System and method for analyzing drilling events

1 . A method for analysing undesired drilling events in a well, the method comprising:

receiving input data comprising daily drilling reports (DDRs) related to the well, wherein the DDRs comprise data related to one or more undesired drilling events in the well;

converting respective first portions of the DDRs related to the one or more undesired drilling events into vectors, wherein each vector comprises an embedded semantic meaning;

plotting each of the vectors on a two-dimensional (2D) graph, wherein plotting each of the vectors comprises determining an x-axis and a y-axis of the 2D graph with the vectors, wherein determining the x-axis and the y-axis of the 2D graph comprises:

extracting eigenvectors from the vectors, wherein each of the eigenvectors comprises an eigenvalue; and

identifying the eigenvectors comprising two largest eigenvalues, wherein the eigenvectors comprising the two largest eigenvalues represent the x-axis and the y-axis;

determining semantic similarities between the vectors with the 2D graph, wherein a respective distance between the vectors represents a degree of semantic similarities therebetween;

generating clusters of the vectors plotted on the 2D graph based upon the semantic similarities, wherein a first cluster of the clusters is related to loss circulation events, wherein a second cluster of the clusters is related to mitigation actions, and wherein a third cluster of the clusters is related to dynamic tests;

identifying a single vector of each of the clusters that represents a summary of the respective cluster, wherein identifying the single vector comprises:

identifying a centroid of each of the clusters, wherein the respective centroid of each of the clusters is a weighted average position of the vectors of the respective cluster; and

identifying the single vector of each of the clusters closest to the respective centroid thereof, wherein the single vector of each of the clusters represents the summary of the respective cluster;

displaying the clusters, the centroids, and the identified single vectors on the 2D graph to provide a visual quality control;

generating a summary of the one or more undesired drilling events with the identified single vectors; and

generating and transmitting a signal to perform a physical action in response to the summary of the one or more undesired drilling events, wherein the physical action comprises at least actuating a valve to vary at least a flow rate, a pressure, or a temperature of the well associated with the one or more undesired drilling events.

2 . The method of claim 1 , further comprising removing respective second portions of the DDRs that are unrelated to the one or more undesired drilling events, wherein the respective first portions of the DDRs are converted into vectors with a model, and wherein the model is a pre-trained embedding model.

3 . The method of claim 1 , wherein a sentence associated with the single vector of each of the clusters represents the summary of the respective cluster.

4 . The method of claim 1 , wherein generating the summary of the one or more undesired drilling events with the identified single vector of each of the clusters comprises concatenating the single vectors to build the summary.

5 . The method of claim 1 , wherein the action in response to the summary of the one or more undesired drilling events comprises one or more of a prevention action, a mitigation action, a contingency action, a remediation action, or a combination thereof.

6 . The method of claim 1 , wherein the summary of the one or more undesired drilling event only comprises information present in the DDRs, wherein the summary does not comprise information from a generative Artificial Intelligence (AI) model, and wherein the summary does not comprise hallucinations.

7 . A computing system, comprising:

one or more processors; and

a memory system comprising one or more non-transitory computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computing system to perform operations, the operations comprising:

receiving input data, wherein the input data is related to a drilling operation in a well and comprises daily drilling reports (DDRs) related to the well, wherein the DDRs comprise historical data related to the well, wherein the historical data comprises one or more undesired drilling events in the well;

removing respective first portions of the DDRs that are unrelated to the one or more undesired drilling events;

converting respective second portions of the DDRs that are related to the one or more undesired drilling events into vectors with a pre-trained embedding model, wherein each vector comprises an embedded semantic meaning;

plotting each of the vectors having the embedded semantic meaning on a two-dimensional (2D) graph, wherein plotting each of the vectors comprises determining an x-axis and a y-axis of the 2D graph with the vectors;

determining semantic similarities between the vectors with the 2D graph, wherein the semantic similarities are based upon the embedded semantic meaning, and wherein a respective distance between the vectors on the 2D graph represents a degree of semantic similarities therebetween;

generating clusters of the vectors plotted on the 2D graph based upon the semantic similarities, wherein a first cluster of the clusters is related to loss circulation events, wherein a second cluster of the clusters is related to mitigation actions, and wherein a third cluster of the clusters is related to dynamic tests;

identifying a single vector of each of the clusters that represents a summary of the respective cluster;

generating a summary of the one or more undesired drilling events with the identified single vectors of each of the clusters by concatenating the single vectors to build the summary of the one or more undesired drilling events;

performing an offset well analysis based upon the summary of the one or more undesired drilling events; and

generating and transmitting a signal to perform a physical action in response to the summary of the one or more undesired drilling events or a result of the offset well analysis, wherein the physical action comprises at least actuating a valve to vary at least a flow rate, a pressure, or a temperature of the well associated with the one or more undesired drilling events.

