New Generation Impulse Testing: An Environmentally Friendly Well Testing Solution for Tight Reservoirs to Acquire Accurate Reservoir Information (Russian)

Shumakov, Yakov; Hollaender, Florian; Williams, Andrew; Munro, Jock; Theuveny, Bertrand

doi:10.2118/196840-ru

Cited by 1 publication

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Such conditions can lead to an increased degradation rate of different subsystems in these tools and failures. Consequently, operations may be compromised, as the tools may provide inaccurate information, deliverables maybe be delayed until the tool is repaired, or the whole operation may be cancelled (Shumakov et al, 2019;Gramcko Contreras et al, 2013). This leads to nonproductive time and financial losses.…”

Section: Introductionmentioning

confidence: 99%

Data-Driven Fault Detection Method for Electronic Boards in Intelligent Remote Dual-Valve System

Bhatnagar¹,

Cassou²,

Masry

et al. 2021

PHME_CONF

View full text Add to dashboard Cite

Intelligent remote dual-valve (IRDV) tools are used to open and close two valves downhole in a well. One valve is aimed to control the flow from the formation being tested into the pipe string to the surface, which is known as the test valve. The other one is aimed to control the flow between the annulus and the ID of the test string and is called the circulating valve. Both valves are activated in the same manner using a power piston that responds to the pressure applied on its lower side (cylinder area) by the hydraulic actuator, which is forced up or down to operate the attached valve. The lower section of the tool is the actuator assembly, comprising a hydraulic section and an electronic section. The electronic section responds to annulus pressure commands and drives the hydraulic section to distribute hydraulic pressure to the valves. Such tools operate in extreme environment conditions, such as elevated temperature, shocks, vibrations, and pressures. Such conditions can lead to an increased degradation rate of different subsystems in these tools and failures. Consequently, operations may be compromised, as the tools may provide inaccurate information, deliverables maybe be delayed until the tool is repaired, or the whole operation maybe cancelled. This leads to nonproductive time and financial losses. To avoid such failures, field engineers are required to check the tool condition after each run by analyzing sensor signals recorded during tool operation and determine if the tool is ready for the next run or if it is faulty and should be sent to maintenance. A reliable analysis, however, is extremely difficult to perform due to the immense number of data channels generated at a record rate, which results in millions of data points from a single run. A manual analysis of this data is extremely time consuming in an environment that can be extremely time critical, and the complexity of the signals limits the effectiveness of manual analysis. An alternative approach is to identify critical subsystems in the tool and allow a domain expert to identify the channels that contain information about the tool condition and possible degradation of each subsystem. Statistical features are extracted from channels that indicate degradation of the system with time. Later, these features can be used to build machine learning models that estimate the tool condition from the recorded data. In this tool, an electronics subsystem was identified as the most critical component in the pareto chart, and thus it was decided to develop a fault detection algorithm to help the field users identify whether the electronics behaved as expected or not. In this work we present a data-driven fault detection method for electronic boards in an IRDV system. Data used to build this model consists of electronic sensor channels monitored over time. Characteristics and distribution of the data were analyzed through exploratory data analysis. The method is based on extracting relevant features from multiple channels. Classification prediction modeling is done on these features, and the features are used to a build support vector machine model that monitors and estimates the health status of the IRDV system after a particular job to learn patterns. The IRDV has two kinds of operating channels—slow channels and activation channels—with corresponding different components associated with them. Modeling is done on both these channels to provide a segregated classification. The model demonstrated excellent value for maintenance and field engineers, because in a few minutes, the physical condition of the IRDV electronics subsystem can be determined using run data. This work is part of a long-term project aiming to construct a digital fleet management system for downhole testing tools.

show abstract

Section: Introductionmentioning

confidence: 99%