653. An MDP Design approach to redundancy allocation with dynamic maintenance
Invited abstract in session TC-54: Methods & stochastic modelling , stream Stochastic modelling.
Tuesday, 12:30-14:00Room: Liberty 1.08
Authors (first author is the speaker)
| 1. | Luke Fairley
|
| STOR-i, Lancaster University | |
| 2. | Rob Shone
|
| Management Science, Lancaster University | |
| 3. | Peter Jacko
|
| Ma, Lancaster University | |
| 4. | Jefferson Huang
|
| Operations Research Department, Naval Postgraduate School |
Abstract
A system can be made to be more reliable by adding redundant copies of a component, where these copies serve as back-ups for cases where other components have failed. The inclusion of more copies decreases the probability of all copies failing at the same time, improving reliability. The problem of deciding how much redundancy to add is called the Redundancy Allocation Problem, which is a combinatorial optimisation problem. Previous work has shown that a repairable redundant system can be modelled as a Markov process by assuming that degradation and repair times are exponentially distributed, and assuming repairs are always started straight away. A natural question arises as to whether or not this second assumption can be relaxed. For example, with three levels of redundancy and only one failure, it need not be the case that this component is repaired straight away. The problem of deciding when to repair can be modelled as a Markov decision process (MDP). As it stands, these two problems are separate: we have one problem for designing a system, and another for determining repair policies. Our work focuses on the integration of these two problems into an MDP Design problem, where we wish to simultaneously design a system and determine the dynamic policy used for its operation. This talk showcases such a model, demonstrates the extent to which instances can be solved exactly using off-the-shelf solvers, and looks towards some more scalable solution methodologies.
Keywords
- Stochastic Optimization
- Reliability
- Combinatorial Optimization
Status: accepted
Back to the list of papers