US-12619682-B2 - Identifying operation anomalies of subterranean drilling equipment

US12619682B2US 12619682 B2US12619682 B2US 12619682B2US-12619682-B2

Abstract

The present disclosure relates to systems, methods, and non-transitory computer-readable media for dynamically utilizing, in potentially real time, anomaly pattern detection to optimize operational processes relating to well construction or subterranean drilling. For example, the disclosed systems use time-series data combined with rig states to automatically detect and split similar operations. Subsequently, the disclosed systems identify operation anomalies from a field-data collection utilizing an automated anomaly detection workflow. The automated anomaly detection workflow can identify operation anomalies at a more granular level by determining which process behavior contributes to the operation anomaly (e.g., according to corresponding process probabilities for a given operation). In addition, the disclosed systems can present graphical representations of operation anomalies, process behaviors (procedural curves), and/or corresponding process probabilities in an intuitive, user-friendly manner.

Inventors

Diego Fernando PATINO VIRANO
Darine Mansour
Sai Venkatakrishnan Sankaranarayanan
Yingwei Yu

Assignees

SCHLUMBERGER TECHNOLOGY CORPORATION

Dates

Publication Date: 20260505
Application Date: 20211022

Claims (12)

1 . A non-transitory computer readable storage medium comprising instructions that, when executed by at least one processor, cause a computing device to: while operating subterranean drilling equipment having a plurality of operating states, each of the plurality of operating states defined by a plurality of operation features, wherein the subterranean drilling equipment includes a pump, each operation feature of the plurality of operation features defining operation of the subterranean drilling equipment over time: collect, from the pump, first sensor data related to the operation of the subterranean drilling equipment, wherein the first sensor data includes a pump flowrate; identify an operating state of the plurality of operating states of the subterranean drilling equipment, wherein the operating state includes a pre-connection activity, the pre-connection activity including at least one of drilling off, moving a drill string to a connection point, or stopping rotation and the pump; identify first time-series data for the subterranean drilling equipment from the first sensor data; identify contextual data that provides context to the first time-series data, wherein the contextual data includes drilling fluid data of a drilling fluid; using the first time-series data and the contextual data, partition the first time-series data by splitting the first time-series data into a plurality of data buckets based on the operating state of the subterranean drilling equipment, each data bucket of the plurality of data buckets including partitioned first time-series data; for one of the plurality of data buckets, filter the partitioned first time-series data to estimate feature signals based on smoothing one or more data spikes from signal noise of the partitioned first time-series data in a manner that reduces temporal lag when provided as an input to a feature extraction model, the feature signals describing the plurality of operation features; extract, utilizing the feature extraction model and from the feature signals, the plurality of operation features, wherein the feature extraction model includes a feature machine-learning model trained to determine the plurality of operation features; generate a plurality of first feature probabilities, each of the plurality of first feature probabilities associated with an operation feature of the plurality of operation features, wherein the plurality of first feature probabilities are estimated likelihoods that the operation feature of the plurality of operation features associated with a feature probability of the plurality of first feature probabilities corresponds to a certain value; provide, for display within a graphical user interface, a feature curve representation of a plurality of feature curves for the plurality of operation features, the feature curve representation including a plurality of feature curve clusters, wherein the plurality of feature curves are displayed on the feature curve representation based on an anomaly threshold, and wherein each of the plurality of feature curve clusters are selectable; receive, at the graphical user interface, a selection of a feature curve cluster of the plurality of feature curve clusters; based on the selection of the feature curve cluster, populate an anomaly visualization on the graphical user interface, the anomaly visualization including the plurality of feature curves based on the selection of the feature curve cluster; receive a user input at a feature probability slider on the graphical user interface, wherein the user input increases a probability threshold for the plurality of first feature probabilities; using the user input, update the anomaly visualization by removing a feature curve of the plurality of feature curves based on removed feature probability of the feature curve being less than the probability threshold set by the user input; identify, on the anomaly visualization and using the probability threshold set by the user input, a first anomaly of the operation of the subterranean drilling equipment based on the plurality of first feature probabilities for the plurality of operation features, wherein the first anomaly is an abnormality of the partitioned first time-series data, wherein identifying the first anomaly includes identifying a probability of the first anomaly being anomalous using an anomaly machine-learning model for probability density estimation; based on the first anomaly and the operation feature, identify a lost time for a drilling operation associated with the operating state and the contextual data; based on the first anomaly and the lost time, determine a corrective action to be performed by the subterranean drilling equipment, the corrective action including adjusting the drilling fluid; based on the corrective action, adjust the drilling fluid by switching at least a portion of the drilling fluid for use by the subterranean drilling equipment; after switching the drilling fluid, collect second sensor data and identify second time-series data from the second sensor data; partition the second time-series data, filter the partitioned second time-series data, and extract the plurality of operation features from the filtered partitioned second time-series data; and generate a plurality of second feature probabilities and identify a second anomaly of the operation of the subterranean drilling equipment.
