System reliability simulations in asset performance management (APM)
Reliability, availability, and modeling (RAM) analysis is a methodology used to assess and predict performance over the life cycle of an asset for a given system design taking into consideration factors affecting production, operations, and maintenance. RAM models are very useful for asset managers because they allow stochastic evaluations of the likelihood of an event (failure) over time to be combined into an overall picture of system performance that can be represented as a series of cash flows. This takes the esoteric knowledge of asset performance and degradation mechanisms and translates it into the language of finance through life cycle costs associated with alternative scenarios.
System reliability models are constructed using historical data against different known risks on the assets. In the following example, we create a model of a system which collects water from a source via a pump and then runs the water through a production process:
The first step is to identify the key risks and associated mitigating actions in the system. In the example scenario, the identified risks for the pump are failure to the impeller, to the seals, and to the bearings. The mitigating actions considered are time-based seal replacement, periodic vibration analysis, and a time-based impeller redesign. Model parameters for reliability distributions and associated costs are estimated from historical data within APM.
The model outcomes are generated through Monte Carlo simulations over a defined period and incorporates inherent uncertainty in the system model. The chart below shows the results of a Monte Carlo simulation for a period of five years. The results show that while the total costs in the mitigating actions scenario are much less than the no-actions scenario, the cost due to production losses is higher. The higher production losses are due to the system shutdown required to execute the mitigating actions.
Monte Carlo simulations are repeated computational experiments where uncertain model factors are represented by a range of a possible values through a probability distribution. Model outputs are simulated through random sampling of the different distributions and the final results account for the corresponding possible outcome values.