustrated. In addition, as an example, the method for predicting the composition of the modified formulation of this embodiment can be executed using the prediction system for the composition of the modified formulation of an embodiment of the present invention described later. Here, Figure 1 Similar to the embodiment of the present invention, it is preferred that the coating properties at least include the color of the coating, and further preferably include the gloss and / or viscosity of the coating in addition to the color. In addition, in addition to color, gloss and viscosity, the coating properties may also include, for example, film thickness, contrast, angular chromatic aberration, smoothness, visual appearance, electrical resistance, contact angle, contamination, solar reflectance, UV transmittance, weather resistance, viscoelasticity, coating abnormality, wire drawing, foaming, etc. In addition, as an example, the predicted corrected formula composition is the numerical value of the index of the amount of each primary color coating (black primary color coating, white primary color coating, blue primary color coating, yellow primary color coating, red primary color coating, etc.) of the raw material or the numerical value of the color, the numerical value of the index of the amount of the gloss regulator and / or the viscosity regulator. For other details, refer to Figure 1 The same embodiment as in the embodiment of the present invention is omitted for brevity. Figure 1 The implementation method is the same, so the description is omitted.
[0133] like Figure 2 As shown, in the present embodiment, first, relationship data representing the relationship between the paint properties of the paint before adjustment, the corrected formula composition of the raw materials used for adjusting the paint properties, and the paint properties after adjustment (or the amount of change in the paint properties) are prepared (step S201). Next, the relationship data representing the relationship between the paint properties of the paint before adjustment, the corrected formula composition of the raw materials used for adjusting the paint properties, and the paint properties after adjustment (or the amount of change in the paint properties) are input into the computer (step S202). Next, by using the machine learning of the computer, a prescribed artificial intelligence model is created, which takes the data of the paint properties of the paint before adjustment and the data of the corrected formula composition of the raw materials used for adjusting the paint properties as input, and takes the paint properties after adjustment (or the amount of change in the paint properties) as output (step S203). For these steps S201 to S203, Figure 1 The implementation method is the same as that of FIG. 1 , so the detailed description is omitted.
[0134] like Figure 2 As shown, in this embodiment, next, by using a computer to input the data of the paint properties of the paint before adjustment and the modified formula composition into the above-mentioned artificial intelligence model, the paint properties after adjustment (or the change in the paint properties) are calculated for prediction (paint property prediction process: step S204). Figure 2 In the coating property prediction process (step S204) of the embodiment of the present invention, a plurality of modified formula compositions pre-selected as candidates are input, and a plurality of adjusted coating properties (or changes in coating properties) corresponding to each modified formula composition are predicted. When a plurality of modified formula compositions are pre-selected, historical data such as a job management table can be referenced.
[0135] For example, the coating properties of the coating composition before adjustment, the standard formula of the raw materials for adjusting the coating properties, and the formula of the raw materials added when manufacturing the coating composition before adjustment are input into the computer. For the data input of the formula prediction, the lower limit and upper limit of the variation range of the modified formula of the raw materials for adjusting the coating properties are set, and based on the set variation range of the modified formula, candidate data of (for example) 100,000 combinations of candidates for the modified formula composition of the raw materials for adjusting the coating properties (combinations of the addition amounts of each primary color coating, matte material, and viscosity modifier) are prepared. Candidate data for the modified formula composition of the raw materials for adjusting the coating properties can be prepared based on the variation range of the modified formula in which the lower limit and upper limit of the variation range of the modified formula of the raw materials for adjusting the coating properties are set to the maximum possible range (for example, the amount of the primary color coating and the gloss modifier is in the range of 0 to 2% relative to the total amount of the coating before adjustment, and the viscosity modifier is in the range of 0 to 10% relative to the total amount of the coating before adjustment). Furthermore, instead of preparing a large amount of candidate data in advance, a sequential search algorithm such as Bayesian optimization or a genetic algorithm can be used to obtain a plurality of candidate data and predicted values of the paint property (or the amount of change in the paint property).
[0136] In this embodiment, next, a corrected recipe composition obtained based on the predicted multiple adjusted paint properties (or changes in paint properties) is determined as the recipe composition of raw materials for adjusting the target paint properties (corrected recipe prediction process: step S205).
[0137] Specifically, as an example, in the correction formula prediction process (step S205), the correction formula composition corresponding to the adjusted paint property (or its change amount) whose difference (e.g., E value) from the target paint property is less than a predetermined reference value (e.g., ΔE is less than 0.1, gloss is less than 0.2, etc.) among the predicted multiple adjusted paint properties (or the change amount of the paint property) is determined as the formula composition of the raw material for achieving the target paint property adjustment. As an example, it can be selected in a way that ΔE is minimized.
[0138] According to the prediction method of the modified formula composition of the present embodiment, the predicted value of the modified formula composition can be obtained by using the artificial intelligence model input data, so that efficiency can be achieved. In addition, the artificial intelligence model uses the data of the coating properties of the coating before adjustment and the data of the modified formula composition of the raw materials for adjusting the coating properties as input, and uses the adjusted coating properties (or the change in coating properties) as output. Furthermore, in the present embodiment, a plurality of modified formula compositions selected in advance as candidates are input, and the predicted values of a plurality of coating properties (or the change in coating properties) are obtained as output, and based on the predicted value (in this example, the difference with the target value is less than the prescribed reference value), the modified formula composition corresponding to the adjusted coating properties (or the change in coating properties) is determined as the formula composition of the raw materials for adjusting the coating properties to achieve the target. Therefore, regardless of whether the information on the raw materials of the paint before adjustment is useful for predicting the paint properties at the current moment, the data on the paint properties of the paint before adjustment actually measured can be used as input to obtain the above-mentioned prediction value. Even if the paint properties at the current moment change from the paint properties obtained from the initial raw materials, this will not affect the calculation of the prediction value, and the corrected formula composition for achieving the target paint properties can be predicted with high accuracy.
[0139] In addition to the color, gloss and / or viscosity are also predicted simultaneously as coating properties, so that a prediction can be made with greater accuracy taking into account their mutual influence.
[0140] In addition, by using the changes in paint properties as the output of a specified artificial intelligence model, errors or deviations caused by measuring instruments can be offset, further improving the accuracy of predictions.
[0141] Furthermore, these methods can be programmed to simulate without actually performing color adjustment, and trial and error can be performed to determine whether conditions such as the proximity rate or difference rate, the correction formula value input range, and the number of candidates for the correction formula composition are appropriate, and used for preliminary confirmation before actual color adjustment.
[0142] Here, when preparing paint using artificial intelligence, when predicting the required correction formula based on paint properties such as color, it may happen that the given solution is not the same, but a predicted formula that is far from the actual record is suggested (for example, although the target color can be obtained, the amount of the original color formula is increased, the additive ratio not included in the original color is reduced, and the man-hours for additive correction are generated, etc.). Therefore, in order to verify the accuracy of the correction formula predicted by artificial intelligence, the paint is prepared and confirmed each time, and there is a problem that the efficiency is reduced because unreasonable predictions such as the above are also prepared.