2 . The non-transitory computer readable storage medium of claim 1 , wherein the filtering the partitioned first time-series data includes filtering with a zero-lag Difference of Gaussian filter.
3 . The non-transitory computer readable storage medium of claim 2 , wherein the zero-lag Difference of Gaussian filter includes a half-filter defined in a temporal domain.
4 . The non-transitory computer readable storage medium of claim 1 , further comprising instructions that, when executed by the at least one processor, cause the computing device to generate the plurality of first feature probabilities for the plurality of operation features by determining probability density functions for discrete feature datasets partitioned from the first time-series data.
5 . The non-transitory computer readable storage medium of claim 1 , further comprising instructions that, when executed by the at least one processor, cause the computing device to provide, for display within the graphical user interface, a plain-text description of one or more of the plurality of operation features contributing to the first anomaly.
6 . A system comprising: one or more memory devices; and one or more server devices configured to cause the system to, while operating subterranean drilling equipment having a plurality of operating states, wherein the subterranean drilling equipment includes a pump, each of the plurality of operating states defined by a plurality of operation features, each operation feature of the plurality of operation features defining operation of the subterranean drilling equipment over time: collect, from the pump, first sensor data related to the operation of the subterranean drilling equipment, wherein the first sensor data includes a pump flowrate; identify an operating state of the plurality of operating states of the subterranean drilling equipment, wherein the operating state includes a pre-connection activity, the pre-connection activity including at least one of drilling off, moving a drill string to a connection point, or stopping rotation and the pump; identify first time-series data for the subterranean drilling equipment from the first sensor data; identify contextual data that provides context to the first time-series data, wherein the contextual data includes drilling fluid data of a drilling fluid; using the first time-series data and the contextual data, partition the first time-series data by splitting the first time-series data into a plurality of data buckets based on the operating state of the subterranean drilling equipment, each data bucket of the plurality of data buckets including partitioned first time-series data; extract, utilizing a feature extraction model, the plurality of operation features, each operation feature of the plurality of operation features defining the operation of the subterranean drilling equipment over time by filtering the partitioned first time-series data to estimate feature signals based on smoothing one or more data spikes from signal noise of the first time-series data in a manner that reduces temporal lag when provided as an input to the feature extraction model, the feature signals comprising at least one of velocity, acceleration, waveform peaks, or waveform troughs, wherein the feature extraction model includes a feature machine-learning model trained to determine the plurality of operation features; generate a plurality of first feature probabilities, each of the plurality of first feature probabilities associated with an operation feature of the plurality of operation features, wherein generating the plurality of first feature probabilities includes converting discrete feature data from the feature signals to continuous feature data, wherein the plurality of first feature probabilities are estimated likelihoods that the operation feature of the plurality of operation features associated with a feature probability of the plurality of first feature probabilities corresponds to a certain value; provide, for display within a graphical user interface, a feature curve representation of a plurality of feature curves for the plurality of operation features, the feature curve representation including a plurality of feature curve clusters, wherein the plurality of feature curves are displayed on the feature curve representation based on an anomaly threshold, and wherein each of the plurality of feature curve clusters are selectable; receive, at the graphical user interface, a selection of a feature curve cluster of the plurality of feature curve clusters; based on the selection of the feature curve cluster, populate an anomaly visualization on the graphical user interface, the anomaly visualization including the plurality of feature curves based on the selection of the feature curve cluster; receive a user input at a feature probability slider on the graphical user interface, wherein the user input increases the anomaly threshold for the plurality of first feature probabilities; using the user input, update the anomaly visualization by removing a feature curve of the plurality of feature curves based on removed feature probability of the feature curve being less than the anomaly threshold set by the user input; identify, on the anomaly visualization and using the anomaly threshold set by the user input, a first anomaly of the operation of the subterranean drilling equipment by comparing the plurality of first feature probabilities for the plurality of operation features to the anomaly threshold, wherein the first anomaly is an