Granted Patent B2

US 12,468,693 · App. 18/538,947 · Granted Nov 11, 2025

Inventors: Tabassum J. Qureshi (Houston, TX); David M. Phillips (Houston, TX); Alberto Ancona (Houston, TX); Andreas Sadlier (Houston, TX)

Assignee: Halliburton Energy Services, Inc.

G06F16/2425G06F16/283G06F16/903G06F16/9535G06F16/958G06F40/205

Abstract

Claims (62)

1 . A method comprising:

receiving a user prompt that was parsed by a natural language processor;

obtaining, by a trained large language model (LLM), a plurality of drilling reports;

generating, by the trained LLM, a database created from a plurality of drilling reports from at least one wellbore, wherein each report includes sections and tables, wherein the trained LLM retrains itself after generating the database;

generating a first summary of a plurality of drilling reports for at least one wellbore based on the database created by the trained LLM from the plurality of drilling reports;

indicating a sentiment score of the first summary; and

generating a final summary based on the first summary and the user prompt.

2 . The method of claim 1 , wherein generating, by the trained LLM, a database includes:

segmenting, by the LLM, the sections and tables using first key-value pairs;

generating, by the LLM, metadata for the plurality of drilling reports;

collecting, by the LLM, keywords for the plurality of drilling reports;

generating, by the LLM, timestamps; and

creating, by the LLM, a hash table of the timestamps using second key-value pairs.

3 . The method of claim 2 , wherein the metadata includes Well ID identification information, Report Date, latitude and longitude of drilling location, Rig ID, and location of the wellbore.

4 . The method of claim 2 , wherein the keywords are tokenized keywords which include Status, Summary, Remarks, Measured Depth, Total Depth, a ratio of measured depth to total depth (Measured Depth/Total Depth), Bit Depth, Health Safety and Environmental (HSE) data.

5 . The method of claim 2 , wherein the timestamps include report start date time, and report end date time.

6 . The method of claim 1 further comprising:

determining the sentiment score of the first summary by analyzing the first summary based on oilfield lexicons, status identifiers included in the first summary, and ranges included in the first summary.

7 . The method of claim 1 , wherein the final summary includes one of a plurality of report grades and wherein a plurality of colors indicate whether the grade is positive, critical, or cautionary.

8 . The method of claim 7 , wherein positive is indicated by green, critical is indicated by red, and cautionary is indicated by yellow.

9 . The method of claim 1 , further comprising modifying a downhole drilling operation in the at least one wellbore based on the final summary.

10 . A system comprising:

a trained large language model (LLM) configured to generate a database created from a plurality of drilling reports from at least one wellbore, wherein each report includes sections and tables, and wherein the trained LLM retrains itself after generating the database; and

a summary generator configured to:

receive a user prompt that was parsed by a natural language processor;

generate a first summary of the plurality of drilling reports based on the database created by the trained LLM from the plurality of drilling reports;

indicate a sentiment score of the first summary; and

generate a final summary based on the first summary and the user prompt.

11 . The system of claim 10 , wherein the trained LLM to generate the database is configured to:

obtain the plurality of drilling reports;

segment the sections and tables using first key-value pairs;

generate metadata for the plurality of drilling reports;

collect keywords for the plurality of drilling reports;

generate timestamps; and

create a hash table of the timestamps using second key-value pairs.

12 . The system of claim 11 , wherein the metadata includes Well ID identification information, Report Date, latitude and longitude of drilling location, Rig ID, and location of the wellbore.

13 . The system of claim 11 , wherein the keywords are tokenized keywords which include Status, Summary, Remarks, Measured Depth, Total Depth, a ratio of measured depth to total depth (Measured Depth/Total Depth), Bit Depth, Health Safety and Environmental (HSE) data.

14 . The system of claim 11 , wherein the timestamps include report start date time, and report end date time.

15 . The system of claim 10 , wherein the summary generator is further configured to:

determine the sentiment score of the first summary by analyzing the first summary based on oilfield lexicons, status identifiers included in the first summary, and ranges included in the first summary.

16 . The system of claim 10 , wherein the final summary includes one of a plurality of report grades and wherein a plurality of colors indicate whether the grade is positive, critical, or cautionary.

17 . One or more machine-readable mediums including instructions executable by one or more processors, the instructions comprising:

instructions to receive a user prompt that was parsed by a natural language processor;

instructions to obtain, by a trained large language model (LLM), a plurality of drilling reports;

instructions to generate, by the trained LLM, a database created from a plurality of drilling reports from at least one wellbore, wherein each report includes sections and tables, wherein the trained LLM retrains itself after generating the database;

instructions to generate a first summary of a plurality of drilling reports based on the database created by the trained LLM from the plurality of drilling reports from at least one wellbore;

instructions to indicate a sentiment score of the first summary; and

instructions to generate a final summary based on the first summary and the user prompt.

18 . The one or more machine-readable mediums of claim 17 , wherein the instructions to generate, by the trained LLM, a database include:

instructions to obtain, by a large language model (LLM), the plurality of drilling reports;

instructions to segment, by the LLM, the sections and tables using first key-value pairs;

instructions to generate, by the LLM, metadata for the plurality of drilling reports;

instructions to collect, by the LLM, keywords for the plurality of drilling reports;

instructions to generate, by the LLM, timestamps; and

instructions to create, by the LLM, a hash table of the timestamps using second key-value pairs.

19 . The one or more machine-readable mediums of claim 18 ,

wherein the metadata includes Well ID identification information, Report Date, latitude and longitude of drilling location, Rig ID, and location of the wellbore;

wherein the keywords are tokenized keywords which include Status, Summary, Remarks, Measured Depth, Total Depth, a ratio of measured depth to total depth (Measured Depth/Total Depth), Bit Depth, Health Safety and Environmental (HSE) data; and

wherein the timestamps include report start date time, and report end date time.

20 . The one or more machine-readable mediums of claim 17 , further comprising:

instructions to determine the sentiment score of the first summary by analyzing the first summary based on oilfield lexicons, status identifiers included in the first summary, and ranges included in the first summary; and

wherein the final summary includes one of a plurality of report grades and wherein a plurality of colors indicate whether the grade is positive, critical, or cautionary.

Assignments (1)

Continuity (1)

References Cited (18)