[0143] Therefore, it is preferred that the process further includes setting a numerical value generation range for the formula quantity of each raw material in the modified formula composition, and in the coating property prediction process (step S204), a plurality of the modified formula compositions formed by combining the numerical values of the formula quantities of each raw material generated within the set numerical value generation range are input. Thus, prediction can be performed efficiently.
[0144] In addition, it is preferred that the method further includes a step of determining an allowable range relative to the predicted correction formula composition, and in the correction formula prediction step (step S205), the formula composition within the allowable range among the predicted correction formula composition is determined as the formula composition of the raw material for adjusting the properties of the coating to achieve the target. Thus, the correction formula can be predicted after the allowable range is set in advance, so the possibility of generating useless formulas is reduced, and thus a prediction result of the correction formula composition with higher accuracy can be efficiently obtained.
[0145] Alternatively, the numerical value generation range may be set to be the same as the allowable range, and a plurality of corrected recipe compositions may be input by combining numerical values of recipe quantities of each raw material generated within the set numerical value generation range (allowable range).
[0146] The above-mentioned numerical value generation range can be obtained based on a certain range from a preset standard value or actual value of the paint recipe after adjustment. Thus, a predicted value with a small deviation from the standard value or actual value can be obtained.
[0147] Alternatively, the above-mentioned numerical value generation range can also be obtained based on at least one of the qualified range of the weather resistance test of the coating film of the adjusted coating formula and the qualified range of the metamerism test. The coating film with the coating properties adjusted using this prediction is often used outdoors, and sometimes needs to pass a long-term durability test (10 to 30 years) called weather resistance or corrosion resistance. However, since there are multiple recipes for achieving a specific color, if a different situation occurs during the color matching process from the recipe when the durability test is performed, it is possible that the weather resistance or corrosion resistance is reduced and deterioration occurs earlier than expected. Therefore, by using the qualified range of the weather resistance test as the numerical value generation range, such a problem can be dealt with. In addition, this is because, from the perspective of metamerism, if the primary color ratio deviates, it may occur that although the same color is shown in the colorimeter, the color is different when exposed to the sun, so by using the qualified range of the metamerism test as the numerical value generation range, such a problem can be dealt with.
[0148] Here, it is preferred that the plurality of corrected recipe compositions input in the coating property prediction step (step S204) consist of candidates equal to or greater than the number of raw materials in the corrected recipe composition of 10. For example, when the number of raw materials is 4, it becomes 10 4 =Total number of combinations = 1000. This 1000 is equivalent to the total number when 4 raw materials are used for production. When the allowable correction formula amount of the yellow raw material is 0-10%, the candidate formula is produced in 10 stages with an interval of 1%, and the same operation is performed for the red raw material, the black raw material, and the white raw material. This is because when more detailed fine-tuning is strictly required for the difference in color, the number of candidates is the number of raw materials composed of 10 correction formulas + 1 (10000 in the case of 4 raw materials). Furthermore, in the case of coil coating and other applications, especially when the difference in color is strict or the allowable correction formula range is large, the number of candidates is the number of raw materials composed of the correction formula + 2 (100000 in the case of 4 raw materials), so that predictions corresponding to such fine-tuning can be made.
[0149] In addition, it is preferred to generate a numerical value of the formula amount of each raw material for adjusting the properties of the coating, generate a candidate of the correction formula composition formed by randomly combining the generated numerical values, and predict the formula composition of the raw materials for adjusting the properties of the coating to achieve the target for the candidate. In particular, when it is necessary to perform fine color adjustment as described above or when the number of raw materials is large, when the correction formula range is large, the total number of combinations is sometimes more than 1 trillion (for example, when 4 kinds of raw materials are used at intervals of 0.01%), and there is a problem that it is difficult to obtain a predicted value and it is not practical. Therefore, by using a random combination as described above, set to 1 million at intervals of 0.01%, for example, combining in the order of arrangement, etc., the number of combinations can be reduced, and the calculation time for obtaining the predicted value can be shortened. In other cases, the total combination may also be 100,000, 10 million, or 100 million digits.
[0150] The value itself can also be randomly generated within the value generation range.
[0151] Here, it is preferred to perform multiple adjustments by adding raw materials for adjusting paint properties to the paint before adjustment, perform multiple predictions of the formula composition of the raw materials for adjusting paint properties for achieving the target, perform prescribed operations on the predicted formula composition of the raw materials for adjusting paint properties, or perform prescribed operations on the difference formula composition of the raw materials for adjusting paint properties obtained by using a difference target value with the target paint property as a baseline value, thereby determining the formula composition of the raw materials for adjusting paint properties for achieving the target.
[0152] As a case where a prescribed calculation is performed on the formula composition of the raw materials used to adjust the properties of the predicted paint, for example, a case where the predicted value is multiplied by a value such as 70% is illustrated. In the color matching of the paint, in the case of excessive addition of secondary primary colors, for example, when 1% of black raw material should be added to the whitish beige paint, but 2% of black raw material is added, it is necessary to add 99% of white raw material to restore the original state. It is known that the primary colors have deviations between batches. In addition, the prediction accuracy of artificial intelligence is not necessarily 100%, so it is easy to cause problems such as excessive addition of primary colors as mentioned above. Therefore, even if it is known that it is unqualified, by performing operations such as the above, it is possible to prevent excessive addition caused by deviations between batches, prediction errors, etc.
[0153] As a first example of using a target difference value, it can be set to a point that is 70% of the target ΔL*. For example, if ΔL*=3, and the target difference value is set to a point that is 70% of the target ΔL*, then through three adjustments, the difference from the target value can be reduced to less than 0.1 in the manner of 3→0.9→0.27→0.08. Even if the adjustment is performed by a colorist, it usually takes 3 to 4 times or more, so it can be completed using an equal or less number of adjustments. As a second example of using a target difference value, it can also be set to a point that is 800% of the target ΔL*, and the predicted value obtained can be multiplied by 10%.
[0154] In addition, it is preferred that the raw materials for adjusting the properties of the coating are added to the coating before adjustment multiple times, and the optimal formulation composition of the raw materials for adjusting the properties of the coating is predicted multiple times to achieve the target, and the formulation composition of 3 to 99% of the optimal formulation composition of the raw materials for adjusting the properties of the coating determined in the correction formulation prediction step is used. This method can also prevent excessive addition caused by deviations between batches, prediction errors, etc.
[0155] In addition, when it is known in advance that the prediction accuracy of a large change is higher than that of a small change, the target change value can be set to be larger than the difference from the target. For example, the difference target value is set to a point of 300% of the amount of the L* value to be changed, and a formula composition of 1 to 33% of the difference formula composition of the raw materials for adjusting the best coating properties determined in the correction formula prediction process can be used.