8 . The computing system of claim 7 , wherein determining the x-axis and the y-axis of the 2D graph comprises:

extracting eigenvectors from the vectors, wherein each of the eigenvectors comprises an eigenvalue; and

identifying the eigenvectors comprising two largest eigenvalues, wherein the eigenvectors comprising the two largest eigenvalues represent the x-axis and the y-axis.

9 . The computing system of claim 7 , wherein identifying the single vector comprises:

identifying a centroid of each of the clusters, wherein the respective centroid of each of the clusters is a weighted average position of the vectors of the respective cluster; and

identifying the single vector of each of the clusters closest to the respective centroid thereof, wherein the single vector of each of the clusters represents the summary of the respective cluster.

10 . A non-transitory computer-readable medium storing instructions that, when executed by one or more processors of a computing system, cause the computing system to perform operations, the operations comprising:

receiving input data, wherein the input data is related to a drilling operation in a well, wherein the input data comprises daily drilling reports (DDRs) related to the well, wherein the DDRs comprise historical data related to the well, wherein the DDRs comprise sentences related to the historical data, and wherein the historical data comprises one or more undesired drilling events in the well;

removing respective first portions of the DDRs comprising sentences that are unrelated to the one or more undesired drilling events, wherein the first portions of the DDRs comprise day-to-day operational context of the well;

converting respective second portions of the DDRs comprising sentences that are related to the one or more undesired drilling events into vectors with a pre-trained embedding model, wherein each of the sentences that are related to the one or more undesired drilling is converted to a respective vector, and wherein each of the vectors comprises an embedded semantic meaning;

plotting each of the vectors having the embedded semantic meaning on a two-dimensional (2D) graph, wherein plotting each of the vectors comprises:

displaying each of the vectors on the 2D graph, wherein each of the vectors is annotated with the sentence related to the respective vector; and

determining an x-axis and a y-axis of the 2D graph with the vectors, wherein determining the x-axis and the y-axis of the 2D graph comprises:

extracting eigenvectors from the vectors, wherein each of the eigenvectors comprises an eigenvalue; and

identifying the eigenvectors comprising two largest eigenvalues, wherein the eigenvectors comprising the two largest eigenvalues represent the x-axis and the y-axis;

determining semantic similarities between the vectors with the 2D graph, wherein the semantic similarities are based upon the embedded semantic meaning, wherein the semantic similarities are determined via a cosine similarity model, and wherein a respective distance between the vectors on the 2D graph represents a degree of semantic similarities therebetween;

generating clusters of the vectors on the 2D graph based upon the semantic similarities, wherein a first cluster of the clusters is related to loss circulation events, wherein a second cluster of the clusters is related to mitigation actions, and wherein a third cluster of the clusters is related to dynamic tests;

identifying a single vector of each of the clusters that represents a summary of the respective cluster, wherein identifying the single vector comprises:

identifying a centroid of each of the clusters, wherein the respective centroid of each of the clusters is a weighted average position of the vectors of the respective cluster; and

identifying the single vector of each of the clusters closest to the respective centroid thereof, wherein the sentence associated with the single vector of each of the clusters represents the summary of the respective cluster;

displaying the clusters, the centroids, and the identified single vectors on the 2D graph to provide a visual quality control;

generating a summary of the one or more undesired drilling events with the identified single vectors of each of the clusters by concatenating the single vectors to build the summary of the one or more undesired drilling events,

performing an offset well analysis based upon the summary of the one or more undesired drilling event; and

generating and transmitting a signal to perform a physical action in response to the summary of the one or more undesired drilling events or a result of the offset well analysis, wherein the physical action comprises at least actuating a valve to vary at least a flow rate, a pressure, or a temperature of the well associated with the one or more undesired drilling events.

11 . The non-transitory computer-readable medium of claim 10 , wherein the sentences of the DDRs relate to the one or more undesired drilling events in the well, one or more actions applied to the well, day-to-day operational information of the well, or a combination thereof, wherein the one or more undesired drilling events comprise a stuck pipe event, a loss circulation event, or a combination thereof, and wherein each of the one or more undesired drilling events comprises details about a failure mode, remedial attempts, or any combination thereof.

12 . The non-transitory computer-readable medium of claim 10 , wherein the day-to-day operational context of the well comprises a rate of penetration (ROP) or a bottom hole assembly (BHA).

13 . The non-transitory computer-readable medium of claim 10 , wherein the pre-trained embedding model is a Bidirectional Encoder Representation from Transformers (BERT) model, and wherein the BERT model is a Sentence-BERT (sBERT) model.

14 . The non-transitory computer-readable medium of claim 10 , wherein the summary of the one or more undesired drilling event only comprises information present in the DDRs, wherein the summary does not comprise information from a generative Artificial Intelligence (AI) model, and wherein the summary does not comprise hallucinations.