# US 12,761,394 B2 · Determining downhole operation transitions from wellbore measurement data using machine learning > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,761,394 B2 - Document ID: US12761394B2 - Status: Granted - Kind: B2 - Application number: 18/442,429 - Filed: 2024-02-15 - Published: 2025-08-21 - Granted: 2026-10-13 - Earliest claimed: 2024-02-15 - Assignees: Schlumberger Technology Corporation - Applicants: Schlumberger Technology Corporation - Inventors: Mallikarjuna Rao Gudipati; Kjetil Westre - Web page: https://patentgrep.com/patent/US12761394B2 ## CPC - G01V 99/00: Subject matter not provided for in other groups of this subclass - E21B 41/00: Equipment or details not covered by groups E21B15/00 - E21B40/00 - E21B 47/00: Survey of boreholes or wells (monitoring pressure or flow of drilling fluid E21B21/08) - G06N 20/00: Machine learning - E21B 2200/22: Fuzzy logic, artificial intelligence, neural networks or the like - G01V: GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS (means for indicating the location of accidentally buried, e.g. snow-buried, persons A63B29/02) - E21B: EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract This application relates to a downhole system that uses a transition detection system to determine transitions between downhole activities or operations for a wellbore based on wellbore measurement data. In various implementations, the transition detection system uses a transition identification machine learning model to generate downhole transition types between downhole operations from wellbore measurement data. Additionally, the transition detection system identifies errors and inaccuracies with activity transitions reported in a downhole operation report based on comparing the downhole operation report to the determined transition times generated by the transition identification machine learning model. ## Claims ### Claim 1 A computer-implemented method for determining transitions in downhole operations, comprising: performing a downhole sequence for a wellbore, the downhole sequence including a first downhole operation and a second downhole operation with a first transition between the first downhole operation and the second downhole operation; measuring, by one or more downhole sensors or one or more surface sensors during the downhole sequence, time-series wellbore data for the downhole sequence of the wellbore; identifying, in a downhole operation report for the wellbore, the first downhole operation, the second downhole operation, and a reported time signature for the first transition, wherein the reported time signature is inaccurate relative to an actual time signature when the first transition occurred between performance of the first downhole operation and performance of the second downhole operation in the downhole sequence for the wellbore; determining a plurality of time points within the time-series wellbore data and a plurality of time windows around the plurality of time points, wherein each time window of the plurality of time windows corresponds to a time point of the plurality of time points and includes, for the time point of the plurality of time points, a first portion before the time point of the plurality of time points and a second portion after the time point of the plurality of time points; for each time point of the plurality of time points, generating, from the time-series wellbore data, a statistical attribute set for a corresponding time window of the plurality of time windows, the statistical attribute set including a first statistical attribute subset for the first portion of the corresponding time window and a second statistical attribute subset for the second portion of the corresponding time window; for each time point of the plurality of time points, predicting a transition type at the time point using a transition identification machine learning model, the statistical attribute set for the corresponding time window being input to the transition identification machine learning model, the transition identification machine learning model having been trained to classify transitions between downhole operations based on statistical attributes of wellbore data at time windows around transition times; identifying, from the transition types for the plurality of time points, a predicted transition type corresponding to the first transition between the first downhole operation and the second downhole operation at a predicted time point with a predicted time signature; and in response to determining that the reported time signature for the first transition between the first downhole operation and the second downhole operation differs from the predicted time signature, updating the reported time signature in the downhole operation report to the predicted time signature to reconcile the reported time signature for the first transition with the actual time signature at which the first transition occurred based on the time-series wellbore data for the downhole sequence. ### Claim 2 (depends on 1) The computer-implemented method of claim 1, wherein generating the statistical attribute set for each time point of the plurality of time points is based on measurement data associated with the downhole operations for the wellbore. ### Claim 3 (depends on 2) The computer-implemented method of claim 2, wherein generating the statistical attribute set for each time point of the plurality of time points is based on different measurement data types measured for the wellbore. ### Claim 4 (depends on 1) The computer-implemented method of claim 1, wherein: the transition identification machine learning model determines the transition types by classifying the corresponding time windows for the plurality of time points based on a group of candidate transition types; and the predicted transition type corresponding to the first transition is selected based on having a highest probability among the group of candidate transition types. ### Claim 5 (depends on 4) The computer-implemented method of claim 4, wherein: the group of candidate transition types includes a null transition type indicating no transition of the downhole operations; and the transition identification machine learning model determines the null transition type when no downhole operation transitions occur within a one of the corresponding time windows for one of the plurality of time points. ### Claim 6 (depends on 5) The computer-implemented method of claim 5, wherein the transition identification machine learning model determines the null transition type for one of the corresponding time windows for one time point of the plurality of time points based on determining that a transition probability for the one of the corresponding time windows is below a transition threshold value. ### Claim 7 (depends on 1) The computer-implemented method of claim 1, further comprising generating training data for the transition identification machine learning model based on: receiving downhole operation reports of downhole operations for a set of reference wellbores; receiving one or more sets of wellbore measurement data for the set of reference wellbores; correlating the downhole operation reports with the one or more sets of wellbore measurement data based on time windows; and generating the training data by combining correlated transition types with the one or more sets of wellbore measurement data based on time signatures of downhole operation transition times from the downhole operation reports. ### Claim 8 (depends on 7) The computer-implemented method of claim 7, further comprising: for a set of time windows around time points in the one or more sets of wellbore measurement data for the set of reference wellbores, generating a statistical attribute set for each time window in the set of time windows; providing the statistical attribute sets for the set of time windows to the transition identification machine learning model to determine transition types for the time points; and comparing the determined transition types for the time points to the correlated transition types to