[0156] <Method for producing paint>
[0157] Figure 3 1 is a flow chart of a method for producing a coating material according to one embodiment of the present invention. Figure 4 1 is a flowchart showing a sub-process of a method for producing a coating material according to an embodiment of the present invention. Figure 3 , Figure 4 , an embodiment of a method for preparing a coating having a target coating property by adding a raw material for coating property adjustment to a coating before adjustment is exemplified. Figure 1 , Figure 2The same as the implementation method, it is preferred that the coating properties at least include the color of the coating, and further, preferably include the gloss and / or viscosity of the coating in addition to the color. In addition, in addition to color, gloss and viscosity, the coating properties can also include, for example, film thickness, contrast, angular chromatic effect, smoothness, visual appearance, resistance, contact angle, contamination, solar reflectance, UV transmittance, weather resistance, viscoelasticity, coating abnormality, wire drawing, foaming, etc. In addition, as an example, the predicted corrected formula composition is the numerical value of the amount of each primary color coating (black primary color coating, white primary color coating, blue primary color coating, yellow primary color coating, red primary color coating, etc.) of the raw material or the numerical value of the color, the numerical value of the amount of gloss regulator and / or viscosity regulator. For other details, refer to Figure 1 , Figure 2 The same embodiment as in the embodiment of the present invention is omitted for brevity. Figure 1 , Figure 2 The implementation method is the same, so the description is omitted.
[0158] like Figure 3 As shown, in this embodiment, first, relationship data representing the relationship between the coating properties of the coating before adjustment, the corrected formula composition of the raw materials used to adjust the coating properties, and the coating properties after adjustment (or the change in the coating properties) are prepared (step S301). Next, the relationship data representing the relationship between the coating properties of the coating before adjustment, the corrected formula composition of the raw materials used to adjust the coating properties, and the coating properties after adjustment (or the change in the coating properties) are input into the computer (step S302). Next, by using the machine learning of the computer, a prescribed artificial intelligence model is created that takes the data of the coating properties of the coating before adjustment and the data of the corrected formula composition of the raw materials used to adjust the coating properties as input, and takes the coating properties after adjustment (or the change in the coating properties) as output (step S303). Next, the data of the coating properties of the coating before adjustment and the corrected formula composition of the raw materials used to adjust the coating properties are input into the above-mentioned artificial intelligence model through the computer, and the coating properties after adjustment (or the change in the coating properties) are calculated for prediction (step S304). Next, the modified formula composition obtained based on the predicted multiple adjusted paint properties (or the changes in paint properties) is determined as the formula composition of the raw materials for adjusting the paint properties to achieve the target (step S305). Figure 2 The implementation method is the same as (for steps S301 to S303, also the same as Figure 1 ), so the detailed description is omitted.
[0159] like Figure 3As shown, in this embodiment, next, raw materials having the predicted corrected recipe composition (the recipe composition determined in step S305) are added to the paint before adjustment to adjust the paint (adjustment process: step S306).
[0160] Next, the properties of the adjusted coating are measured to determine whether it is qualified (qualified or not determination process: step S307). The color can be measured using any known colorimeter, etc. In addition, the gloss can be measured using any known gloss meter, and the viscosity can be measured using any known viscometer. Other coating properties can also be measured by any known measurement means. The measurement method of each coating property is not particularly limited, and it is also possible to measure, for example, an object on which the coating is actually applied to a predetermined object to be coated, or to directly measure the coating before coating (liquid colorimetry). In the case of coating (applying), there can be various coating methods such as rod coating, roller coating, spray coating, dip coating, electrophoretic coating, printing coating, curtain flow coating, brush coating, roller coater coating, airless spray coating, powder coating, offset printing, flexographic printing, electrostatic coating, air spray coating, smear coating, spin coating, etc.
[0161] Then, the correction formula composition prediction process (step S305), the adjustment process (step S306), and the pass / fail judgment process (step S307) are repeated until the judgment result in the pass / fail judgment process (step S307) is passed. From among the predicted multiple adjusted paint properties, the correction formula composition corresponding to the adjusted paint property whose difference with the target paint property (for example, with the E value) is within a specified range (for example, ΔE is less than 0.1, gloss is less than 0.2, etc.) can be re-determined as the formula composition of the raw materials for adjusting the paint property to achieve the target (as an example, it is re-determined in a way that ΔE is the second smallest), or by using other criteria, the formula composition of the raw materials for adjusting the paint property to achieve the target can also be re-selected. Such a series of color adjustments can also be set up on a continuous flow production line with a colorimeter, while continuously monitoring the color changes, while adjusting the deviation from the standard color.
[0162] In this embodiment, it is preferred that after the pass / fail judgment process (step S307), relational data representing the relationship between the data of the coating properties of the coating before adjustment, the predicted modified formula composition, and the data of the coating properties of the adjusted coating measured in the pass / fail judgment process (step S307) are input into the computer, and the relational data is updated. Thus, by updating the relational data, an artificial intelligence model based on more data can be produced. The update can be performed just after new data is obtained, or it can be performed regularly or in a timely manner. However, this process is not necessary and can be omitted.
[0163] In this embodiment, when the result of the pass / fail judgment process (step S307) is that the number of times the pass / fail judgment has been reached, a sub-process (step S308) is performed. Figure 4 Therefore, the following reference Figure 4 Provide explanation.
[0164] like Figure 4 As shown, in the sub-process (step S308), through computer-based machine learning, another artificial intelligence model is produced, which takes the data of the paint properties of the paint before adjustment and the data of the corrected formula composition of the raw materials used to adjust the paint properties as input, and the paint properties of the paint after adjustment (or the change in the paint properties) as output (artificial intelligence model production sub-process: step S308-1). Next, through the computer, in the above-mentioned other artificial intelligence model, the data of the paint properties of the paint before adjustment and the corrected formula composition are input, so as to calculate the other adjusted paint properties (or the change in the paint properties) for prediction (paint property prediction sub-process: step S308-2). In the paint property prediction sub-process (step S308-2), multiple corrected formula compositions selected in advance as candidates are input, and multiple adjusted paint properties (or changes in paint properties) corresponding to each corrected formula composition are predicted. Next, based on the predicted multiple adjusted paint properties (or the amount of change in paint properties), another optimal correction formula composition is determined as the formula composition of the raw materials for adjusting the paint properties to achieve the target (correction formula prediction sub-process: step S308-3). Next, the raw materials with the predicted other correction formula composition are added to the paint before adjustment to adjust the paint (adjustment sub-process: step S308-4). Next, the adjusted paint properties are measured and judged whether they are qualified (qualified or not judgment sub-process: step S308-5). Then, the paint property prediction sub-process (step S308-2), the correction formula prediction sub-process (step S308-3), the adjustment sub-process (step S308-4), and the qualified or not judgment sub-process (step S308-5) are repeated until the judgment result of the qualified or not judgment sub-process (step S308-5) is qualified.
[0165] In addition to using another artificial intelligence model to obtain another prediction result, each sub-process can also Figure 3 The corresponding steps of the present step shown are performed in the same manner.
[0166] exist Figure 3 The pass / fail judgment process of this process shown or Figure 4When the result of the judgment of the sub-process is that the sub-process is qualified, the flow of this embodiment ends (step S309). In addition, in the present invention, it is not necessary to perform the sub-process.
