CML Optimization from Inspectioneering.com

Piping CML Optimization: Optimizing the Definition

By Mike Sparago, Principal Engineer II at The Equity Engineering Group, and Chris Massengale, Consultant Engineer II at The Equity Engineering Group. This article appears in the January/February 2021 issue of Inspectioneering Journal.

 Introduction

“Condition monitoring location (CML) optimization” is a term we frequently hear, particularly for piping circuits. Everyone wants their circuits to be “CML optimized,” but what does this truly mean? Borrowing a term from API RP 581, Risk-based Inspection Methodology, the broad definition of CML optimization as it relates to internal thinning is easy: we wish to establish the minimum CML coverage to correctly identify the true damage state.[1] Note that the true damage state is related to remaining life in traditional, time, or condition-based inspection programs. Although we sometimes focus on having sufficient CML coverage to correctly define the corrosion environment, corrosion rates are only part of the picture. Projected wall thickness is the ultimate goal when it comes to inspection, repair, and replacement planning. 

While the broad definition for CML optimization is easy, breaking this definition down to specific methods, algorithms, and business logic is not so straightforward. What is the basis for a given “optimization?” How precisely do we need to characterize the corrosion environment so we can quantify the amount of wall loss? Will CML quantities and selections be based on expert opinion, statistical methods, or a blend of both? How is the inspector, facility, company, and industry experience incorporated into the analysis? Are we optimized for a snapshot in time or is there an allowance for future degradation? There are as many answers to these questions as there are methods presently employed.

Setting the Ground Rules

For the purpose of this discussion, we will be focusing on wall thickness data taken at prescribed locations to monitor internal corrosion. In this context, “CMLs” are actually thickness monitoring locations. We normally generate this data using ultrasonic thickness (spot or scanning) and profile radiographic techniques. 

At a minimum, we expect CML optimization to provide guidance in proper CML quantities and placement. It must be consistent with the assigned, internal thinning damage mechanisms, while also considering site and industry experience. Additional benefits resulting from this dedicated, focused review of historical thickness data and inspection practices will often include the following:

• Identification of and follow up inspection for potential localized corrosion areas
• Recommendations to establish new CMLs or archive / inactivate unneeded CMLs
• Identification and resolution of data anomalies, with resulting database “cleanup”
• Opportunities to reduce measurement error and improve repeatability
• Adjustments to circuit boundaries: combining, splitting, or moving boundaries, as appropriate
• Prioritization for circuits that are candidates for 100% component inspection

As will be discussed later, the specific benefits will depend on many factors, including the quality and quantity of historical data, the corrosion environments, and the capabilities of the data analysis process.

How can we be so certain?

Let’s break down why we perform piping on-stream thickness inspections. We are going to continue focusing on piping because most other equipment types (vessels, exchangers, tanks) offer an unfair advantage. That advantage, of course, is internal inspection. For major fixed equipment, periodic internal inspections not only drive CML quantities and placement but offer an additional layer of protection since internal inspections can provide early clues for localized corrosion. If we encounter unexpected thinning during an internal inspection, we routinely add appropriate on-stream monitoring. In terms of sample size, an internal inspection provides orders of magnitude of greater “coverage” than can be realistically achieved via ultrasonics (even including scanning UT methods) or profile radiography testing. With piping circuits, except for exceptionally large diameter components (or perhaps incredibly small inspectors), spot UT, UT scanning/AUT, and profile RT provide our only glimpse of a component’s inner wall, from which we must interpret internal thickness profiles and thinning vulnerabilities. 

Read Related Articles

• CML Optimization: Effective and Efficient Coverage and Placement
• Avoiding 5 Common Pitfalls of Pressure Vessel Thickness Monitoring
• A Discussion on the Piping Thickness Management Process - Part 2: Determining Corrosion Monitoring Locations
• CML Reduction – Are you planning to fail?
• Improving Confidence in On-Stream Inspections of Pressure Equipment

The big picture, of course, is that we inspect to reduce uncertainty. We can, and sometimes do, estimate the remaining life or probability of failure (PoF) for components using only nominal thicknesses and estimated corrosion rates. However, we generally do not rely on such assumptions because there is considerable uncertainty in the corrosion rate and component thickness estimates. We recognize that obtaining and analyzing wall thickness measurements allows us to better understand the corrosion environment, which improves our ability to plan repairs/replacements and to minimize failures. Despite obstacles such as measurement and location repeatability errors, imperfect circuitization/CML placement, and non-uniform corrosion, the total uncertainty is reduced with appropriate on-stream inspection and analysis. 

Inspectors know that more CMLs, more examination points (EPs), and more frequent surveys produce less uncertainty in our quest to define the corrosion environment. However, since the amount of surface area we can reasonably inspect using these inspection techniques is extremely small, and given that all internal corrosion environments are not created equal, we must rely on other sources of data to achieve the needed uncertainty reductions. Commonly used data sources to reduce uncertainty are industry standards, proper circuitization, damage mechanism assignment, and inspector/plant/industry experience.

