Granted Patent B2

US 12,467,874 · App. 17/476,983 · Granted Nov 11, 2025

Inventors: Philipp Isken (Altenberge, DE); Sandra Bittorf (Witten, DE); Oliver Kroehl (Cologne, DE); Claudia Bramlage (Essen, DE); Markus Vogel (Kamp-Lintfort, DE); Stefan Silber (Krefeld, DE); Gaetano Blanda (Haltern am See, DE); Olivia Lewis (Berlin, DE); Daniel Haake (Potsdam, DE)

Assignee: EVONIK OPERATIONS GMBH

G01N21/8851B25J9/1697G01N21/8422G01N21/8806G05B19/41875G06F16/51G06N20/00G06T7/0004G06T7/001G06T7/11G16C60/00G01N2021/8427G01N2021/8854G01N2021/8861G01N2021/8864G01N2021/8874G01N2021/888G01N2021/8887G05B2219/37206G05B2219/37451G05B2219/45013G06T2207/20081G06T2207/20084G06T2207/30108G06T2207/30156G16C20/30G16C20/70Y02P90/02

Abstract

Claims (58)

1 . A computer-implemented method for providing a coating composition-related prediction program, the method comprising:

providing a database comprising associations of one or more of qualitative and quantitative characterizations of coating surfaces and one or more parameters selected from a group comprising one or more of: one or more of components of a coating composition used for producing a respective coating surface; one or more of relative and absolute amounts of one or more of the components; manufacturing-process parameters of the coating composition; and application-process parameters used for creating the coating surfaces;

training a machine learning model on the associations of the coating surface characterizations with the one or more parameters in the database for providing a predictive model having learned to correlate one or more of the qualitative and the quantitative characterizations of one or more coating surfaces with one or more of the parameters stored in association with respective coating surface characterizations; and one or more of:

providing a composition-quality-prediction program which comprises a first predictive model, the composition-quality-prediction program being configured for using the first predictive model for predicting the properties of a coating surface to be produced from one or more input parameters selected from a group comprising one or more of: one or more components of the coating composition to be used for producing the coating surface; one or more of relative and absolute amounts of one or more of the components; manufacturing-process parameters to be used for preparing the coating composition; application-process parameters to be used for creating the coating surface; and

providing a composition-specification-prediction program which comprises a second predictive model, the composition-specification-prediction program being configured for using the second predictive model for predicting, based on an input specifying at least a desired coating surface characterization, one or more output parameters related to the coating composition predicted to generate the coating surface having input surface characterizations, the one or more output parameters being selected from a group comprising one or more of: one or more components of the coating composition; one or more of relative and absolute amounts of one or more of the components; manufacturing-process parameters to be used for preparing the coating composition; and application-process parameters for creating the coating surface,

wherein the training the machine learning model comprises using an active learning module for selecting a candidate coating composition specification from a set of candidate composition specifications which provides the strongest learning effect for the first predictive model and the second predictive model, the learning effect being provided as a result of the preparation of the selected candidate composition, empirical measurement of properties of the selected candidate composition, and retraining of one or more of the first predictive model and the second predictive model on training data comprising the empirically measured properties of the selected and prepared candidate composition,

wherein the selection of the candidate coating composition specification by the active learning module comprises

for each of the multiple candidate coating composition specifications:

predicting properties of a coating surface of the specified candidate coating composition using the first predictive model of the composition-quality-prediction program; and

determining an uncertainty score being indicative of the uncertainty of the first model with respect to the predicting; and

using the one of the multiple candidate coating compositions having the maximum uncertainty score as the selected candidate coating composition specification.

2 . The method of claim 1 , the method comprising:

providing a images depicting coating surfaces made from multiple different coating compositions, wherein at least some of the coating surfaces respectively have one or more coating defects of one or more different defect types;

applying a defect-identification program on the images for recognizing patterns in the images, for obtaining the measures of the coating defects represented by the identified patterns in the images and for computing one or more of the qualitative and a the quantitative characterization of the coating surfaces depicted in the images;

storing one or more of the qualitative and the quantitative characterizations of the coating surfaces in association with one or more parameters related to the coating composition used for creating the coating surface comprising these defects in the database.

3 . The method of claim 1 , wherein the coating composition comprises a liquid basic component and one or more solids dispersed in the liquid basic component, wherein the liquid basic component comprises one or more of water, a water-based solution or an organic solvent, wherein the one or more solids are selected from a group comprising pigments, dyes, and fillers.

4 . The method of claim 1 , wherein the properties of the coating surface comprise an indication of one or more coating defects selected from a group comprising one or more of: a cratering defect, an abrasion defect, an adhesion failure defect, an alligatoring defect, a bleeding defect, a blistering defect, a bloom defect, a bridging defect, a bubbling defect, a cathodic disbanding defect, a checking defect, a cissing defect, a cobwebbing defect, a cracking defect, a crazing defect, a crowsfooting defect, a delamination defect, a fading defect, a flaking defect, a grinning defect, a heat defect, an impact defect, an intercoat contamination defect, a mud cracking defect, an orange peeling defect, a peeling defect, a pinholes defect, a rippled coating defect, a runs defect, a rust rashing defect, a rust spotting defect, a rust staining defect, a sags defect, a settlement defect, a skinning defect, a solvent lifting defect, a solvent popping defect, a stress cracking defect, an undercutting defect, and a wrinkling defect.

5 . The method of claim 1 , further comprising using of the composition-specification-prediction program for predicting a coating composition specification meeting a desired coating surface characterization provided as input, the using comprising:

providing at least a specification of a desired coating surface characterization as input to the composition-specification-prediction program;

predicting, by the composition-specification-prediction program, a specification of a coating composition adapted to provide a coating surface having the desired surface characteristics, wherein the specification comprises one or more parameters selected from a group comprising one or more of: one or more coating composition components; absolute or relative amounts of one or more of the coating composition components; manufacturing process parameters; and application process parameters;

preferably further comprising one or more of: outputting the predicted specification of the coating composition to a human; and inputting the specification of the selected candidate coating composition to a processor which controls a facility for one or more of producing and testing compositions for coating compositions, wherein the processor drives the facility to produce the input coating composition.

6 . The method of claim 1 , further comprising:

a. performing the training of the machine learning model on the associations of the measures and the coating components in the database for providing the first predictive model and the second predictive model, where a loss function is minimized for the training,

b. testing to determine whether a value of the loss function obtained for the predictive model meets a specified criterion,

whereby selectively, in the event that the specified criterion is not met, the following steps are carried out:

i. selection of a candidate coating composition specification from a plurality of candidate coating composition specifications by the active learning module, the selected candidate composition specification specifying the one of the candidate compositions determined to provide the highest learning effect of the first predictive model and the second predictive model regarding the correlation of one or more of the qualitative and the quantitative coating surface characterizations and one or more parameters selected from a group comprising one or more of: coating components; component amounts; manufacturing process parameters; and application process parameters of the coating composition;

ii. driving a facility for automatically producing the candidate coating composition in accordance with the selected specification, for automatically applying the produced candidate composition on a substrate and for automatically acquiring an image of the surface with the applied coating composition;

iii. applying the defects-identification program on the image acquired in step ii for computing and storing one or more of the qualitative and the quantitative characterization of the coating surface depicted in the image in the database, thereby extending the database;

iv. re-training of the first predictive model and the second predictive model on the extended database for providing an improved version of the predictive model,

v. repeated execution of step b using the improved version of the predictive model,

c. one or more of: replacing the first predictive model of the composition-quality-prediction program with the improved version of the first predictive model; and replacing the second predictive model of the composition-specification-prediction program with the improved version of the second predictive model.

7 . The method of claim 1 , wherein the candidate coating composition specification in addition comprises at least a desired coating surface characterization, wherein the selection of the candidate coating composition specification by the active learning module comprises

for each of the multiple candidate coating composition specifications:

Predicting one or more of the output parameters predicted to be related to the candidate coating composition having the desired coating surface characterization using the second predictive model of the composition-specification-prediction program;

Determining an uncertainty score of the prediction; and

using the one of the multiple candidate coating compositions as the selected candidate coating composition specification having the maximum uncertainty score.

8 . The method of claim 1 , wherein the active learning module is operatively coupled to a facility for preparing and testing compositions for paints, varnishes, printing inks, grinding resins, pigment concentrates or other coating materials.

9 . A non-transitory machine-readable medium storing a data structure for implementing the method of claim 1 further comprising: encoding at least one coating composition specification, the data structure being an output of the active learning module, and being interpretable by a facility for producing and testing coating compositions, the data structure being configured to cause, upon being input to the facility, the facility to prepare and test the coating composition specified in the coating composition specification.