[0167] According to the coating manufacturing method of the present embodiment, by using an artificial intelligence model to input data, a predicted value of the modified formula composition can be obtained, and this can be repeated until the coating properties are close to the specified target value with high accuracy, so that efficiency can be achieved compared with the situation where, for example, a colorist repeatedly adjusts and measures each time according to a job management table. In addition, the artificial intelligence model takes the data of the coating properties of the coating before adjustment and the data of the modified formula composition of the raw materials for adjusting the coating properties as input, and takes the adjusted coating properties (or the change in the coating properties) as output. Furthermore, in the present embodiment, a plurality of modified formula compositions selected in advance as candidates are input, and a plurality of predicted values of coating properties (or the change in the coating properties) are obtained as output, and based on the predicted value (in this example, in a manner that minimizes the difference with the target value), the modified formula composition corresponding to the adjusted coating properties (or the change in the coating properties) is determined as the formula composition of the raw materials for adjusting the coating properties for achieving the best target. Therefore, regardless of whether the information about the raw materials of the paint before adjustment is useful for predicting the properties of the paint at the current moment, the data of the paint properties of the paint before adjustment actually measured can be used as input to obtain the above-mentioned prediction value. Even if the paint properties at the current moment change from the paint properties obtained from the initial raw materials, this will not affect the calculation of the prediction value, and the paint properties can be made to approach the specified target value with high precision.
[0168] By simultaneously predicting gloss and / or viscosity in addition to color as paint properties, it is possible to predict with higher accuracy taking into account their mutual influence, so that the paint properties can be closer to the desired paint properties.
[0169] In addition, by using the change in paint properties as the output of a specified artificial intelligence model, the errors or deviations caused by the measuring instrument can be offset, which can further improve the accuracy of the prediction and make the paint properties closer to the specified target value.
[0170] <Prediction system for paint properties>
[0171] Figure 5 1 is a block diagram of a system for predicting coating properties according to an embodiment of the present invention. This system is a system for predicting the coating properties of a coating after adjustment when a raw material for adjusting coating properties is added to the coating before adjustment. Figure 5As shown, the coating property prediction system 10 of the present embodiment has a computer 11. The computer 11 has a machine learning function. In addition, the computer 11 has an artificial intelligence model acquisition unit 12 and a coating property prediction unit 13. The artificial intelligence model acquisition unit 12 uses machine learning to create a prescribed artificial intelligence model that takes the data of the coating property of the coating before adjustment and the data of the modified formula composition of the raw materials used for coating property adjustment as input, and takes the coating property after adjustment (or the change in the coating property) as output. The artificial intelligence model acquisition unit 12 can be set as a communication unit that obtains the artificial intelligence model from the outside through communication or the like. In the present embodiment, it is set to obtain the artificial intelligence model through the artificial intelligence acquisition unit 12 of the computer 11, but in the case of obtaining it from the outside, it is not necessarily necessary to use the computer 11, and it can also be obtained by being transferred by others. The coating property prediction unit 13 inputs the data of the coating property of the coating before adjustment and the modified formula composition in the artificial intelligence model, thereby calculating the coating property after adjustment (or the change in the coating property) for prediction. The artificial intelligence model acquisition unit 12 and the coating property prediction unit 13 can be set as processors.
[0172] The above-mentioned relationship data is input to the computer 11. In the present embodiment, the computer 11 has a storage unit 14 (memory) for storing the relationship data and the artificial intelligence model obtained (produced in this example) and a communication unit 15 for sending and receiving the relationship data. The communication unit 15 is capable of sending and receiving not only the relationship data but also other data. Preferably, the artificial intelligence model acquisition unit 12 has a function of performing machine learning using the input relationship data as learning data (e.g., supervised data). In addition, the computer 11 preferably also has a functional unit for preprocessing the relationship data. The details of the preprocessing are the same as those described in the embodiment of the method for predicting coating properties, so the re-explanation is omitted. In addition, it is preferred that the prediction system 10 has a display unit (display) for displaying the prediction results.
[0173] According to the prediction system 10, data can be inputted using an artificial intelligence model to obtain a predicted value of the coating properties (or the amount of change thereof), thereby achieving high efficiency. In addition, the artificial intelligence model takes the data of the coating properties of the coating before adjustment and the data of the modified formula composition of the raw materials used for adjusting the coating properties as input, and takes the adjusted coating properties (or the amount of change of the coating properties) as output, so regardless of whether the information of the raw materials of the coating before adjustment is useful for predicting the coating properties at the current moment, the data of the coating properties of the coating before adjustment that are actually measured can be used as input to obtain the above-mentioned predicted value, and even if the coating properties at the current moment change from the coating properties obtained from the initial raw materials, this will not affect the calculation of the predicted value, and the adjusted coating properties can be predicted with high accuracy.
[0174] As coating properties, gloss and / or viscosity are predicted simultaneously in addition to color, thereby enabling predictions to be made with higher accuracy taking into account their mutual influence.
[0175] In addition, by using the changes in paint properties as the output of a specified artificial intelligence model, errors or deviations caused by measuring instruments can be offset, further improving the accuracy of predictions.
[0176] <Prediction system for correcting recipe composition>
[0177] Figure 6 This is a block diagram of a prediction system for correcting a formula composition according to an embodiment of the present invention. This system is a system for predicting the formula composition of the raw materials for adjusting the properties of the paint to achieve the target paint properties when the raw materials for adjusting the properties of the paint are added to the paint before adjustment to prepare the paint having the target paint properties. Figure 6 As shown, the prediction system 20 for the modified recipe composition of this embodiment is composed of a computer 21. The computer 21 has a machine learning function. In addition, the computer 21 has an artificial intelligence model acquisition unit 22, a coating property prediction unit 23, a storage unit 24, and a communication unit 25. These components are similar to those in Figure 5 The artificial intelligence model acquisition unit 12, the paint property prediction unit 13, the storage unit 14, and the communication unit 15 in the embodiment of the correction recipe composition shown are the same as those described above, so the detailed description is omitted.
[0178] The paint property prediction unit 23 is configured to receive a plurality of correction recipe compositions selected in advance as candidates, and predict a plurality of adjusted paint properties (or changes in paint properties) corresponding to the respective correction recipe compositions.
[0179] The computer 21 also has a modified formula composition prediction unit 26, which determines the modified formula composition obtained based on the predicted multiple adjusted paint properties (or the change in paint properties) as the formula composition of the raw materials used to adjust the paint properties to achieve the target. The modified formula composition prediction unit 26 can be set as a processor. In addition, the modified formula composition prediction unit 26 is configured to be able to determine the modified formula composition corresponding to the adjusted paint properties (or its change) as the formula composition of the raw materials used to adjust the paint properties to achieve the target from among the predicted multiple adjusted paint properties, in a manner that the difference with the target paint properties is within a specified range (for example, ΔE is less than 0.1, the gloss is less than 0.2, etc.). In addition, it is preferred that the computer 21 also has a functional unit for preprocessing the relational data. The details of the preprocessing are the same as those described in the embodiment of the method for predicting paint properties, so the description is omitted again. In addition, it is preferred that the present prediction system 20 has a display unit (display) that displays the prediction results.