abnormality of the partitioned first time-series data, wherein identifying the first anomaly includes identifying a probability of the first anomaly being anomalous using an anomaly machine-learning model for probability density estimation; based on the first anomaly and the operation feature, identify a lost time for a drilling operation associated with the operating state and the contextual data; based on the first anomaly and the lost time, determine a corrective action to be performed by the subterranean drilling equipment, the corrective action including adjusting the drilling fluid; based on the corrective action, adjust the drilling fluid by switching at least a portion of the drilling fluid for use by the subterranean drilling equipment; after switching the drilling fluid, collect second sensor data and identify second time-series data from the second sensor data; partition the second time-series data, filter the partitioned second time-series data, and extract the plurality of operation features from the filtered partitioned second time-series data; and generate a plurality of second feature probabilities and identify a second anomaly of the operation of the subterranean drilling equipment.
7 . The system of claim 6 , wherein the one or more server devices are configured to cause the system to smooth the one or more data spikes from the signal noise of the first time-series data using a zero-lag Difference of Gaussian filter.
8 . The system of claim 6 , wherein the one or more server devices are configured to cause the system to generate the plurality of first feature probabilities for the plurality of operation features by utilizing a non-parametric model to estimate a probability density function based on the discrete feature data from the feature signals.
9 . The system of claim 6 , wherein the one or more server devices are configured to cause the system to compare the plurality of first feature probabilities for the plurality of operation features to the anomaly threshold by: identifying a minimum feature probability of the plurality of first feature probabilities; and comparing the minimum feature probability to the anomaly threshold, the anomaly threshold being a preset or user-configurable value.
10 . A computer-implemented method comprising: while operating subterranean drilling equipment having a plurality of operating states, each of the plurality of operating states defined by a plurality of operation features, wherein the subterranean drilling equipment includes a pump, each operation feature of the plurality of operation features defining operation of the subterranean drilling equipment over time: collecting, from the pump, sensor data related to the operation of the subterranean drilling equipment, wherein the sensor data includes a pump flowrate; identifying an operating state of the plurality of operating states of the subterranean drilling equipment, wherein the operating state includes a pre-connection activity, the pre-connection activity including at least one of drilling off, moving a drill string to a connection point, or stopping rotation and the pump; identifying time-series data for the subterranean drilling equipment from the sensor data; identifying contextual data that provides context to the time-series data, wherein the contextual data includes drilling fluid data of a drilling fluid; using the time-series data and the contextual data, partitioning the time-series data by splitting the time-series data into a plurality of data buckets based on the operating state of the subterranean drilling equipment, each data bucket of the plurality of data buckets including partitioned time-series data; filtering the partitioned time-series data to estimate feature signals based on smoothing one or more data spikes from signal noise of the time-series data in a manner that reduces temporal lag when provided as an input to a feature extraction model, the feature signals describing the plurality of operation features; extracting, utilizing the feature extraction model and from the feature signals, the plurality of operation features, wherein the feature extraction model includes a feature machine-learning model trained to determine the plurality of operation features; generating a plurality of feature probabilities, each of the plurality of feature probabilities associated with an operation feature of the plurality of operation features, wherein the plurality of feature probabilities are estimated likelihoods that the operation feature of the plurality of operation features associated with a feature probability of the plurality of feature probabilities corresponds to a certain value; providing, for display within a graphical user interface, a feature curve representation of a plurality of feature curves for the plurality of operation features, the feature curve representation including a plurality of feature curve clusters, wherein the plurality of feature curves are displayed on the feature curve representation based on an anomaly threshold, and wherein each of the plurality of feature curve clusters are selectable; receiving, at the graphical user interface, a selection of a feature curve cluster of the plurality of feature curve clusters; based on the selection of the feature curve cluster, populating an anomaly visualization on the graphical user interface, the anomaly visualization including the plurality of feature curves based on the selection of the feature curve cluster; receiving a user input at a feature probability slider on the graphical user interface, wherein the user input increases the anomaly threshold for the plurality of feature probabilities; using the user input, updating the anomaly visualization by removing a feature curve of the plurality of feature curves based on removed feature probability of the feature curve being less than the anomaly threshold set by the user input; identifying, on the anomaly visualization and using the anomaly threshold set by the user input, an anomaly of the operation of the subterranean drilling equipment based on the plurality of feature probabilities for the plurality of operation features, wherein the anomaly is an abnormality of the partitioned time-series data, wherein identifying the anomaly includes identifying a probability of the anomaly being anomalous using an anomaly machine-learning model for probability density estimation; based on the anomaly and the contextual data, determining a corrective action to be performed by the subterranean drilling equipment, the corrective action including adjusting the drilling fluid; and implementing the corrective action with the subterranean drilling equipment.
