AMORE-MIO - Advanced MethOds for Reliability and maintenance Evaluation : Models, Inference, Optimization
Two research teams are involved in this project ; they have complementary skills in reliability theory, maintenance modeling, system monitoring and control, and optimization.
- LJK : ASAR (Applied Statistics and Reliability) team bringing its competencies in stochastic modelling and statistical inference in reliability and maintenance and in joint modelling of ageing and maintenance efficiency of complex systems.
- GIPSA-lab : SAFE (Safe controlled and monitored systems) team (formerly SAIGA team) with competencies in deterioration and predictive maintenance modelling, prognosis and remaining useful life estimation, system monitoring and diagnosis, post-prognosis decision making and control systems.
Two industrial partners EDF R&D and GRT Gaz support scientifically this project and will participate to the working meetings: they will bring some of their challenging open issues as motivating examples for the theoretical works developed in the project, and they will be potential end-users of the developed methods. They are also potentially interested by the possibility of developing more specific contractual collaborations in order to applied the developed techniques on specific data sets.
AMORE-MIO project team is built in the continuation of the PERSYVAL-Lab exploratory project AMORE (2016-2018). AMORE-MIO objective is to build a platform to structure and further strengthen the joint work between LJK/ASAR and GIPSA-lab/SAFE teams and to increase our visibility and potential for collaboration with different external partners from industry and foreign universities.
Context and background
For complex industrial systems, an issue consists in maintaining equipment in working order conditions in accordance with safety, availability and cost constraints. Nowadays, the wide access to information and data on systems opens new research perspectives to this classical reliability and maintenance issue. Those information and data can be very heterogeneous. They can arise from reliability data base and describe the successive failures, maintenances and inspections of the different parts of the considered systems. Practically, those data base can simultaneously consider several, and possibly correlated, failure mechanisms. Maintenance actions can have different efficiencies and inspections can also be more or less extensive. Information can also be collected thanks to the presence of sensors or modern monitoring technics and describe the specific working conditions of each system, such as for example humidity or temperature. Those sensors can also record the severity of the usage conditions of each piece of equipement and even measure deterioration levels and health indicators. AMORE-MIO aims at exploring solutions and developing a comprehensive approach to optimally manage the health state of deteriorating systems based on this diversity of available information, and resorting to a wide range of possible actions, from optimal control of the operating conditions to maintenance actions.
The project is structured around the three following axes:
- to propose complex models that can consider simultaneously all this heterogeneous information,
- to develop the corresponding inference methods in order to assess from data the different model parameters,
- to implement those models and to develop decision-making approaches in order to both optimize the mainte- nance and control the working conditions of the different systems.
It is important to highlight that "black box" machine learning algorithms seem not to be perfectly relevant in AMORE-MIO application context. Indeed, if the interpretability of the models is not necessary in various application fields, they become significant arguments when trackability, auditability, transparency or physical justification are required, as in reliability and safety context. AMORE-MIO aims at developping "white box" fully intelligible models and decision-making methods.