[0180] According to the prediction system of the modified formula composition of the present embodiment, it is possible to obtain the predicted value of the modified formula composition by using the artificial intelligence model input data, so that it is possible to achieve high efficiency. In addition, the artificial intelligence model uses the data of the coating properties of the coating before adjustment and the data of the modified formula composition of the raw materials for adjusting the coating properties as input, and uses the adjusted coating properties (or the change in coating properties) as output. Furthermore, in the present embodiment, a plurality of modified formula compositions selected in advance as candidates are input, and the predicted values of a plurality of coating properties (or the change in coating properties) are obtained as output, and the modified formula composition corresponding to the adjusted coating properties is determined according to the predicted value (in this example, in a manner such that the difference with the target value becomes within a prescribed range (for example, ΔE is less than 0.1, the gloss is less than 0.2, etc.) as the best formula composition of the raw materials for adjusting the coating properties to achieve the target. Therefore, regardless of whether the information about the raw materials of the paint before adjustment is useful for predicting the paint properties at the current moment, the data of the paint properties of the paint before adjustment actually measured can be used as input to obtain the above-mentioned prediction value. Even if the paint properties at the current moment change from the paint properties obtained from the initial raw materials, this will not affect the calculation of the prediction value, and the corrected formula composition for achieving the target paint properties can be predicted with high accuracy.
[0181] By simultaneously predicting gloss and / or viscosity as paint properties in addition to color, it is possible to predict with even higher accuracy in consideration of their mutual influence.
[0182] In addition, by using the changes in paint properties as the output of a specified artificial intelligence model, it is possible to offset the errors and deviations caused by the measuring instrument, further improving the accuracy of the prediction.
[0183] Here, the computer 21 preferably further includes a pass / fail judgment unit 27, which adds raw materials having a predicted modified formula composition to the paint before adjustment, and judges whether the paint is pass / fail based on the result of measuring the paint properties after adjustment. The pass / fail judgment unit 27 can be set as a processor. Thus, the process of the paint manufacturing method of the above-mentioned embodiment can be executed.
[0184] Here, preferably, similarly to the implementation method, a numerical value generation range of the recipe quantity is set for each raw material of the modified recipe composition, and the coating property prediction unit is input with multiple modified recipe compositions formed by combining the numerical values of the recipe quantities of each raw material generated within the set numerical value generation range.
[0185] Furthermore, it is preferred that an allowable range relative to the predicted correction recipe composition is determined, and the correction recipe prediction unit determines a correction recipe composition within the allowable range from the predicted correction recipe composition as a recipe composition of raw materials for adjusting target coating properties.
[0186] Alternatively, the numerical value generation range may be set to be the same as the allowable range, and a plurality of corrected recipe compositions may be input by combining numerical values of recipe quantities of each raw material generated within the set numerical value generation range (allowable range).
[0187] In addition, it is also preferred to perform multiple adjustments of adding raw materials for adjusting paint properties to the paint before adjustment, perform multiple predictions of the formula composition of the raw materials for adjusting paint properties to achieve the target, and determine the formula composition of the raw materials for adjusting paint properties to achieve the target by performing prescribed operations on the predicted formula composition of the raw materials for adjusting paint properties, or performing prescribed operations on the difference formula composition of the raw materials for adjusting paint properties obtained by using a difference target value with the target paint property as a baseline value.
[0188] The details of these are the same as those described in the embodiment of the method, and thus re-description is omitted. In addition, similarly to the embodiment of the method, (the absolute value of) the coating property can be used instead of the amount of change in the coating property.
[0189] Hereinafter, examples of the present invention will be described, but the present invention is not limited to the following examples at all.
[0190] Example
[0191] In order to clarify the effects of the present invention, predictions of Examples 1 to 11 and predictions of Comparative Examples 1 and 2 were performed, and thus they will be described below.
[0192] First, the method for producing the coating composition before adjustment used in Examples 1 to 11 and Comparative Examples 1 and 2 will be described. First, the method for producing the primary color coating will be described.
[0193] <Original color production example 1>
[0194] 39 parts by mass of acrylic resin (solid content: 40% by mass), 15 parts by mass of isophorone, and 46 parts by mass of titanium oxide were mixed and dispersed using a sand mill (dispersion medium: glass beads) until the maximum particle size of the pigment coarse particles was 10 μm or less to prepare a white primary paint.
[0195] <Primary color production examples 2 to 5>
[0196] Black primary color paints 1 and 2, yellow primary color paints, and red primary color paints were prepared as shown in Table 1 below by the method of Production Example 1 except that the pigments were changed to black calcined pigments, yellow iron oxide, and red iron oxide, respectively.
[0197] Gloss regulator
[0198] Adjustment was performed using GASIL HP395 (gloss adjuster 1) manufactured by INEOS SILICAS or SYLYSIA 435 (gloss adjuster 2) manufactured by Fuji Silicone Chemical Co., Ltd.
[0199] Viscosity regulator
[0200] Use isophorone.
[0201] [Table 1]
[0202] Table 1 Primary color paint production example
[0203] Contents of coating properties White primary paint Black primary paint 1 Black primary paint 2 Yellow primary paint Red primary paint Acrylic resin 39 39 39 39 39 Isophorone 15 15 15 15 15 Titanium oxide 46 0 0 0 0 Black Burnt Pigment 1 0 46 0 0 0 Black Burnt Pigment 2 0 0 46 0 0 Yellow burnt pigment 0 0 0 46 0 Iron oxide red 0 0 0 0 46
[0204] The coating compositions before adjustment were prepared as follows using the original color coatings shown in Table 1. The coating compositions before adjustment are summarized in Table 2 below.
[0205] ·Preparation Example 1 of Coating Composition Before Adjustment
[0206] 31 parts by mass of acrylic resin (solid content concentration: 40% by mass), 3 parts by mass of cyclohexanone, 11 parts by mass of xylene, 5.5 parts by mass of KTL-20N (produced by Kitamura Co., Ltd.), 85 parts by mass of white primary color, 1 part by mass of black primary color, 13 parts by mass of yellow primary color, 1 part by mass of red primary color, 3 parts by mass of gloss modifier, and 20 parts by mass of isophorone were added and uniformly mixed with a disperser to prepare a coating composition 1 before adjustment (solid content concentration: 40% by mass).
[0207] ·Preparation Example 2 of Coating Composition Before Adjustment
[0208] Pre-adjustment coating compositions 2 to 4 (solid content concentration: 40% by mass) were prepared in the same manner as in Production Example 1 except that the primary color coating was added as shown in Table 2 below.