11 . The computer-implemented method of claim 10 , further comprising determining the plurality of feature curve clusters, wherein the plurality of feature curve clusters represent one or more of the plurality of operation features associated with a plurality of time-series data.
12 . The computer-implemented method of claim 11 , further comprising: determining, for the operation feature of the plurality of operation features, a difference score between two or more of the plurality of feature curve clusters; and updating the graphical user interface with the operation feature of the plurality of operation features together with the difference score between the two or more of the plurality of feature curve clusters.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. provisional application No. 63/199,293 filed on 18 Dec. 2020 and titled “Similarity and Anomaly Recognition in Drilling Operations”, which is hereby incorporated herein in its entirety by reference. BACKGROUND Recent years have seen significant improvements in extracting and identifying operational performance data associated with subterranean drilling. Unfortunately, a number of problems still exist with conventional systems for identifying operation anomalies. For example, conventional drilling anomaly systems implement key performance indicators or other aggregate measures of drilling operation processes that suffer from low interpretability. In addition, certain conventional drilling anomaly systems are not capable of real-time anomaly identification. Moreover, some conventional drilling anomaly systems promote selective (and subjective) review of certain drilling parameters that may appear anomalous but are not. To illustrate, conventional drilling anomaly systems can measure drilling operation processes, but these systems often fail to measure drilling operation processes in a way that provides constructive feedback for improving the measured process. For instance, conventional drilling anomaly systems use key performance indicators (or other aggregate measures). However, these indicators are often averages or other statistical values that, of themselves, are difficult for field personnel to interpret and/or develop improvement plans for the particular drilling operation process. Accordingly, key performance indicators are often perceived as too vague/complex to understand. Other alternatives, such as histograms, likewise fail to provide an effective mechanism for improving a measured drilling operation process. In addition to a lack of interpretability, conventional drilling anomaly systems are often of little use in real-time field operation. For example, some conventional drilling anomaly systems use key performance indicators or other measures that aggregate drilling operation data over time. Accordingly, such conventional drilling anomaly systems are typically incapable of identifying anomalous drilling operation processes as they occur because a key performance indicator is still (over the aggregate) within tolerance or an accepted range. As a result, conventional drilling anomaly systems operate with reduced accuracy and real-time effectiveness. Based in part on the foregoing deficiencies, some conventional drilling anomaly systems promote selective (and subjective) review of certain drilling parameters. For example, a drilling engineer in the field may conduct a post-drilling-session review of a drilling session average for one or more drilling parameters relative to a historical aggregate of drilling sessions. Such manual approaches often fail to produce accurate results. Indeed, identified anomalies are rarely actual anomalies, and perceived normal data is not necessarily normal. These common discrepancies are due to the myriad different variables that mere observation and the human mind cannot practically capture with any consistent degree of accuracy. Indeed, the complex interplay between the various drilling parameters (e.g., hookload, block position, revolutions per minute, pump flow rate, rate of penetration, etc.), rig states (e.g., pre-connection activities, connection activities, and post-connection activities), contextual data (e.g., drilling operator, date and time, geological formation, drilling metric, bottom-hole assembly, drilling fluid, etc.), and other contributing factors is beyond the mental capacity of the human mind to evaluate—let alone determine anomalies. SUMMARY Aspects of the present disclosure can include methods, computer-readable media, and systems that dynamically utilize a feature extraction model to determine behavior anomalies in time-series drilling data. In particular, the disclosed systems partition the time-series drilling data into similar activities, such as pre-connection activities, connection activities, and rotary drilling. From the partitioned data, the disclosed systems extract a collection of features using a feature extraction model. Such a collection of features includes, for instance, maximum or minimum velocity (and/or acceleration) of a traveling block, maximum and minimum block height, total block time moving upwards and downwards, etc. In one or more embodiments, the disclosed systems determine a corresponding probability density function for each feature. Subsequently, the disclosed systems determine an anomaly based on a minimum probability for one or more features satisfying an anomaly threshold. Additional features and advantages of one or more embodiments of the present disclosure are outlined in the following description. BRIEF DESCRIPTION OF THE DRAWINGS The detailed description provides one or more embodiments with additional specificity and detail t