Ph. D. Project
Dates:
2023/10/01 - 2026/09/30
Student:
Supervisor(s):
Description:
Scientific context and goals:
Since recent years, prognostics of the remaining useful life (RUL) of manufacturing systems is a research area of growing scientific interest because it plays a crucial role in Prognostics and Health Management (PHM) community [1]. Indeed, accurate and confident prediction of the RUL are very meaningful and important for maintaining system operational efficiency by taking the right action (maintenance and/or control), at the right time [2,3]. In that way, by integrating RUL information as input of the decision-making process, scheduling of production and/or maintenance, and logistic supports are being moved from reactive form to proactive one based on failure anticipation. In the literature, a large number of prognostics approaches have been proposed and successfully applied in different applications [4,5,6] with a nowadays trend leading to data driven approaches.
In these approaches, a major issue impacting the performance of the prognostics models and, by the way, their usage, is the quality of data provided by the monitoring systems. It is shown that existence of sensor degradation may significantly influence data uncertainty and therefore, the effectiveness of system health prognostics [11]. In this way, measurement noise has been widely investigated in the literature with respect to statistical errors with constant or time-varying variance [7,8]. Nevertheless, in real applications the degradation state of the systems "it-self" may affect the monitoring system. Such relation has not been considered yet. For instance, in practice, due to the varying operating environment and cumulative damage to the system, the embedded sensors (i.e., monitoring system) may suffer such damage leading to consider interaction between the equipment and embedded sensors performances despite sensors usually deteriorates over a long period of operating time [9,10].
To face the above challenges, the objective of this PhD project is to develop appropriate prognostics approaches for failure/RUL prediction of manufacturing system considering measurement errors caused by sensors degradation. More precisely, the purpose of the PhD will address the three scientific issues:
- How to model the degradation processes of the system and the embedded sensors system considering both operating condition and the state interaction between the system and the embedded sensors?
- How to model the degradation impact of the embedded sensors on the measurement errors?
- What are the appropriate prognostics approaches for failure/RUL prediction of the system from data uncertainty caused by degradation sensor?
The above scientific issues are identified as main challenges and issues with the PHM (Prongostic and Health Management) communities such as PHM IEEE, PHM society and IFAC (TC5.1., TC6.4) ou CIRP).
To face with these scientific issues, the work developed in this PhD will structured into three major phases: (1)-modelling of degradation processes for both manufacturing system and its embedded sensors considering not only the operating condition but also the degradation interaction/dependencies between the system and its sensors; (2)- development of observation models for measurement errors considering the degradation states of embedded sensors; (3)- based on the degradation and measurement errors models, the development of appropriate data-driven approaches for failure/RUL prediction of the manufacturing system.
These contributions provide answers to the scientific issues referring to the initial objective targeted:
- Modelling and qualification of the degradation interaction impacts between the manufacturing system and its sensors
- Development of degradation models for both system and its sensors considering the degradation interaction impacts and operating condition
- Development of measurement errors models (linear or non- linear) considering the degradation impact of sensors
- Elaboration of prognostics algorithms using particle filters or/and deep learning for failure/RUL prediction of the system from data uncertainty caused by degradation sensors
These contributions to these scientific issues are supported by different scientific tools, which are mainly mathematical tools (e.g., copula models, mixing distributions, stochastic process) and data-driven prognostic approaches (particle filters, deep learning ...).
These different modeling and development proposals will be implemented on Matlab software. Several performance metrics will be selected to evaluate the performance of the proposed models/approaches. Numerical examples will be used to validate these proposals (at least in the laboratory) while also trying to extend this validation step by using real data from industrial use cases (ex., Continental, INEOS, SEW USOCOME dans le cadre de projets européens comme AI-PROFICIENT (https://ai-proficient.eu/) et MODAPTO (2023-2025).
Since recent years, prognostics of the remaining useful life (RUL) of manufacturing systems is a research area of growing scientific interest because it plays a crucial role in Prognostics and Health Management (PHM) community [1]. Indeed, accurate and confident prediction of the RUL are very meaningful and important for maintaining system operational efficiency by taking the right action (maintenance and/or control), at the right time [2,3]. In that way, by integrating RUL information as input of the decision-making process, scheduling of production and/or maintenance, and logistic supports are being moved from reactive form to proactive one based on failure anticipation. In the literature, a large number of prognostics approaches have been proposed and successfully applied in different applications [4,5,6] with a nowadays trend leading to data driven approaches.
In these approaches, a major issue impacting the performance of the prognostics models and, by the way, their usage, is the quality of data provided by the monitoring systems. It is shown that existence of sensor degradation may significantly influence data uncertainty and therefore, the effectiveness of system health prognostics [11]. In this way, measurement noise has been widely investigated in the literature with respect to statistical errors with constant or time-varying variance [7,8]. Nevertheless, in real applications the degradation state of the systems "it-self" may affect the monitoring system. Such relation has not been considered yet. For instance, in practice, due to the varying operating environment and cumulative damage to the system, the embedded sensors (i.e., monitoring system) may suffer such damage leading to consider interaction between the equipment and embedded sensors performances despite sensors usually deteriorates over a long period of operating time [9,10].
To face the above challenges, the objective of this PhD project is to develop appropriate prognostics approaches for failure/RUL prediction of manufacturing system considering measurement errors caused by sensors degradation. More precisely, the purpose of the PhD will address the three scientific issues:
- How to model the degradation processes of the system and the embedded sensors system considering both operating condition and the state interaction between the system and the embedded sensors?
- How to model the degradation impact of the embedded sensors on the measurement errors?
- What are the appropriate prognostics approaches for failure/RUL prediction of the system from data uncertainty caused by degradation sensor?
The above scientific issues are identified as main challenges and issues with the PHM (Prongostic and Health Management) communities such as PHM IEEE, PHM society and IFAC (TC5.1., TC6.4) ou CIRP).
To face with these scientific issues, the work developed in this PhD will structured into three major phases: (1)-modelling of degradation processes for both manufacturing system and its embedded sensors considering not only the operating condition but also the degradation interaction/dependencies between the system and its sensors; (2)- development of observation models for measurement errors considering the degradation states of embedded sensors; (3)- based on the degradation and measurement errors models, the development of appropriate data-driven approaches for failure/RUL prediction of the manufacturing system.
These contributions provide answers to the scientific issues referring to the initial objective targeted:
- Modelling and qualification of the degradation interaction impacts between the manufacturing system and its sensors
- Development of degradation models for both system and its sensors considering the degradation interaction impacts and operating condition
- Development of measurement errors models (linear or non- linear) considering the degradation impact of sensors
- Elaboration of prognostics algorithms using particle filters or/and deep learning for failure/RUL prediction of the system from data uncertainty caused by degradation sensors
These contributions to these scientific issues are supported by different scientific tools, which are mainly mathematical tools (e.g., copula models, mixing distributions, stochastic process) and data-driven prognostic approaches (particle filters, deep learning ...).
These different modeling and development proposals will be implemented on Matlab software. Several performance metrics will be selected to evaluate the performance of the proposed models/approaches. Numerical examples will be used to validate these proposals (at least in the laboratory) while also trying to extend this validation step by using real data from industrial use cases (ex., Continental, INEOS, SEW USOCOME dans le cadre de projets européens comme AI-PROFICIENT (https://ai-proficient.eu/) et MODAPTO (2023-2025).
Department(s):
Modeling and Control of Industrial Systems |