[0209] [Table 2]
[0210] Table 2 Production Example
[0211] Coating composition before adjustment 1 Coating composition before adjustment 2 Coating composition before adjustment 3 Coating composition before adjustment 4 Acrylic resin 31 31 31 31 Cyclohexanone 3 3 3 3 Xylene 11 11 11 11 KTL-20N 5.5 5.5 5.5 5.5 White primary paint 85 2 50 71 Black primary paint 1 1 0 25 4 Black primary paint 2 0 77 0 0 Yellow primary paint 13 7 18 11 Red primary paint 1 14 7 14 Gloss regulator 1 3 0 4 0 Gloss Regulator 2 0 3 0 2 Isophorone 20 20 20 20
[0212] <Method for producing coating film>
[0213] As the primer, Fine Tough G primer (primer 1) was applied by bar coating to a dry film thickness of 5 μm, and then sintered for 60 seconds at a condition where the maximum temperature of the material (galvanized-aluminum alloy steel plate) was 210° C. to form a primer coating film. Next, the coating composition 1 was applied by bar coating or roller coating to a dry film thickness of 15 μm, and then sintered for 60 seconds at a predetermined maximum temperature of the material, and then immediately cooled to form a coating film of the pre-adjustment coating composition.
[0214] <Measurement method>
[0215] Each measurement described in each Example and Comparative Example was measured by the method described below.
[0216] Determination of color data
[0217] For each coating film described in the method described in the method for producing the coating film, the color was measured using various color difference meters such as a colorimeter CR-300 (manufactured by Konica Minolta Inc.), a spectrophotometer LabScan XE (manufactured by HunterLab, Inc.), a spectrocolorimeter SE6000 (manufactured by Nippon Denshoku Industries Co., Ltd.), or a colorimeter SM-T (manufactured by Suga Testing Instruments Co., Ltd.).
[0218] Determination of gloss data
[0219] The gloss of each coating film described in the method described in the method for producing a coating film was measured using VG7000 (manufactured by Nippon Denshoku Industries, Ltd.).
[0220] Determination of viscosity data
[0221] The viscosity of each coating was measured in seconds using a Ford Cup #4 at 25°C.
[0222] · Obtaining coating properties of coating composition before and after adjustment
[0223] The primer (primer) is made of Nippon Paint Co., Ltd.
[0224] Fine Tough G Primer (Primer 1)
[0225] Fine Tough C Primer (Primer 2)
[0226] NSC 4800NC Primer (Primer 3)
[0227] Any one of the three epoxy resin primers having different color tones is used to form a coating film of the coating composition before adjustment and the coating composition after adjustment by the method described in the coating film production method.
[0228] The color of the coating film obtained is used
[0229] CR-400 (manufactured by Konica Minolta) (Colorimeter 1)
[0230] SE6000 (manufactured by Nippon Denshoku Industries Co., Ltd.) (Colorimeter 2)
[0231] Lab Scan XE (manufactured by HunterLab) (Colorimeter 3)
[0232] The colorimeter is used to measure the L*, a*, and b* values, or the difference in L*, a*, and b* values with the target standard plate.
[0233] ·Obtaining the coating composition after adjustment
[0234] The obtained pre-adjustment coating composition and raw materials for adjusting coating properties (each primary color coating, a gloss adjuster, and isophorone) are added to obtain a post-adjustment coating composition having target coating properties.
[0235] <Acquisition of learning data>
[0236] Acquisition of learning data for AI model building
[0237] 27 colors of paints (paint properties before adjustment: L* value 24-94, a* value -3-10, b* value -2-37, gloss 3-55, viscosity 130-280 seconds), 14-22 μm film thickness, 6 batches each, a total of 162 paints were prepared by the same method as the above-mentioned "Pre-adjustment Paint Composition Manufacturing Example". The coating method is 3 batches of rod coating and 3 batches of roller coating. Adjustment is repeated until qualified. According to the adjustment history of the paint properties, all 5829 relationship data for the production of the artificial intelligence model of the paint properties of the paint composition before adjustment, the corrected formula composition of the raw materials used for adjusting the paint properties, and the paint properties of the paint composition after adjustment are input into the computer. At this time, the color value, gloss, viscosity and film thickness are used in the paint composition properties before adjustment, and the color value, gloss, viscosity, and film thickness changes due to adjustment are used in the paint properties of the paint composition after adjustment. The corrected formula composition of each raw material used to adjust the properties of the coating material uses the value obtained by dividing each corrected formula amount by the total amount of the coating material. Furthermore, as the related condition data, the sintering temperature of the sintering process (varied between 240°C and 260°C), the type of primer, and the colorimeter are recorded.
[0238] At this time, colorimeter 1 to 3 were used randomly for each batch. Similarly, primer 1 to 3 were used randomly for each batch.
[0239] <Evaluation Method>
[0240] (Estimated accuracy)
[0241] The estimation accuracy is calculated based on the error obtained by comparing the measured value of the color data of this embodiment with the predicted value. Based on the measured value and the predicted value, the determination coefficient of the scatter plot composed of the measured value and the predicted value of each estimation object is calculated. The determination coefficient is represented by (1-(residual sum of squares / total sum of squares)).
[0242] In Examples 1 to 11 and Comparative Examples 1 and 2, calculation was performed based on the determination coefficients of the estimated objects (L* value, a* value, b* value, gloss, viscosity, and film thickness) shown in Table 3.
[0243] It can be said that the closer the value of the determination coefficient is to 1, the higher the accuracy of the learned artificial intelligence model.
[0244] Here, the high accuracy of the learned artificial intelligence model means that the accuracy of the prediction of the paint property prediction system is high.
[0245] (Color adjustment times)
[0246] The number of color adjustments indicates the average number of color adjustments until a qualified color is obtained when 5 pre-adjustment paints are prepared with slightly changed initial material addition formulas and adjusted according to the corrected formula composition predicted by artificial intelligence. Here, a small number of adjustments means that the prediction accuracy of the paint property prediction system and the calculation method of the corrected formula composition are high.
[0247] (Production rate)
[0248] The production volume ratio indicates the ratio of the adjusted production volume when the predetermined production volume is set to 1. Here, the production volume ratio does not greatly exceed 100%, which means that the accuracy of the prediction of the paint property prediction system and the accuracy of the calculation method of the corrected formula composition are high, and although a slight error is allowed, it is an efficient system that does not produce useless paint.
[0249] <Criteria for judging the quality of paint adjustment>
[0250] Satisfaction of all three of the following conditions will be considered a pass.
[0251] The differences in L* value, a* value, and b* value between the standard plate and the plate coated with the adjustment paint are denoted as ΔL*, Δa*, and Δb*.
[0252] ·ΔE=√(ΔL *2 +Δa *2 +Δb *2 ) is less than 0.1, and ΔL*, Δa*, and Δb* are each less than 0.1.
[0253] For the Δ gloss value, the difference from the standard plate is 0.5 or less when the gloss value is 10 or less, 1 or less when the gloss value is 10 to 20, and 3 or less when the gloss value is 30 or more.
[0254] The difference between the target value and the measured viscosity is less than 10 seconds.
[0255] Next, predictions of each of the Examples and Comparative Examples will be described.
[0256] <Example 1>
[0257] Data cleaning, zero point introduction, combination, exponentiation, normalization and other preprocessing were performed in advance. A neural network was used to create a prescribed artificial intelligence model in a computer, using the L* value, a* value, b* value, gloss, and viscosity data of the paint properties of the paint composition before adjustment of the relational data, the data of the corrected formula of the raw materials for adjusting the paint properties, the colorimeter, and the type of primer as explanatory variables, and the change in the L* value of the paint composition after adjustment as the target variable.