10 . The method of claim 1 , wherein the composition-specification-prediction program is configured for receiving an incomplete coating composition specification and for using the incomplete coating composition specification for limiting a solution space of a predicted output parameters.

11 . The method of claim 8 , wherein the active learning module is configured to automatically drive the facility iteratively to generate and empirically analyze one or more candidate coating compositions until the accuracy of the first predictive model and the second predictive model reaches a predefined minimum accuracy threshold or until a maximum number of iterations is reached.

12 . The method of claim 8 , wherein the active learning module is configured to generate a graphical user interface (GUI), the GUI comprising a plurality of selectable GUI elements, each GUI element representing one of the candidate coating composition specifications, the GUI comprising a model quality indicator being indicative of the accuracy of the first predictive model and the second predictive model, the active learning module being configured for

driving the facility, upon a selection of one of the GUI-elements by a user, to prepare and test the candidate coating composition represented by the selected GUI element,

use the empirical properties of the prepared candidate coating composition for retraining the predictive model; and

update the model quality indicator of the GUI to indicate the accuracy of the retrained predictive model.

13 . A computer system comprising at least one processor and computer-interpretable instructions which, when executed by the at least one processor, cause the at least one processor to:

access a database comprising associations of one or more of qualitative and quantitative characterizations of coating surfaces and one or more parameters selected from a group comprising one or more of: one or more of the components of the coating composition used for producing the respective coating surface; one or more of relative and absolute amounts of one or more of the components; manufacturing-process parameters of the coating composition; and application-process parameters used for creating the coating surfaces;

train a machine learning model on the associations of the coating surface characterizations with the one or more parameters in the database for providing a predictive model having learned to correlate the one or more of the qualitative and the quantitative characterizations of one or more coating surfaces with one or more of the parameters stored in association with the respective coating surface characterizations; and one or more of:

provide a composition-quality-prediction program which comprises a first predictive model, the composition-quality-prediction program being configured for using the first predictive model for predicting the properties of a coating surface to be produced from one or more input parameters selected from a group comprising one or more of: one or more components of a coating composition to be used for producing a coating surface; one or more of relative and absolute amounts of one or more of the components; manufacturing-process parameters to be used for preparing the coating composition; and application-process parameters to be used for creating the coating surface; and

provide a composition-specification-prediction program which comprises a second predictive model, the composition-specification-prediction program being configured for using the second predictive model for predicting, based on an input specifying at least a desired coating surface characterization and outputting one or more parameters related to a coating composition predicted to generate a coating surface having the input surface characterizations, the one or more output parameters being selected from a group comprising one or more of: one or more components of the coating composition; one or more of relative and absolute amounts of one or more of the components; manufacturing-process parameters to be used for preparing the coating composition; and application-process parameters for creating the coating surface,

wherein the computer-interpretable instructions which, when executed by the at least one processor, further causes the at least one processor to training the machine learning model by using an active learning module for selecting a candidate coating composition specification from a set of candidate composition specifications which provides the strongest learning effect for the first predictive model and the second predictive model, the learning effect being provided as a result of the preparation of the selected candidate composition, empirical measurement of properties of the selected candidate composition, and retraining of one or more of the first predictive model and the second predictive model on training data comprising the empirically measured properties of the selected and prepared candidate composition,

wherein the selection of the candidate coating composition specification by the active learning module comprises

for each of the multiple candidate coating composition specifications:

predicting properties of the specified candidate coating composition using the first predictive model of the composition-quality-prediction program; and

determining an uncertainty score being indicative of the uncertainty of the first model with respect to the predicting; and

using the one of the multiple candidate coating compositions having the maximum uncertainty score as the selected candidate coating composition specification.

14 . The computer system of claim 13 further comprising;

facility for producing and testing compositions for paints, varnishes, printing inks, grinding resins, pigment concentrates or other coating materials, where the facility comprises at least two workstations, where the at least two workstations are connected to one another via a transport system on which self-propelled transport vehicles are able to run for transporting one or more of the components of the composition and of the composition produced between the workstations.

15 . The system of claim 13 , wherein the composition-specification-prediction program is configured for receiving an incomplete coating composition specification and for using the incomplete coating composition specification for limiting a solution space of a predicted output parameters.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (52)