Keywords: Prescriptive maintenance, Bayesian networks, Dynamic bayesian networks, Continuous degradation
Maintenance decision-making is of prime importance for improving the global performance of industrial systems and structures during their useful life. Maintenance and replacement models use different methods starting from static analysis like Fault Tree analysis, random variables for modeling the failure time, or employ stochastic processes for depicting the degradation behavior of the system. Today, due to improvements in industrial methods and also to fulfill the missions that are designed for, the systems are more and more complicated.
Therefore, more complex mathematical and statistical methods are needed to model the behavior of such systems. Recently, Probabilistic graphical models have been proposed to model the complexity of industrial systems and are widely used in dependability, risk analysis, and maintenance applications. Bayesian Networks (BN) and their temporal version, i.e.; Dynamic Bayesian Networks (DBN) which is suitable to capture the time dependency structure of system condition, are members of this family.
The primary interest of this project is to find a way to merge continuous degradation and DBN to benefit from more straightforward and more realistic modeling of the degradation along with the interaction between the system components.
The key questions, which arise in this project, are:
1. How to develop models for a complex system when the continuous degradation of some components can be measured directly and we already have for these components models of gradual degradation and optimized maintenance strategies?
2. How to develop global maintenance strategies for a system of the previous type and how to integrate the proposed strategies to the available model to analyze their performance?
3. How to model the quality of the available monitoring information and evaluate its impact on the performance of a smart maintenance strategy, which is less robust to monitoring faults than a systematic strategy that does not exploit the monitoring information.
4. How to evaluate the reliability or operational availability of a complex system with dependencies between its components?
5. How to deal with the problem of modeling imperfect repair and its effects on a complex system subject to a prescriptive maintenance policy?
The objective of the research topic is to address all these questions by focusing primarily on the first three points which constitute a prerequisite for the rest.
Candidate skills:
- A professionally relevant background in probability, statistics, Markov models, etc.
- Strong implementation and programming knowledge ( R , Python, …)
- Good written and oral English language skills
In addition, familiarity with Bayesian Networks and dynamic Bayesian networks is appreciated.
To apply, please send a CV, a motivation letter, and your master’s transcripts by April 26, 2023.