Risk modelling for critical pipeline assets

The better a utility can predict failures, the better it can avoid them by targeting the pipes most in need of repair or replacement.

By Graham Bell and Peter Martin

Utilities across North America face rising costs due to pipe failures, as these assets continue to deteriorate and cities continue to grow. More utilities are now using risk models to plan their pipe replacement programs. Recent emphasis is away from highly-subjective factor-based systems and towards solutions that measure risk in a quantitative way to allocate funds to at-risk assets.

Xylem’s Decision Intelligence approach to smart water management is helping utilities see a higher return on investment (ROI) for their capital and operational spending through a combination of sound data collection and management, machine learning and loss of service simulations.

Since 2013, Xylem’s team of data scientists, engineers and asset management experts have been generating advanced risk pipeline models for utilities around the world. By combining advanced data models like machine learning with a utility’s existing dataset, they can produce a “probability of failure” (POF) for each asset.


Subscribe to our Newsletter!

The latest environmental engineering news direct to your inbox. You can unsubscribe at any time.

While most risk models are made up of “likelihood of failure” (LOF) and “consequence of failure” (COF), Xylem’s includes POF and COF. Using POF produces results as a probability-based percentage, rather than a subjective factor. Additionally, each POF value is linked to the utility’s geographic information system, representing the chance a pipeline or section of pipe could fail in the near future.

To help a utility understand why an asset might fail, Xylem takes a “glass box” approach to machine learning to help utilities understand their POF results. This transparent approach allows utilities to understand the contributing factors driving POF for their pipelines and to better target mitigation strategies that have the most impact (rather than simply relying on a replace-only approach).

A utility that is able to mitigate failures in high POF areas by addressing key contributing factors can significantly increase the ROI for their programs and reduce outage and financial impacts to their customers. It has also been proven that the higher accuracy of machine learning to forecast future failures, compared with subjective factor-based systems, results in better choices for pipelines that need replacement.

In other words, the better a utility can predict failures, the better they can avoid them by targeting the pipes most in need of repair or replacement. Without an objective and quantitative approach to POF, a utility might not be realizing the highest possible ROI from their decisions.

A quantitative approach

The most common practice for calculating risk is using factors for likelihood and consequence of failure, then weighting each factor based on judgement. Risk (sometimes referred to as “business risk exposure” or BRE) is then calculated as LOF x COF. Another way to describe this approach is a quantitative risk model. The problem with these factor-based methodologies is that results can misrepresent risk, regardless of the scale used for each category.

For example, on a standard factor-based LOF and COF matrix, an asset with a LOF of 1 and COF of 5 will have the same risk score as an asset with a POF of 5 and a COF score of 1.

A simple matrix fails to identify differences between results because COF doesn’t properly scale with the importance of an asset. To address COF from a quantitative perspective, Xylem utilizes a monetized COF to combat the problems brought on by factor-based solutions.

Their COF follows the “triple bottom line (TBL) approach and focuses on three main categories for estimating the impact a failure could have on a community:

  • Economic – costs to the utility including repair costs and revenue losses.
  • Social – the costs to the community and businesses due to loss of service or flooding.
  • Environmental – remediation costs or fines from failures affecting a protected or sensitive habitat.

The combined output of the TBL approach is then ranked into tiers that are, in turn, prioritized by individual asset POF. Using this approach eliminates the possibility that critical assets with high COFs and low POFs will be overshadowed when ranked against assets with high POFs and low COFs.

For example, Asset A that has a POF of 0.05% and a COF of $1,000,000 would have a score of $500, which is similar to Asset B with a POF of 50% and a COF of $1,000. Although they have the same risk (or BRE) score, a utility can choose different asset management strategies for A and B.

Asset A may be considered too consequential to fail and requires more rigorous O&M policies, while Asset B may be better off running until failure. By properly tiering a quantitative COF, a utility can use the results to manage assets more effectively. (See Figure 1)

Figure 1 chart
Figure 1. Tiering a quantitative consequence of failure.

Grouping sections of pipe into impact tiers helps designate certain assets for different strategies and increase the ROI for each dollar spent mitigating risk.

Dynamic COF analysis

“Consequence of failure” values vary based on a utility’s ability in respond to, isolate and repair failures. Xylem uses a utility’s existing data, along with extensive proprietary asset databases to estimate response, shutdown and repair time. By simulating best- and worst-case scenarios, a dynamic range of COF values are calculated to help the utility understand which isolation valves need assessment first for their critical assets.

When a range of COF values are calculated, an already established quantitative approach to risk takes on an additional dimension that is unique to each asset. A utility can use the range of values to further separate critical assets from each other and prioritize projects with the greatest ROI, and reduce risk at the same time.

A side-benefit of quantitative risk models is that, as more data is collected or repairs are conducted, updating the risk model does not require tedious re-scoring of subjective factors. This is important to help utility management track the improvement in risk mitigation strategies over time and justify funding for future programs.

Graham Bell and Peter Martin are with Xylem Inc. This article appears in ES&E Magazine’s October 2019 issue.


Please enter your comment!
Please enter your name here