This PhD focuses on the optimization of the logistic of predictive maintenance.
Device failures is one of the main sources of cost in the industry. It is notably the case for airlines, because if a failure occurs, flights must be canceled, which leads to costs that are often much higher than the price of replacing this equipment. The objective of predictive maintenance is to exploit available data to predict when the next failure will occur. Based on this information, the airline must schedule the next maintenance operations that are going to be done.
This problem can be formulated as a Partially Observed Markov Decision Process (POMDP). We study some methods to solve a POMDP with a Linear Program using probablistic graphical models.
Cohen, V., Parmentier, A. (2018). Linear Programming for Decision Processes with Partial Information . arXiv preprint arXiv:1811.08880.
Generalization of Markov Blanket to understand dependences in Graphical Models, JFRB , May 2018, Toulouse
Linear Programming for Influence Diagrams, ISMP , July 2018, Bordeaux
Linear Programming for Influence Diagrams, Conference on Discrete Optimization and Machine Learning , July 2018, Tokyo
Linear Programming for Influence Diagrams, PGMO days , November 2018, Saclay
CERMICS, École Nationale des Ponts et Chaussées
6 et 8, av. Blaise Pascal
Cité Descartes - Champs-sur-Marne
77455 Marne-la-Vallée cedex 2