Ph. D. Project
Title:
Failure prognosis and decision-making for predictive maintenance of a hydrogen power using Artificial Intelligence
Dates:
2022/10/14 - 2025/10/15
Student:
Supervisor(s):
Other supervisor(s):
Dr. VU Hai-Canh (hai-canh.vu@utc.fr)
, Bonrnand Bastien (bastien.bornand@eo.dev)
Description:
In the context of promoting renewable energy, the GEH2 (zero-emission hydrogen power generator) developed by EODev is now the most compact and efficient hydrogen generator on the market in terms of power output. To ensure its proper performance/service and maximize the energy produced, various maintenance operations, including periodic preventive maintenance, have been planned. However, this inadequate maintenance planning can lead to significant disadvantages (e.g., unnecessary maintenance operations, high failure frequency, random/reduced performance, additional maintenance/operating costs). In this context, an evolution towards "just in time" predictive maintenance is therefore important for EODev to deploy only the right maintenance actions according to the real state of the GEH2 system and its equipment. The deployment of this new maintenance strategy is mainly based on the two following processes: (1)- failure prediction of the GEH2 components (fuel cell, cooling system, battery, control system, power conversion system) and (2)- optimization of predictive maintenance decision-making of the components.
To answer these industrial challenges, the first objective of this thesis is to develop prognostic approaches to predict the state of health (degradation level) of critical components of the GEH2. Indeed, in the literature, several prognostic approaches have been proposed for predicting the failure time of an equipment/system. However, the existing approaches are mainly proposed for production systems or machine tools. They are not directly applicable to specific equipment, such as the GEH2 system recently developed by EODev. In addition, the knowledge and data about the degradation/failure process of the GEH2 components are not yet mature. And finally, a GEH2 can be considered as a complex system composed of the different sub-systems/components with strong dependencies. Given the complexity of the GEH2 system, the various types of data, the different nature of the components, as well as the dependencies between them, the prognosis of the health state of the GEH2 (system level) and its components (component level) is, therefore, a complex problem to solve.
The second objective of this thesis is to propose/develop a predictive maintenance strategy adapted to the GEH2 based on the prognosis results obtained from the previous step while respecting the properties of the machine, its usage profile and constraints related to both operation and logistical support (e.g., availability of spare parts, maintenance budget, limited maintenance duration). A multi-criteria optimization (e.g., minimizing maintenance costs and/or maximizing GEH2 performance) with constraints will be developed, based on conventional optimization algorithms or/and reinforcement learning algorithms to find the optimal maintenance schedule. In the literature, several decision support models (maintenance strategies and optimization algorithms) have been developed. However, the development and implementation of adequate solutions on an industrial scale still lack foundations, methods, and tools.
The major scientific originalities of this thesis are, therefore:
The development of a methodology to identify the key/critical components of a GEH2 system according to a specific usage profile.
The construction of advanced prognostic algorithms based on Machine learning/Deep learning to predict the degradation evolution and/or remaining useful life of the critical components of the GEH2. The developed algorithms must allow for the exploitation of the available data and the domain knowledge.
Development of hybrid prognostic algorithms to predict the health state and remaining useful life of the GEH2 (system level prognosis).
The development of a decision support model for the predictive maintenance of the GEH2 based on the prognostic results.
A multi-criteria optimization approach provides an adapted maintenance schedule based on the developed decision support model while respecting the properties and usage profiles of the GEH2 and the constraints related to the operation and the logistic support.
The development of a model to evaluate the effectiveness of the predictive maintenance actions implemented.
Implementation in a commercial software solution.
To answer these industrial challenges, the first objective of this thesis is to develop prognostic approaches to predict the state of health (degradation level) of critical components of the GEH2. Indeed, in the literature, several prognostic approaches have been proposed for predicting the failure time of an equipment/system. However, the existing approaches are mainly proposed for production systems or machine tools. They are not directly applicable to specific equipment, such as the GEH2 system recently developed by EODev. In addition, the knowledge and data about the degradation/failure process of the GEH2 components are not yet mature. And finally, a GEH2 can be considered as a complex system composed of the different sub-systems/components with strong dependencies. Given the complexity of the GEH2 system, the various types of data, the different nature of the components, as well as the dependencies between them, the prognosis of the health state of the GEH2 (system level) and its components (component level) is, therefore, a complex problem to solve.
The second objective of this thesis is to propose/develop a predictive maintenance strategy adapted to the GEH2 based on the prognosis results obtained from the previous step while respecting the properties of the machine, its usage profile and constraints related to both operation and logistical support (e.g., availability of spare parts, maintenance budget, limited maintenance duration). A multi-criteria optimization (e.g., minimizing maintenance costs and/or maximizing GEH2 performance) with constraints will be developed, based on conventional optimization algorithms or/and reinforcement learning algorithms to find the optimal maintenance schedule. In the literature, several decision support models (maintenance strategies and optimization algorithms) have been developed. However, the development and implementation of adequate solutions on an industrial scale still lack foundations, methods, and tools.
The major scientific originalities of this thesis are, therefore:
The development of a methodology to identify the key/critical components of a GEH2 system according to a specific usage profile.
The construction of advanced prognostic algorithms based on Machine learning/Deep learning to predict the degradation evolution and/or remaining useful life of the critical components of the GEH2. The developed algorithms must allow for the exploitation of the available data and the domain knowledge.
Development of hybrid prognostic algorithms to predict the health state and remaining useful life of the GEH2 (system level prognosis).
The development of a decision support model for the predictive maintenance of the GEH2 based on the prognostic results.
A multi-criteria optimization approach provides an adapted maintenance schedule based on the developed decision support model while respecting the properties and usage profiles of the GEH2 and the constraints related to the operation and the logistic support.
The development of a model to evaluate the effectiveness of the predictive maintenance actions implemented.
Implementation in a commercial software solution.
Department(s):
| Modeling and Control of Industrial Systems |