[0258] Similarly, artificial intelligence is created to predict the change values of a* and b*. After obtaining a total of three artificial intelligences, the estimation accuracy of each is calculated, and the average estimation accuracy is confirmed.
[0259] Then, in the revised formula composition of the raw materials used to adjust the properties of each paint, the four raw materials, white primary color, black primary color 1, yellow primary color, and red primary color, were set with an allowable range, and 100,000 numbers were generated within each range, and then randomly combined to obtain 100,000 candidate data. The candidate data was input to the three artificial intelligences that had completed learning to obtain the predicted values of the three paint properties.
[0260] The point where the color difference from the standard plate is 70% is taken as the target value, and a correction formula of 4 raw materials that meets the target value and the change amount of L, a, and b is obtained from 100,000, and the color is adjusted. The 4 raw materials other than the set allowable range are not used for correction.
[0261] Since the target gap value was used, it was unqualified. The result of the second color adjustment with the target gap value of 90% was unqualified. The target gap value was used again, so it was unqualified. The result of the third color adjustment with the target gap value of 100% (no target gap) was qualified. The results are shown in Table 3.
[0262] <Comparative Example 1>
[0263] The same method as in Example 1 was followed except that the formulation composition data used for prediction was used as the overall ratio of the paint before adjustment, the predicted paint properties after adjustment were used as absolute values, and data related to the paint properties of the original paint were not used.
[0264] <Example 2>
[0265] The gloss and viscosity were estimated by adding gloss modifier 1 and isophorone to the raw materials within the allowable range. The same procedure as in Example 1 was used as the estimated accuracy of the viscosity.
[0266] <Comparative Example 2>
[0267] The same procedure as in Example 2 was carried out except that the number of corrected recipe compositions was changed to 1,000.
[0268] <Example 3>
[0269] The same procedure as in Example 2 was carried out except that the raw material 3 with a set allowable range was used and the number of corrected recipe compositions was changed to 1000.
[0270] <Example 4>
[0271] The same procedure as in Example 2 was carried out except that the predicted coating properties after adjustment were expressed as absolute values.
[0272] <Example 5>
[0273] The same procedure as in Example 2 was carried out except that the learning data and the prediction target were set to one paint color and data on the paint properties of the original paint were not used.
[0274] <Example 6>
[0275] The same procedure as in Example 2 was carried out except that PLS was used.
[0276] <Example 7>
[0277] The same procedure as in Example 2 was followed except that gradient boosting was used.
[0278] <Example 8>
[0279] The same procedure as in Example 2 was carried out except that random numerical value generation and random combination were not performed, and the correction formulation amounts were set at equal intervals within the allowable range so that the total amount of correction formulations was 100,000, with 10 stages for the white primary color, the black primary color 1, the yellow primary color, and the red primary color, 5 stages for the gloss adjuster 1, and 2 stages for isophorone.
[0280] <Example 9>
[0281] The same method as in Example 2 was carried out except that no difference target was set and the color difference with the standard plate was targeted to be zero from the first time.
[0282] <Example 10>
[0283] The same method as in Example 2 is used except that the numerical generation range and the allowable range of the recipe quantity are not set, random combination is not performed, and candidate correction recipes calculated by the operator with reference to past adjustment records are input into the artificial intelligence for prediction to determine the correction recipe.
[0284] <Example 11>
[0285] The same procedure as in Example 1 was carried out except that the sintering temperature in the sintering step was used as an explanatory variable and the prediction of the change in film thickness was added as the change in coating material properties.
[0286] [Table 3]
[0287]
[0288] In addition, "NN" in Table 3 means neural network, "PLS" means partial least squares regression, and "GB" means gradient boosting.
[0289] Description of Reference Numerals
[0290] 10: Prediction system;
[0291] 11: Computer;
[0292] 12: Artificial intelligence model acquisition department;
[0293] 13: Coating properties prediction department;
[0294] 14: storage unit;
[0295] 15: Ministry of Communications;
[0296] 20: Prediction system;
[0297] 21: Computer;
[0298] 22: Artificial intelligence model acquisition department;
[0299] 23: Coating property prediction department;
[0300] 24: storage unit;
[0301] 25: Ministry of Communications;
[0302] 26: Modify the formula composition prediction department;
[0303] 27: Pass / Fail Judgment Department.
Claims
1. A method for predicting coating properties, characterized in that: The coating properties after adjustment are predicted by a machine learning method when a raw material for adjusting the coating properties is added to the coating before adjustment. The coating properties prediction method comprises: The paint property prediction process is a process in which the paint property data of the paint before adjustment and the data of the corrected formula composition of the raw materials used for adjusting the paint properties are input as input, and the paint property after adjustment is output, and prediction is performed by inputting the paint property data of the paint before adjustment and the corrected formula composition into a computer, thereby calculating the paint property after adjustment.
2. A method for predicting coating properties, characterized in that: The method of predicting the change in the properties of the coating after adjustment when adding a raw material for adjusting the properties of the coating to the coating before adjustment is performed by a machine learning method, wherein the method of predicting the properties of the coating comprises: The paint property prediction process is a process in which the paint property data of the paint before adjustment and the data of the corrected formula composition of the raw materials used for adjusting the paint properties are input as input, and the change amount of the paint property after adjustment is output, and prediction is performed by inputting the paint property data of the paint before adjustment and the corrected formula composition into a computer, thereby calculating the change amount of the paint property after adjustment.
3. A method for predicting a modified recipe composition, characterized in that: By using a machine learning method, when a raw material for adjusting the properties of a coating is added to a coating before adjustment to prepare a coating having a target coating property, a formula composition of the raw material for adjusting the properties of the coating to achieve the target coating property is predicted. The method for predicting the corrected recipe composition comprises calculating the adjusted paint properties for prediction by the paint property prediction step of the paint property prediction method according to claim 1, In the coating property prediction step, a plurality of data on coating property of the coating before adjustment and the corrected recipe composition are input, and a plurality of coating properties after adjustment corresponding to each of the corrected recipe compositions are predicted. The method for predicting a corrected recipe composition further includes a corrected recipe prediction step of determining the corrected recipe composition obtained based on the predicted plurality of paint properties after adjustment as a recipe composition of raw materials for adjusting paint properties for achieving the target.
4. A method for predicting a modified recipe composition, characterized in that: By using a machine learning method, when a raw material for adjusting the properties of a coating is added to a coating before adjustment to prepare a coating having a target coating property, a formula composition of the raw material for adjusting the properties of the coating to achieve the target coating property is predicted. The method for predicting the corrected recipe composition comprises: calculating the change amount of the adjusted paint property to make a prediction through the paint property prediction step of the paint property prediction method according to claim 2; In the coating property prediction step, a plurality of data on the coating property of the coating before adjustment and the correction formula composition are input, and a plurality of changes in the coating property after adjustment corresponding to each of the correction formula compositions are predicted. The method for predicting a corrected recipe composition further includes a corrected recipe prediction step of determining the corrected recipe composition obtained based on the predicted changes in the plurality of adjusted paint properties as a recipe composition of raw materials for adjusting the paint properties to achieve the target.
