Contacts
Chris Steele, Head of Digital Products and Services | Head of Data Science and Analytics SteeleC@binnies.com +44(0)1737 856543

Asset survival modelling

Quantifying the balance between reactive and proactive maintenance

PAM Analytics, in partnership with Binnies and utilising the ECO-X platform, helps asset-rich companies to change their asset management policies from reactive fail and fix to proactive predict and prevent by using predictive analysis and simulation to model, simulate and optimise their asset management policies.

This is done at the individual asset level and at the operational, tactical and strategic levels. The outputs are asset management policies that optimise the trade-off between the risk costs of asset failure and replacement, and the direct costs of asset maintenance.

Key features

The HELIX system can generate real-time visualisations of asset health, unit costs and key performance indicators across the targeted assets. These data-driven, real-time dynamic intelligence dashboards enable understanding of system reliability, asset performance and cost characteristics from facility/system level down to asset level.

HELIX utilises PAM Analytics to provide tactical-level risk management-through dynamic asset deterioration curves enabling:

  • Predictive analytics rather than business intelligence
  • Dynamic risk of asset-failure modelling
  • Asset to portfolio management.

The HELIX system can simulate the financial implications of a range of asset maintenance and replacement policies to determine the optimal refurbishment and replacement strategies policy, subject to a range of operational constraints, for example, the organisation’s maintenance capacity and attitude to the risk of asset failure.

Proof-of-concept (PoC) method

All models will begin with a PoC model to understand the data, parameters, asset behaviour and ability to model that behaviour. For a typical PoC, one or two asset types of the same class, for example, wet well submersible pumps, are selected. These will display typical behaviours or specific behaviours that need to be understood. Data are prepared for the selected classes.

The team goes through the data very carefully to gain a deep and thorough understanding of them. This is the first part of the exploratory data analysis stage and usually results in many questions about the data. This phase may take a couple of iterations to understand the data fully, but its importance cannot be underestimated.

Using the understanding that has been developed and the desired outcomes, the data are prepared for the analysis and modelling in the time-to-failure transformations module.

The key performance indicator, asset deterioration curve and predicted maintenance intervention modules are utilised to gain insights and understanding into asset performance and failure.

Full-scale-method case study

Once a PoC is complete, the model developed can be reused on individual assets or an entire asset to provide strategic asset management direction. The following is an anonymised real example of asset survival modelling in which 1000 pumps in sewage pumping stations were simulated. The simulation model combined the Cox proportional hazards model (the model at the centre of predictive asset management) and discrete event simulation.

The model was developed using 12 years of data. The simulations show the results for the following five years sampled at monthly intervals and are, thus, the aggregation of results over 60 months. The state of the pumps in any month is dependent on their state in previous months.

Risk tolerance based on a five-point scale ranging from risk averse to risk tolerant was applied to the model. Risk tolerance is adjustable but is typically based on the maximum acceptable level of repeated pump, for example, failure and is defined as the number of consecutive monthly failures a pump can have before it is replaced.

Costs were set for new pumps and for annual proactive and reactive maintenance.

The model demonstrated that the total cost of maintenance policies that are only reactive is much greater than the total cost of maintenance policies that have just a small proactive element. The total cost decreases as the proportion of proactive maintenance increases for all risk tolerances, except at high maintenance capacities and very risk tolerant policies; the extent of the decrease depends on the maintenance capacity and risk tolerance.

These conclusions are unlikely to be news to many, but the model quantifies this benefit by providing a clear set of points where regular proactive maintenance reduces the cost of reactive maintenance and capital expenditure. The model does this taking into account various risk tolerances, thereby enabling asset managers to make informed decisions and provide justification and demonstrate the probable consequence of taking those decisions.

In this example, the following conclusions were reached and quantified:

  • If the maintenance capacity of the organisation is very low, the optimal pump management policy could not be achieved.
  • If the maintenance capacity of the organisation is very high, the risk tolerance is very high, so the maintenance policy is almost exclusively proactive.
  • The total cost would be greater than is optimal.
  • The optimal pump management policy does not require all the maintenance capacity; too much maintenance is being carried out and unnecessary costs are being incurred.
  • For all other maintenance capacities, the optimal pump management policy is achieved at high levels of proactive maintenance. The reduction in the total cost becomes progressively smaller as the level of proactive maintenance increases. When other costs such as lower consequence costs associated with high levels of proactive maintenance are included, the curves become steeper and then flatten out, thereby making the optimal point or region clearer.