train the transition identification machine learning model. ### Claim 9 (depends on 1) The computer-implemented method of claim 1, wherein generating the statistical attribute set for each time point of the plurality of time points includes determining one or more of a mean, median, maximum, minimum, or standard deviation based on the time window for the time point of the plurality of time points. ### Claim 10 (depends on 1) The computer-implemented method of claim 1, wherein the transition identification machine learning model uses a decision leaf-based architecture to determine the predicted transition type. ### Claim 11 (depends on 1) The method of claim 1, further comprising training the transition identification machine learning model using statistical attribute sets generated from time windows around time points in reference time-series wellbore data as training samples to predict transition types for the time points, and comparing the predicted transition types to corresponding ground-truth transition types for the time points. ### Claim 12 A system, comprising: a processor; memory in electronic communication with the processor; and instructions stored in the memory, the instructions being executable by the processor to: perform a downhole sequence for a wellbore, the downhole sequence including a first downhole operation and a second downhole operation with a first transition between the first downhole operation and the second downhole operation; measure, by one or more downhole sensors or one or more surface sensors during the downhole sequence, time-series wellbore data for the downhole sequence of the wellbore; identify, in a downhole operation report for the wellbore, the first downhole operation, the second downhole operation, and a reported time signature for the first transition, wherein the reported time signature is inaccurate relative to an actual time signature when the first transition occurred between performance of the first downhole operation and performance of the second downhole operation in the downhole sequence for the wellbore; determine a plurality of time points within the time-series wellbore data and a plurality of corresponding time windows around the plurality of time points, wherein each time window of the plurality of corresponding time windows corresponds to a time point of the plurality of time points and includes, for the time point of the plurality of time points, a first portion before the time point of the plurality of time points and a second portion after the time point of the plurality of time points; for each time point of the plurality of time points, generate, from the time-series wellbore data, a statistical attribute set for a corresponding time window of the plurality of corresponding time windows, the statistical attribute set including a first statistical attribute subset for the first portion of the corresponding time window and a second statistical attribute set for the second portion of the corresponding time window; for each time point of the plurality of time points, predict a transition type at the time point using a transition identification machine learning model, the statistical attribute set for the corresponding time window being input to the transition identification machine learning model, the transition identification machine learning model having been trained to classify transitions between operations based on statistical attributes of wellbore data at time windows around transition times; identify, from the transition types for the plurality of time points, a predicted transition type corresponding to the first transition between the first downhole operation and the second downhole operation at a predicted time point with a predicted time signature; and in response to determining that the reported time signature for the first transition between the first downhole operation and the second downhole operation differs from the predicted time signature, update the reported time signature in the downhole operation report to the predicted time signature to reconcile the reported time signature for the first transition with the actual time signature at which the first transition occurred based on the time-series wellbore data for the downhole sequence. ### Claim 13 A computer-implemented method for determining transitions in downhole operations, comprising: receiving reference time-series wellbore data measured for a set of reference wellbores; receiving reference downhole operation reports for the set of reference wellbores; identifying, from the reference downhole operation reports, reference time points corresponding to reference transitions between downhole operations in the set of reference wellbores; determining reference time windows around the reference time points in the reference time-series wellbore data; generating, from the reference time-series wellbore data, reference statistical attribute sets for the reference time windows; labeling the reference statistical attribute sets corresponding to the reference time points with corresponding ground-truth transition types from the reference downhole operation reports to create training data samples; training a transition identification machine learning model using the training data samples to predict transition types for the reference time points, including comparing predicted transition types for the reference time points to the corresponding ground-truth transition types for the reference time points; receiving time-series wellbore data measured for a downhole sequence in a wellbore by one or more downhole sensors or one or more surface sensors, the downhole sequence including a first downhole operation, a second downhole operation, and a first transition between the first downhole operation and the second downhole operation; identifying, in a downhole operation report for the wellbore, a reported time signature for the first transition between the first downhole operation and the second downhole operation, wherein the reported time signature is inaccurate relative to an actual time signature at which the first transition occurred between the first downhole operation and the second downhole operation in the downhole sequence for the wellbore; determining a plurality of time points within the time-series wellbore data and a plurality of corresponding time windows around the plurality of time points, wherein each time window of the plurality of time windows corresponds to a time point of the plurality of time points and includes, for a time point of the plurality of time points, a first portion before the time point of the plurality of time points and a second portion after the time point of the plurality of time points; for each time point of the plurality of time points, generating, from the time-series wellbore data, a statistical attribute set for a corresponding time window of the plurality of time windows, the statistical attribute set including a first statistical attribute subset for the first portion of the corresponding time window and a second statistical attribute subset for the second portion of the corresponding time window; for each time point of the plurality of time points, predicting a transition type at the time point using the transition identification machine learning model, the statistical attribute set for the corresponding time window being input to the transition identification machine learning model; identifying, from the transition types for the plurality of time points, a predicted transition type corresponding to the first transition between the first downhole operation and the second downhole operation at a predicted time point with a predicted time signature; and in response to determining that the reported time signature for the first transition differs from the predicted time signature, updating the reported time signature in the downhole operation report to the predicted time signature to reconcile the reported time signature with the actual time signature at which the first transition occurred, based on the time-series wellbore data for the wellbore.