ADVERTISEMENT

Risk-Based Statistical Approach To Predict Casing Leaks

Source: Getty Images.

You have access to this full article to experience the outstanding content available to SPE members and JPT subscribers.

To ensure continued access to JPT's content, please Sign In, JOIN SPE, or Subscribe to JPT

Currently, the predominant technology for measuring casing integrity is electromagnetic corrosion logging. While this technology has improved the ability to measure and monitor corrosion, the findings usually are not conclusive and need to be integrated with other data to enable qualitative assessments. A probabilistic approach was introduced to improve interpretation of data from electromagnetic corrosion logs. This paper presents a statistical risk-based approach to predicting casing leaks using electromagnetic corrosion logs.

Introduction

Sound well-integrity-management strategy in mature fields, where wells can sustain economic production for 30 to 50 years, is vitally important. Failing to implement this strategy can lead to a catastrophic loss of both assets and human life. One example of such loss is surface leaks that are caused by downhole multiple casing impairments caused by active shallow aquifer corrosion.

Corrosion logging provides the most-direct measurement of casing integrity and can be used as a predictive measure as well. Mechanical, ultrasonic, and electromagnetic tools are three main types of corrosion-logging tools. A mechanical multifinger tool uses multiple high-resolution calipers to measure slight changes in the internal diameter of tubing and casing strings. The tool deploys an array of hard-surfaced fingers that monitors the inner pipe wall. Each of the sensors generates an independent signal that is recorded against depth.

Electromagnetic Induction Tool (EMIT)

This paper focuses on the interpretation challenges associated with EMITs (Fig. 1), a casing-corrosion-logging technology. The tool uses three types of noninvasive electromagnetic measurements to characterize well casings, using low-, medium-, and high-frequency induction currents that are related to the casing-wall thickness, inside diameter, and permeability or conductivity. The lower the frequency, the deeper the penetration is to the outer casings.

Fig. 1—EMIT.T=Transmitter;  R=Receiver.

 

Each parameter is averaged around the pipe circumference. The tool has multiple transmitters and receivers to send and receive electromagnetic signals. It detects average metal loss and changes in casing geometry irrespective of fluid type.

Despite its ability to assess metal loss in multiple casings, the tool can only read an azimuthal average loss across multiple strings. Consequently, wells with casing failures definitely will show average-metal-loss values of less than 100% unless the failure occurs around the 360° circumference. In other words, 50% average metal loss could mean a failure if one part of the casing is completely lost and another is intact.

Average Remaining Barriers Ratio (ARBR)

Data to be interpreted in a typical EMIT corrosion log are total remaining thickness and average-metal-loss percentage at all hot spots. The presiding assumption, made on the basis of which EMIT data are analyzed, is that casing corrosion is external where casing strings are lost one by one. The metal-loss breakdown per each string thus can be easily calculated. The ARBR is the average number of remaining strings between the corrosive zones, usually water-bearing formations, and the wellbore divided by the number of nominal strings at a certain corrosion-growth hot spot. It is a normalized parameter that accounts for different combinations and sizes of downhole casings.

Uncertainty Quantification of EMIT Metal-Loss Data

Data. A cross-field data collection was conducted to assess the variation of EMIT responses across known leaking and nonleaking average-metal-loss hot spots. A leaking average-metal-loss hot spot is defined as the one measured by EMIT across confirmed multiple casing leaks or failures. Casing failures are confirmed by different well-integrity diagnostic methods, some of which are rigless, such as annuli and temperature surveys, and some of which are performed during rig intervention, such as positive and negative pressure tests. On the other hand, a nonleaking average-metal-loss hot spot is the absolute opposite, where the EMIT measurements are taken across confirmed healthy multiple casing conditions.  

Average-Metal-Loss ­Uncertainty. EMIT average-metal-loss measurements are subject to interpretation uncertainty. The fact that EMIT does not provide directional metal-loss readings presents the challenge of defining an average metal-loss cutoff beyond which a maintenance-workover decision is made before the well casing fails. The first step toward solving this challenge is to quantify the uncertainty using well-known statistical methods. In this work, a fundamental assumption is made that both leaking and nonleaking average-metal-loss data are normally distributed. Normality tests were performed on both data sets to validate this assumption. The normality test is a plot of the data ranked in an ascending order vs. their percentiles on a normal probability scale. A straight-line trend of the plotted data is the condition upon which the data set is classified as normally distributed.