“Refined” Defined

Recognizing that our goal is to reduce uncertainty, let’s refine our earlier CML optimization definition to state that we wish to establish the minimum CML coverage in order to project the true damage state with an acceptable level of uncertainty. This provides some clarity, but our definition is not yet quantitative. Like a dependable inspection technique, we need to have a documented, quantitative basis with sound acceptance criteria before embarking on a CML optimization project. In general, CML optimization may identify locations requiring additional CMLs or higher inspection effectiveness techniques, but it may also reveal historical CMLs that can be archived or potentially eliminated. The use of a quantitative optimization process is particularly important in the latter case to ensure that a rigorous basis was used to archive or remove CMLs from a circuit.

Note that this refined definition references thickness projections, not just the current thickness state. Indeed, we prefer an optimized inspection plan to be valid for multiple inspections; ideally, as far as 20 years into the future. We do not want to make unnecessary adjustments (i.e., database, isometrics, insulation removal, etc.) to our inspection plans after they have been “optimized.”

Quantitatively Speaking

In order to define CML optimization more quantitatively, we need a means to model total uncertainty for all component thickness distributions in a circuit. This requires a better understanding of the individual sources of uncertainty, including:

1. Component original/nominal thickness distributions
2. The corrosion process, including non-idealities in circuit design
3. The various sources of measurement error

These are challenging inspection program and analytical elements, not only for CML optimization but for piping thickness programs in general. Such challenges represent some of the main reasons that piping experiences higher failure frequencies.

In developing a more quantitative definition for CML optimization, we need to consider the basis for the resulting inspection planning decisions. Experience will always play a vital role in inspection planning; however, relying on experience alone yields a qualitative, subjective CML optimization, or at best, one that is semi-quantitative. The other extreme is a purely statistical approach, which often fails because statistics may not address the practical insights needed for successful inspection planning. The best solution to the CML optimization problem is one that combines our knowledge and experience about corrosion environments, component thicknesses, and measurement processes with the quantitative power of advanced modeling. The optimal solution also requires a model that is carefully constructed to mirror our understanding of how thickness degradation occurs over time, for the wide range of corrosion environments we encounter.

Leveraging Knowledge and Experience

Every day, we make inspection and reliability decisions by combining field data with knowledge and experience. For example, simply confirming that measured corrosion rates are consistent with the assigned damage mechanisms increases the power of both our data and our beliefs. Recent improvements in computing power and analytical techniques allow us to take this data/knowledge combination process to a new level, by making Bayesian modeling practical for inspection planning—and specifically CML optimization.

Transforming our knowledge and experience of component thicknesses, corrosion environments, and measurement processes into Bayesian prior distributions is straightforward and intuitive. These “priors” allow us to express our degree-of-belief for important CML optimization elements in a quantitative and documented manner. In most cases, we can assign reasonably “informative” priors, given the assigned damage mechanisms and our specific experience with the environment. Combining these prior distributions with historical thickness data yields posterior distributions for model parameters, providing improved wall thickness versus time projections. These thickness projections form the basis for a state-of-the-art CML optimization.

So, Bayesian modeling gives us the best of both worlds. We can quantitatively integrate our knowledge and experience with historical data, providing for more accurate corrosion rate distributions and future thickness projections. But how does Bayesian modeling address our concerns about uncertainty?

New Approach to an Old Practice

We have seen how uncertainty principles, visualized through credible intervals, are used to determine if CMLs can be archived or eliminated. We have also seen how uncertainty can be used to guide re-inspection schedules and model inspection effectiveness for existing CMLs. In a CML optimization study, however, we want to account for all components—not just monitored CMLs—in a circuit. Many of the early CML quantity practices, some of which are still in use today, begin with a count of pipe lengths plus fittings as a measure of circuit complexity. Then, a fixed percentage of components to be monitored is often assigned based on the expected corrosion environment.

The more rigorous modeling equivalent to this practice is to include all unmonitored components in the CML optimization modeling.

Unmonitored components are considered to have the specified nominal thickness per the relevant pipe specification and the original thickness distribution is estimated based on typical values for the specific size/component combination. Hierarchical models allow unmonitored components to “borrow strength” from CML and group level corrosion rate distributions elsewhere in a circuit.

All for One, One for All

We have thus far ignored the fact that uncertainty for any CML is impacted by the historical data for all CMLs in that circuit. Given reasonable circuitization, we recognize that every valid reading in a circuit should reduce the uncertainty for all CMLs. Accordingly, the (95%) upper, and more importantly, the (5%) lower credible intervals on our thickness projection plots will shrink for both CMLs and unmonitored components, as new readings are added to the circuit. 

Top: All Historical Data, Bottom: Removed Data from Three CMLs

Since we need to include all components in our CML Optimization and recognize that uncertainty is reduced with incremental historical data, we can complete our definition as follows:

CML optimization is a data-driven process used to establish the minimum CML coverage for all circuit components, with the goal of projecting the true damage state within a specified uncertainty.

Conclusion

We have seen how the concept of uncertainty can be used to effectively solve CML optimization and related inspection planning problems. Credible intervals provided by hierarchical Bayesian modeling allow us to visualize uncertainty while utilizing our knowledge and experience in a quantitative manner. As a result, these modeling techniques provide improved CML optimization results when compared to experience or statistically based approaches.

References

1. API RP 581, 2016, Risk-based Inspection Methodology, 2016, American Petroleum Institute