Cementing lab data validation based on machine learning

Patent metadata

Field	Value
Publication number	US-12595715-B2
Application number	US-202117558810-A
Country	US
Kind code	B2
Filing date	Dec 22, 2021
Priority date	Dec 22, 2021
Publication date	Apr 7, 2026
Grant date	Apr 7, 2026

Abstract

Techniques of the present disclosure relate to validating data for a composition design. A method comprises applying a machine learning model to at least two inputs comprising parameters of a cement composition and experimental conditions such that the machine learning model outputs at least one predicted property of the cement composition; performing a laboratory experiment to determine at least one experimental property of the cement composition; calculating an error between the at least one predicted property and the at least one experimental property; and recording the experimental data in a cement property database if the error is within an error range or repeating the performing the laboratory experiment if the error is outside the error range.

First claim

What is claimed is:

1 . A method comprising:
applying a machine learning model trained on previously recorded experimental data comprising parameters of a plurality of historical cement composition and historical experimental condition, to at least two inputs comprising parameters of a cement composition comprising a cement and water and experimental conditions such that the machine learning model outputs at least one predicted property of the cement composition, wherein properties of a cement composition comprise at least one of compressive strength, thickening time, or gel strength and wherein experimental conditions comprise at least of pressure, temperature, or time; performing a laboratory experiment to determine at least one experimental property of the cement composition; calculating an error between the at least one predicted property and the at least one experimental property; recording the experimental data in a cement property database if the error is within an error range or repeating the performing the laboratory experiment if the error is outside the error range; and preparing an additional cement composition based on the cement composition, such that at least one of the parameters of the additional cement composition has a second error within the error range.

2 . The method of claim 1 , further comprising training the machine learning model to predict properties of cement compositions.

3 . The method of claim 1 , further comprising sending a notification to a user about deviation in material performance if the error is outside the error range.

4 . The method of claim 1 , further comprising determining a range of values for the two or more inputs to screen the inputs prior to an application of the machine learning model to the inputs.

5 . The method of claim 1 , wherein the parameters of the cement composition each individually comprise at least one parameter selected from the group consisting of an experimental condition, an amount of a component of the cement composition, an identify of a component of the cement composition, and combinations thereof.

6 . The method of claim 1 , wherein the at least one predicted property comprises at least one cement property selected from the group consisting of compressive strength, thickening time, fluid loss, and combinations thereof.

7 . The method of claim 1 , further comprising training the machine learning model with previous recorded cement data.

8 . The method of claim 1 , further comprising screening the at least two inputs with a predetermined range.

9 . The method of claim 1 , further comprising pumping the additional cement composition in a wellbore.

10 . The method of claim 1 wherein the laboratory experiment comprises a QA/QC test, wherein the QA/QC test determines the performance of cement composition materials and/or wherein the laboratory experiment comprises a formulation test, wherein the formulation test determines cement composition properties.

11 . The method of claim 1 further comprising allowing the additional cement composition to set to form a set composition.

12 . A method comprising:
calculating an acceptable range of values for properties of cement compositions based on previously recorded experimental data; receiving new experimental data for a cement composition, wherein the experimental data comprises at least one experimental property of the cement composition comparing the at least one experimental property to determine if the experimental data is within the acceptable range of values; applying a machine learning model to at least two inputs comprising parameters of a cement composition comprising a cement and water and experimental conditions such that the machine learning model outputs at least one predicted property of the cement composition, wherein properties of a cement composition comprise at least one of compressive strength, thickening time, or gel strength and wherein experimental conditions comprise at least of pressure, temperature, or time; calculating an error between at least one predicted property and the at least one experimental property; recording the experimental data in a cement property database, wherein-if the error is within an error range or repeating receiving new experimental data from a repeated laboratory experiment if the error is outside the error range; and producing an additional cement composition based on the cement composition, such that at least one of the parameters of the additional cement composition has a second error within the error range.

13 . The method of claim 12 , further comprising training the machine learning model with previously recorded cement data, prior to application of the machine learning model to the inputs.

14 . The method of claim 12 , further comprising training the machine learning model with previously recorded data.

15 . The method of claim 12 , further comprising sending a notification to a user about deviation in material performance if the error is outside the error range.

16 . The method of claim 12 , further comprising disposing the additional cement composition into a wellbore.

17 . A method comprising:
training a machine learning model to predict properties of cement compositions based on previously recorded experimental data; calculating an acceptable range of values for one or more properties in the experimental data; performing a laboratory experiment to determine at least one experimental property of a cement composition; applying the machine learning model to at least two inputs comprising parameters of a cement composition comprising a cement and water and experimental conditions such that the machine learning model outputs at least one predicted property of the cement composition, wherein properties of a cement composition comprise at least one of compressive strength, thickening time, or gel strength and wherein experimental conditions comprise at least of pressure, temperature, or time; calculating an error between the at least one predicted property and the at least one experimental property, wherein the at least one experimental property is within the acceptable range of values; recording the experimental data in a cement property database if the error is within a range or repeating the performing the laboratory experiment if the error is outside the error range; and preparing an additional cement composition based on the cement composition, such that at least one of the parameters of the additional cement composition has a second error within the error range.

18 . The method of claim 17 , wherein the parameters of the cement composition each individually comprise at least one parameter selected from the group consisting of an experimental condition, an amount of a component of the cement composition, an identify of a component of the cement composition, and combinations thereof.

19 . The method of claim 17 , wherein the at least one predicted property comprises at least one cement property selected from the group consisting of compressive strength, thickening time, fluid loss, and combinations thereof.

20 . The method of claim 17 , further comprising sending a notification to a user about deviation in material performance if the error is outside the error range.

Assignees

• Halliburton Energy Services Inc

Inventors

• Amini Shohreh
• Singh John Paul Bir
• Jandhyala Siva Rama Krishna

Classifications

• G16C20/30

  Prediction of properties of chemical compounds, compositions or mixtures · CPC title

• E21B47/005

  Monitoring or checking of cementation quality or level · CPC title

• C04B40/0032

  Controlling the process of mixing, e.g. adding ingredients in a quantity depending on a measured or desired value (B28C7/00 takes precedence) · CPC title

• C09K8/46

  containing inorganic binders, e.g. Portland cement · CPC title

• G06N20/00

  Machine learning · CPC title