Both plots characterize the two data sets (leaking and nonleaking) as normally distributed. After identifying the distribution of both data sets, their probability density functions (PDFs) can be written in a mathematical form to generate their normal probability distribution curves. These curves are plotted to relate the PDF with average metal loss for both leaking and nonleaking hot spots.

Analyzing the plots reveals and quantifies the uncertainty of EMIT average-metal-loss measurements. The average-metal-loss range defines the interval within which a nonuniform external corrosion is likely to occur, causing multiple casing failures. In fact, a casing leak is expected whenever an average-metal-loss value is between 30 and 70%. Also, the probability of a casing failure can be inferred to be unity above 70% and zero below 30% average metal loss.  

ARBR Uncertainty. ARBR is a normalized mathematical transformation of the average-metal-loss parameter. The reason for introducing ARBR is that the metal-loss value does not address the number of nominal casings or their thicknesses or combinations. This research assumes that EMIT response to average thickness loss in multiple casings is a strong function of these factors and not just the average-metal-loss percentage. Accordingly, the subsequent modeling of multiple casing failures will focus on ARBR as the statistical random variable of the probability distributions.

ARBR data will be categorized into leaking and nonleaking data sets. A normality test also is performed to decide whether the two data sets are normal.

Both plots characterize the two ARBR data sets as normally distributed. Generally, and similar to the average-­metal-loss data sets, leaking-hot-spot data correlate more linearly on a normal probability scale than nonleaking-hot-spot data.

After identification of the distribution of both data sets, their PDFs can be written in a mathematical form to generate their normal probability distribution curves. These curves are plotted to relate the PDF with ARBR for both leaking and nonleaking hot spots.

Similar to the average-metal-loss analysis, these plots reveal the uncertainty associated with ARBR. The ARBR range defines the interval within which a nonuniform external corrosion is likely to occur, causing multiple casing failures. In fact, a casing leak is expected whenever an ARBR is between 0.4 and 0.8. Also, the probability of a casing failure can be inferred to be unity when ARBR values are below 0.4 and zero when they are above 0.8.

Conclusions

  • Most EMIT data acquired are more qualitative than quantitative because the tool provides circumferential average readings of metal-thickness loss across multiple strings rather than directional readings.
  • EMIT technology uncertainty can be quantified by studying the average-metal-loss measurements across leaking and nonleaking metal-loss hot spots.
  • Introducing the ARBR concept normalizes the effects of multiple casing combinations, grades, and thicknesses.
  • The assumption that the EMIT average-metal-loss and ARBR data are normally distributed is fairly applicable.
  • A casing leak is expected whenever an average-metal-loss value is between 30 and 70%. Also, the probability of a casing failure can be inferred to be unity when average metal loss is above 70% and zero when it is below 30%.  
  • A casing leak is expected whenever an ARBR is between 0.4 and 0.8. Also, the probability of a casing failure can be inferred to be unity when an ARBR value is below 0.4 and zero when it is above 0.8.
This article, written by Special Publications Editor Adam Wilson, contains highlights of paper SPE 183948, “Risk-Based Statistical Approach To Predict Casing Leaks,” by Mohammed D. Al-Ajmi, SPE, Saudi Aramco; Dhafer Al-Shehri, SPE, King Fahd University of Petroleum and Minerals; and Nasser M. Al-Hajri, Abdullrahman T. Mishkes, Muhammad A. Al-Hajri, SPE, and Nayef S. Al-Shammari, SPE, Saudi Aramco, prepared for the 2017 SPE Middle East Oil and Gas Show and Conference, Manama, Bahrain, 6–9 March. The paper has not been peer reviewed.

Risk-Based Statistical Approach To Predict Casing Leaks

01 June 2018

Volume: 70 | Issue: 6

STAY CONNECTED

Don't miss out on the latest technology delivered to your email weekly.  Sign up for the JPT newsletter.  If you are not logged in, you will receive a confirmation email that you will need to click on to confirm you want to receive the newsletter.

 

ADVERTISEMENT

ADVERTISEMENT

ADVERTISEMENT

ADVERTISEMENT

ADVERTISEMENT