Ph. D. Project
AI-based proactive maintenance decision-marking for industry 4.0
2020/11/01 - 2023/10/31
This PhD proposal is part of the European project H2020-ICT AI-PROFICIENT (Artificial Intelligence for improved PROduction efFICIEncy, quality and maintenance) which aims to develop AI-based proactive control strategies to improve the manufacturing process (production efficiency, quality and maintenance) while keeping the human in central position, assuming supervisory (human-on-the-loop) and executive (human-in-command) roles.
In that way, the global objective of this PhD is to develop AI-based proactive maintenance decision-making approaches for manufacturing systems in the context of industry 4.0. The proposed approaches will be applied to the pilot sites of the project, e.g. the Continental - Sarreguemines plant (France) and INEOS - Gladbeck plant & Geel plant (Belgium), in order to validate their feasibility and benefits. As active member of the CRAN team, the PhD student will be really involved in the European project by participating to project development, meetings, writing deliverables, presenting progress and results ...

Research topic: scientific context and goals
In manufacturing system, maintenance plays a key role to sustain the system within its nominal operation space mainly by anticipating failures (e.g. predictive maintenance). Indeed maintenance optimization aims to control optimally, and at lowest possible maintenance costs, Key Performance Indicator (KPIs) of the system both related to conventional performances (e.g. productivity, reliability), and emerging ones (e.g. sustainability). In the context of industry 4.0, manufacturing systems are facing to, not only the new ways of producing goods and services, but also the activities, methods and tools used to maintain industrial equipment and facilities for promoting required agility and resilience. Particularly, with the development of Cyber-Physical Systems (CPS)/Cyber-physical-Production Systems (CPPS) concept, it is appearing massive digital transformation providing huge data related to the health state of CPS/CPPS components, the operational conditions of the system, ... which can help, thanks to advanced AI techniques (e.g., machine learning), to better detect abnormal behaviours of equipment (diagnosis), predict future failure modes (prognosis) and support, by advance, maintenance decisions (proactive decision-making). However, although a lot advanced AI techniques are now available, their application in prognostics and maintenance decision-making is still limited and remains widely open.
To face this issue, the aim of this Ph. D. is to develop AI-based proactive maintenance decision-making ** ? approaches for manufacturing system in the context of industry 4.0. The work developed will provide an effective and efficient solution to the current challenge in predictive maintenance on the use of AI techniques to optimize maintenance actions in dynamic and proactive way (to do the right action, at the right place and in just in time). The PhD program is structured in three major phases:
Phase 1 - Prognostics of KPI at the system (machine) level: The aim is to develop AI-based approaches allowing to predict the KPIs of the system taking into account not only the prognostic results related to KPI (predictive reliability/RUL-remaining useful life) at component level (in base of embedded prognostic algorithms within CPS components), but also the dependence relationships between components under specific context associated to the systems missions/functions. To support this objective, several dependencies in terms of kinds of interactions between components (e.g., structural/functional, stochastic, informational dependence) should be first modelled and formulated (*** sous quell formalisme?). This helps to quantify the impact of one component/group of components on the KPI of other components. Secondly, AI-based approaches (e.g., recurrent neural network) will be investigated to predict the KPIs at the system level from the components' KPI predicted at CPS component level and the formalised dependences between components. The prognostics results of KPIs will be used for predictive maintenance decision-making at the second phase.
Phase 2 - Development of AI-based maintenance decision-making models: The proposed models will be built on a set of appropriate decision rules and advanced AI algorithms (e.g. reinforcement learning) which allow to learn the most relevant decision rules to deal with the current condition of the system and its environment from observed data, the estimated KPI at both components and system level. In that way, the proposed AI-based decision models should enable not only providing optimal maintenance planning taking into account both the requirements associated with the maintained system and its support one (e.g. spare parts, maintenance skill) but also to be able to update efficiently the maintenance planning in a dynamic context (e.g. structure changes on main or support system, occurrence of new maintenance opportunities).
Phase 3 - Validation of the proposed models/approaches on the pilot case studies: The models/approaches proposed in phases 1 and 2 will be applied to the two industrial case studies of the project (Continental - Sarreguemines plant (France) and INEOS - Gladbeck plant & Geel plant (Belgium)). Several performance metrics will be selected to evaluate the performance of the proposed models/approaches. A study on adjustment/validation of the proposed models/ approaches should be also investigated. This phase will allow to underline, from the assessments and adjustments done, the advantages and disadvantages of such approaches for supporting proactive maintenance decision-marking.
predictive maintenance, prognostic, AI, optimization
Modeling and Control of Industrial Systems