Page 182 - Intelligent Digital Oil And Gas Fields

Components of Artificial Intelligence and Data Analytics 139

4.3.3 Event Diagnostics and Failure Analysis
As the applications of Big Data and predictive modeling are rapidly expan-
ding in E&P, they are being used in predictive maintenance, event diag-
nostics, and failure analysis of a wide range of equipment in oil and gas
operations, mainly for drilling and artificial lift (e.g., ESPs) and rotating
equipment diagnostics (e.g., compressors and turbines).
The term failure analysis basically stems from the branch of MA, called
survival analysis (Tabachnick and Fidell, 2013), which is a set of data-driven
statistical techniques (e.g., data mining) for analyzing the length of time until
something happens (i.e., an unplanned event) and for determining if that
time differs for different groups of samples or for groups subjected to differ-
ent treatments. For example, in medical settings, survival analysis is used to
determine the time course of various medical conditions and whether dif-
ferent modes of treatments produce changes over time. In industries such as
oil and gas, such analysis is referred to as failure analysis, and it is used to
determine time until failure of a specific equipment part and whether parts
manufactured differently have different rates of failure.
Mirani and Samuel (2016) have presented data analytics-based workflows
for monitoring and mitigating drill-string failures caused by tool vibrations.
The proposed workflow integrates the modified vibration stability plot with
the data analytics tool to predict drill-string failures caused by torsional and
lateral vibration. In drilling operations, the modified stability plot provides
optimum operating parameters—including weight on bit (WOB), revolu-
tions per minute (RPM), and ROP—to minimize vibration. However,
the actual real-time generated drilling parameters are not always optimum.
Mirani and Samuel (2016) have used the deviation of real-time parameters
from optimum values and the tools of unsupervised statistical learning
(i.e.,dataclustering)tocalculatedeviationvectors,representativeofthemisfit
fromtheoptimumpoint.Thederivedstabilityclustersarethenusedforquan-
titative failure mitigation. Moreover, they address the question of, “how long
can drilling operations be performed if the measured data remains outside
of stability cluster before tool failure occurs.” The data-driven calculation
of the cumulative vibration risk index provides a more sound technique
for risk quantification.
Kale et al. (2015) have proposed another application for optimizing
operational performance and failure prevention management of drilling sys-
tems using real-time data and predictive analytics. They have proposed the
framework and algorithms for constructing data-driven component life
models to optimize operational efficiency and extend the life of a drilling