ALSTOM (Power and Transport) needs to commit to stringent system availability requirements, especially in the context of service-level agreements. A non-adherence to the performance levels often leads to penalties. This requires modeling the system by including failures and repairs but also the support logistics such as maintenance resources and spares. Due to the complexity of the model, usually involving large state spaces and in most of the cases non-Markovian (i.e., not represented by a Markov chain, which is more or less required to get a numerical solution), the only evaluation tool at hand is Monte Carlo simulation. While this method is directly applicable, the fact that the system unavailability is (hopefully) rare means that the simulation time to get a sufficient relative accuracy can be extremely long. Specific techniques, such as Importance Sampling, have been developed to deal with this issue and to obtain an accurate estimation in a reasonable time. From one side, Importance Sampling can lead to very efficient tools, but on the other side, there is no general procedure to follow to automatically produce an Importance Sampling technique. Each problem must be analyzed to find out if there is a way to exploit the idea on that specific case. For the types of models that must be studied, finding appropriate Importance Sampling methods is largely an open problem.

The goal of the PhD thesis is then manyfold:

To model the problems encountered by ALSTOM Power and ALSTOM Transport.
To design valid Importance Sampling methods for these problems. While Importance
Sampling has been extensively studied in the Markovian framework for systems involving failures and repairs, a limited amount of work exists in the non-Markovian case and when logistics is involved.
To develop the corresponding tools: the ultimate goal is to have an implementation of the designed methods that can be used by ALSTOM engineers.

The PhD thesis will be supervised by ALSTOM Transport (Paris) and Inria (Rennes). The
candidate is expected to spend some time on both sites.

Bibliography:

P. Dersin and R. C. Valenzuela. Application of Non-Markovian Stochastic Petri Nets
to the Modeling of Rail System Maintenance and Availability. In the Proceedings of
the 2012 Winter Simulation Conference, Berlin, Germany, December 2012.
G. Rubino and B. Tuffin, editors. Rare Event Simulation using Monte Carlo Methods.
John Wiley & Sons, 2009.
S. Asmussen and P. W. Glynn. Stochastic Simulation. Springer-Verlag, New York,
2007.
P. L’Ecuyer and B. Tuffin. Approximating zero-variance importance sampling in a
reliability setting. Annals of Operations Research, 189:277–297, 2011

PhD Thesis Opportunity with Inria and ALSTOM: Availability Estimation by Simulation for Systems including Logistics

Contacts: Bruno Tuffin & Pierre Dersin Tél. direct : +33 (0)2 99 84 74 94, +33 ( 0)1 5706 1595 Email: bruno.tuffin@inria.fr, pierre.dersin@transport.alstom.com

Bibliography:

Contacts: Bruno Tuffin & Pierre Dersin
Tél. direct : +33 (0)2 99 84 74 94, +33 ( 0)1 5706 1595
Email: bruno.tuffin@inria.fr, pierre.dersin@transport.alstom.com