5. The method according to claim 2 or 4, characterized in that: Also includes: a relational data input step of inputting into the computer relational data indicating the relation between the corrected formulation composition of the raw materials for adjusting the properties of the coating and the amount of change in the properties of the coating after adjustment; and an artificial intelligence model acquisition step, acquiring the predetermined artificial intelligence model, In the artificial intelligence model acquisition step, the predetermined artificial intelligence model is produced by using the computer to perform machine learning using the relationship data input in the relationship data input step as learning data.
6. The method according to any one of claims 1 to 4, characterized in that: The prescribed artificial intelligence model also contains as input data on the manufacturing conditions of the coating, Contains any one or more of the following (a) to (d): (a) any one or more of the viscosity and non-volatile content of the coating; (b) any one or more of a roller peripheral speed in the step of applying the coating to an object, a coating pressure applied to the object, and a flow rate of the coating; (c) one or more of the sintering temperature and the sintering time in the sintering step; and (d) One or more of the temperature and humidity of the production line.
7. The method according to any one of claims 1 to 4, characterized in that: The paint properties include at least the color of the paint.
8. The method according to claim 7, characterized in that The coating properties include the gloss and / or viscosity of the coating in addition to the color.
9. The method according to any one of claims 1 to 4, 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 machines, random forests, gradient boosting, K-nearest neighbor method, neural networks, Bayesian estimation, or ensemble learning thereof.
10. The method according to claim 9, characterized in that The machine learning uses a prediction algorithm based on a neural network.
11. The method according to claim 4, characterized in that From among the multiple changes in the adjusted paint properties predicted in the corrected recipe prediction process, the corrected recipe composition corresponding to the change in the adjusted paint properties whose difference from the target paint properties is below a prescribed baseline value is determined as the recipe composition of the raw materials for adjusting the paint properties to achieve the target.
12. The method according to claim 4 or 11, characterized in that: The method further comprises the step of setting a numerical value generation range of the formula quantity for each raw material in the modified formula composition. In the coating material property prediction step, the plurality of corrected recipe compositions are inputted by combining numerical values of recipe quantities of respective raw materials generated within the set numerical value generation range.
13. The method according to claim 12, characterized in that The method further comprises the step of determining an allowable range of the predicted modified formulation composition. From the corrected recipe compositions predicted in the corrected recipe prediction step, a recipe composition within the allowable range is determined as a recipe composition of a raw material for adjusting the properties of the coating material to achieve the target.
14. The method according to claim 4, characterized in that Adding the raw material for adjusting the properties of the coating to the coating before adjustment is performed multiple times, and predicting the formula composition of the raw material for adjusting the properties of the coating to achieve the target is performed multiple times, The formula composition of the raw materials for adjusting the paint properties to achieve the target is determined by performing prescribed operations on the predicted formula composition of the raw materials for adjusting the paint properties, or by performing prescribed operations on the difference formula composition of the raw materials for adjusting the paint properties obtained by using the difference target value with the target paint property as a baseline value.
15. A method for producing a coating, further comprising: an adjustment step of adding a raw material having the corrected formula composition predicted by the method according to claim 4 to the paint before adjustment, and adjusting the paint; and The pass / fail judgment process measures the properties of the adjusted coating and judges whether it is pass / fail. The modified recipe composition prediction step, the adjustment step, and the pass / fail judgment step are repeatedly performed until the pass / fail judgment step results in a pass / fail judgment result.
16. The method for producing a coating according to claim 15, characterized in that: It also includes: a relationship data updating step of inputting into the computer relationship data representing the relationship between the predicted modified formula composition and the data on the amount of change in the adjusted paint property measured in the pass / fail judgment step, and updating the relationship data.
17. The method according to any one of claims 1 to 4, 11, 14 to 16, characterized in that: The coating material is a coating material for coil coating.
18. A system, characterized in that: By using machine learning methods, the paint properties of the paint after adjustment are predicted when raw materials for adjusting paint properties are added to the paint before adjustment. The system comprises a computer, wherein the computer comprises a paint property prediction unit, wherein the paint property prediction unit predicts the paint property after adjustment by inputting the paint property data of the paint before adjustment and the corrected formula composition data of raw materials for adjusting the paint property as inputs and the paint property after adjustment as outputs, thereby calculating the paint property after adjustment.
19. A system, characterized in that: By using machine learning methods, we can predict the change in the properties of the paint after adjustment when adding raw materials for adjusting the properties of the paint to the paint before adjustment. The system comprises a computer, wherein the computer comprises a paint property prediction unit, wherein the paint property prediction unit predicts by inputting the paint property data of the paint before adjustment and the corrected formula composition data of raw materials for adjusting the paint properties as inputs, and the change amount of the paint properties after adjustment as outputs, thereby calculating the change amount of the paint properties after adjustment.
20. A system, characterized in that: By using a machine learning method, a formula composition of raw materials for adjusting the properties of a coating material is predicted for achieving a target coating material property when a coating material having a target coating material property is prepared by adding raw materials for adjusting the properties of the coating material to the coating material before adjustment. The system is configured such that the paint property prediction unit of the system according to claim 18 calculates and predicts the paint property after adjustment, The paint property prediction unit is configured to receive a plurality of paint property data of the paint before adjustment and the corrected recipe composition, and predict a plurality of paint properties after adjustment corresponding to each of the corrected recipe compositions. The system includes a computer, and the computer also includes a modified recipe composition prediction unit, which determines the modified recipe composition obtained based on the predicted plurality of adjusted paint properties as the recipe composition of raw materials for adjusting the paint properties to achieve the target.
21. A system, characterized in that: By using a machine learning method, a formula composition of raw materials for adjusting the properties of a coating material is predicted for achieving a target coating material property when a coating material having a target coating material property is prepared by adding raw materials for adjusting the properties of the coating material to the coating material before adjustment. The system is configured such that the paint property prediction unit of the system according to claim 19 calculates the amount of change in the paint property after adjustment to perform prediction, The paint property prediction unit is configured to receive a plurality of paint property data of the paint before adjustment and the correction formula composition, and predict a plurality of changes in the paint properties after adjustment corresponding to each of the correction formula compositions. The system includes a computer, which also includes a modified recipe composition prediction unit, which determines the modified recipe composition obtained based on the predicted changes in the plurality of adjusted paint properties as the recipe composition of raw materials for adjusting the paint properties to achieve the target.
22. The system according to claim 21, characterized in that The computer further includes a pass / fail judgment unit configured to judge the pass / fail status of the paint based on the result of adding raw materials having the predicted corrected formula composition to the paint before adjustment and measuring the properties of the paint after adjustment.
23. The system according to any one of claims 18 to 22, characterized in that Also features: An artificial intelligence model acquisition unit acquires the specified artificial intelligence model.
