Ph. D. Project
Title:
Event-based AI predictive analytics for failure prediction of industrial systems
Dates:
2025/03/15 - 2028/03/14
Supervisor(s):
Other supervisor(s):
Cavalcante Cristiano (c.a.v.cavalcante@random.org.br)
Description:
General context
This doctoral project is part of the ANR PRCI project in collaboration with Brazil, TECHMAINT (Event-based predictive analytics for health management of industrial systems), which aims to develop predictive analytics models based on event data for predicting the health status of industrial systems. In this context, the objective of this thesis is to design and develop prognostic models based on Artificial Intelligence (AI) for predicting failures in industrial systems.
Scientific context and goals:
With the fast development of information and sensing technology as advocated by Industry 4.0 initiative, Prognostics and Health Management (PHM) is now widely investigated in industrial systems to contribute to their reliability, supportability, and economic affordability. Indeed, PHM is more an integrated approach for managing the system health in a life cycle perspective with a focus on failure prediction and maintenance optimization. Many prognostics models, especially data-driven ones, have been proposed and successfully applied to various industrial applications thanks to the advancement of predictive analytics techniques (predictive modelling, Machine Learning). The data can be categorized into condition monitoring data (CM) referred to the data measured by sensors in real-time and event data. Although CM data-based prognostic approaches are the most studied, continuous monitoring on an industrial plant is usually expensive or even impossible for certain complex systems despite its prevalence and effectiveness. One alternative for the prognostics purpose is therefore, to effectively incorporate the analysis of (discrete) event data. Event data provides useful information on what happened and what was done to the system. They are via processing supported by PLCs, supervision system, SCADA, etc. or via inspection, and can be considered in most of the cases as information or an indicator because already processed to be semantically consistent. So, this data provides rich information of system operation (without adding sensors even if some events are built from CM) and its degradation process.
The objective of this thesis is to develop innovative predictive approaches based on event data to estimate the Remaining Useful Life (RUL) of industrial systems while overcoming the limitations of traditional predictive maintenance methods. The proposed approaches incorporate advanced Artificial Intelligence (AI)-based prediction techniques as well as hybrid models that combine the strengths of both methodologies. Furthermore, these approaches will be validated using real industrial case studies, ensuring their practical applicability and integration into maintenance decision-making processes. The proposed approaches/models will be tested and validated using experimental datasets and/or real industrial use cases.
This research is part of a joint PhD program between UL and UFPE (Federal University of Pernambuco, Brazil). The thesis work will be conducted at UFPE for the first 18 months and at UL for the remaining 18 months.
This doctoral project is part of the ANR PRCI project in collaboration with Brazil, TECHMAINT (Event-based predictive analytics for health management of industrial systems), which aims to develop predictive analytics models based on event data for predicting the health status of industrial systems. In this context, the objective of this thesis is to design and develop prognostic models based on Artificial Intelligence (AI) for predicting failures in industrial systems.
Scientific context and goals:
With the fast development of information and sensing technology as advocated by Industry 4.0 initiative, Prognostics and Health Management (PHM) is now widely investigated in industrial systems to contribute to their reliability, supportability, and economic affordability. Indeed, PHM is more an integrated approach for managing the system health in a life cycle perspective with a focus on failure prediction and maintenance optimization. Many prognostics models, especially data-driven ones, have been proposed and successfully applied to various industrial applications thanks to the advancement of predictive analytics techniques (predictive modelling, Machine Learning). The data can be categorized into condition monitoring data (CM) referred to the data measured by sensors in real-time and event data. Although CM data-based prognostic approaches are the most studied, continuous monitoring on an industrial plant is usually expensive or even impossible for certain complex systems despite its prevalence and effectiveness. One alternative for the prognostics purpose is therefore, to effectively incorporate the analysis of (discrete) event data. Event data provides useful information on what happened and what was done to the system. They are via processing supported by PLCs, supervision system, SCADA, etc. or via inspection, and can be considered in most of the cases as information or an indicator because already processed to be semantically consistent. So, this data provides rich information of system operation (without adding sensors even if some events are built from CM) and its degradation process.
The objective of this thesis is to develop innovative predictive approaches based on event data to estimate the Remaining Useful Life (RUL) of industrial systems while overcoming the limitations of traditional predictive maintenance methods. The proposed approaches incorporate advanced Artificial Intelligence (AI)-based prediction techniques as well as hybrid models that combine the strengths of both methodologies. Furthermore, these approaches will be validated using real industrial case studies, ensuring their practical applicability and integration into maintenance decision-making processes. The proposed approaches/models will be tested and validated using experimental datasets and/or real industrial use cases.
This research is part of a joint PhD program between UL and UFPE (Federal University of Pernambuco, Brazil). The thesis work will be conducted at UFPE for the first 18 months and at UL for the remaining 18 months.
Keywords:
Prognostics, PHM, AI, Data-driven models, Discret event data
Department(s):
| Modelling and Control of Industrial Systems |
Publications: