At a glance:
- Accurately modelling the degradation of assets is essential for making predictions about their future condition, for forming an efficient plan for their maintenance, and for making infrastructure investment decisions.
- There are a number of degradation modelling approaches that can help asset managers ensure the continued utility and fitness for purpose of their assets.
- Here, we examine various asset degradation modelling methods, and explore why and when asset managers should look to re-evaluate their approach.
From transport links to utility networks, the UK economy relies on the mechanical and geotechnical assets (such as construction machinery or transportation embankments) that form the key linchpins of our national infrastructure. Modelling the degradation of these assets is essential for making predictions about their future condition, forming a robust and efficient plan for their maintenance, and for making infrastructure investment decisions.
Predicting the degradation of such assets requires an understanding of the impact of weather, which is an increasingly complex task as the climate changes. Fortunately, asset managers now have access to increasing volumes of data being collected via intelligent infrastructure schemes, opening many opportunities to improve degradation models with the implementation of more sophisticated modelling techniques.
As with all models that form the foundation of business planning and decision-making activities, the periodic review and improvement of degradation models should be standard practice for asset managers looking to ensure their continued utility and fitness for purpose. But what are the broad approaches to asset degradation modelling, and how and when should these approaches be re-evaluated?
How can I choose the right approach to asset degradation modelling?
Selecting the appropriate degradation modelling approach requires an understanding of the business needs for asset condition predictions, and an appreciation of the relative benefits of a wide range of possible modelling techniques. At a high level, there are three broad approaches to modelling degradation:
If the physical mechanisms for degradation (weather conditions, for example) are well understood, then a deterministic approach may be suitable. Deterministic approaches make estimates of the future condition of an asset by taking into account factors like the asset’s age. However, they provide no information about uncertainty in the estimate.
One advantage of this approach is that models can be easily interpreted, which is often useful for gaining stakeholder buy-in for maintenance or investment plans. The downside is that in cases where the mechanisms of degradation are complex, not fully understood, and/or are influenced by random events, the degradation predictions may not be complete or useful.
Example: Next year the bridge will be in a fair condition because it is 20 years old
Probabilistic models can be designed to output information about an asset’s likely future condition, providing important information about any uncertainty relating to maintenance and investment planning activities. Knowledge of uncertainty within degradation estimates can improve the robustness of maintenance and investment plans and reduce the expenditure associated with any deviation.
A common probabilistic degradation model is a Markov chain model, in which an asset moves between states representing different degradation conditions. Assuming the age and maintenance history of assets are accounted for, there must also be sufficient data relating to asset ages and maintenance activities to accurately calibrate the model.
This has clear advantages over deterministic approaches, particularly when there is access to sufficient asset history data.
Example: Next year the bridge has a 70% chance of being in a fair condition and a 30% chance of being in a poor condition. This reflects the observed outcomes of a cohort of other bridges – varied in terms of age and location – against which we can compare our bridge.
Where standard and probabilistic models have failed to inspire confidence, hybrid models can offer a solution.
Hybrid models are a combination of deterministic and probabilistic approaches that function by incorporating knowledge of degradation mechanisms into probabilistic models.
These models lessen the requirement for comprehensive data inputs and make the model more intelligible to stakeholders. While these are mathematically challenging to construct and are likely to require bespoke software implementations, they are well worth the investment.
Investment in hybrid models is particularly justified in cases where degradation predictions can have a significant detrimental effect on safety, reputation or finances, and where standard and probabilistic models have failed to inspire confidence.
Example: Next year the bridge has a 65% chance of being in a fair condition and a 35% chance of being in a poor state. This reflects the observed outcomes of a cohort of other bridges – varied in terms of age and location – against which we can compare our bridge, while also incorporating the understanding that previously unobserved mechanisms of degradation (extreme wind events, for example), will accelerate the degradation process.
Why should I re-evaluate my approach to asset degradation modelling?
As the foundation for asset management planning, poorly performing degradation models can cause significant budgeting and resource allocation problems downstream. A model review can uncover the need for improvements to the modelling approach, or it can highlight the requirement for more flexibility in downstream processes if performance cannot be improved with available data. But when should a re-evaluation take place?
When confidence has been lost in the existing approach
Confidence can be lost in the existing degradation modelling approach for a number of reasons, including the historical or recent poor performance of predictions.
Alternatively, a model may be performing reasonably well, but there are concerns that its assumptions will become invalid as the physical, regulatory, or business environment changes, or as the application of its outputs evolve. For example, consider a case in which a model’s calibrated parameters depend on climate, and in particular, on the large-scale patterns of climate change. If certain parameters are left static, the model may slowly diverge from observations, causing issues for the longer-term forecasting and planning activities that use the model’s outputs.
Poorly performing degradation models can cause significant budgeting and resource allocation problems downstream.
When new datasets become available
Rudimentary degradation models can be constructed based on nothing more than information about failure times. However, additional degradation information has the potential to considerably improve the model outputs and should be incorporated wherever possible. For example, LIDAR (Light Detection and Ranging) sensors have the potential to exponentially increase the available degradation data for some asset classes, but will likely require changes to existing degradation modelling approaches.
In some cases, new datasets will be noisy, but this should not deter asset managers from investigating their utility to degradation modelling. Stochastic filtering techniques, used in combination with knowledge of degradation mechanisms, can enable the prediction of asset condition from partial and noisy observations.
When ‘state-of-the-art’ has moved on
The research and application of degradation modelling techniques is continually evolving. If several years have passed since a modelling approach was established, it is likely that there are new advances to consider. This was the motivation for the Smith Institute’s review of Network Rail’s earthwork degradation methodology, which resulted in the identification of several quick wins to the existing approach.
In the time that has passed since a methodology was established, it is also likely that the amount of available degradation data has increased. With increased volumes of degradation data, a wider range of techniques may be exploited. For example, explainable machine learning algorithms, based on decision tree methods, can be useful tools for tackling infrastructure deterioration modelling where there is plenty of degradation data.
How can I make the most use of asset degradation predictions?
As the climate changes and extreme weather events increase in their frequency and severity, understanding the condition of our key national infrastructure and the likelihood of their degradation is more important than ever.
Fortunately, new datasets and modelling techniques are becoming available to tackle these challenges. To make the best use of the remarkable tools available, asset managers must work hand-in-hand with mathematical experts to select approaches that best meet the growing complexity of the challenge.
If you would like to learn more, or would like to discuss how the Smith Institute can support you in predicting asset degradation, please get in touch: email@example.com