Trainee Project
Title:
Exploring the combination of data-driven AI, ontologies, and reasoning for prognostics and health management of
production systems
Dates:
2025/03/01 - 2025/07/31
Supervisor(s):
Description:
Predictive Maintenance and Prognostics and Health Management (PM&PHM) approaches aim to intervene in the
equipment of production systems before faults occur. To properly implement a PHM system, data-centric steps
must be taken, including data acquisition and manipulation, detection of machine states, health assessment,
prognosis of future failures, and advisory generation (ISO 13374-2: 2007; Franciosi et al., 2024). Nevertheless,
data-driven approaches require knowledge to be exploited and provide their full "power". Indeed, data embed
relevant information/knowledge in relation to the usage, health and context of systems, which needs to be
exploited. However, revealing the real value of data and discovering useful patterns of knowledge embedded in
maintenance data (encompassing data from Programmable Logic Controller up to Maintenance Management
System) have presented a major challenge due to the heterogeneity of data sources and the variety of data types
(Karray et al., 2019). Ontologies can effectively contribute to resolving this issue through the organization of data,
semantic annotation and integration (Karray et al., 2010). Moreover, as reported in many reviews (see for instance
Biggio & Kastanis, 2020; Fink et al., 2020; Nguyen et al., 2023), all PM&PHM steps will benefit from leveraging
data-driven AI algorithms. Indeed, data-driven AI technologies, such as machine learning and data mining, have
been used in the literature to detect and predict potential anomalies and, hence, improve production efficiency
within manufacturing processes (Cao et al., 2022). However, both the lack and the excess of heterogeneity of data
can impact the predictability of the algorithms (Dalzochio et al., 2020) and therefore their performance.
This highlights the opportunity to combine symbolic AI (as ontologies), reasoning (as SWRL rule-based reasoning),
and data-driven AI for predictive maintenance (Franciosi et al., 2024). In this respect, symbolic AI and data-driven
AI can support each other and be combined toward the development of predictive maintenance approaches. On
the one hand, machine learning algorithms can enable the extraction of concepts and patterns (as machine
degradation models) from data or the calculation of information (as specific indices) that can then be enriched
due to the querying performed by maintenance domain ontologies and rule-based reasoning on this input data.
On the other hand, data from various sources and systems within a facility could be integrated to gather accurate
information in order to improve data-driven approaches.
Therefore, the objectives of this Master Thesis are the following: (1) exploring the scientific literature on
ontologies and machine learning techniques in order to understand in which way they are concretely combined
and what are the applications in the production contexts; (2) proposing a learning and reasoning approach for
PM&PHM of production systems.
equipment of production systems before faults occur. To properly implement a PHM system, data-centric steps
must be taken, including data acquisition and manipulation, detection of machine states, health assessment,
prognosis of future failures, and advisory generation (ISO 13374-2: 2007; Franciosi et al., 2024). Nevertheless,
data-driven approaches require knowledge to be exploited and provide their full "power". Indeed, data embed
relevant information/knowledge in relation to the usage, health and context of systems, which needs to be
exploited. However, revealing the real value of data and discovering useful patterns of knowledge embedded in
maintenance data (encompassing data from Programmable Logic Controller up to Maintenance Management
System) have presented a major challenge due to the heterogeneity of data sources and the variety of data types
(Karray et al., 2019). Ontologies can effectively contribute to resolving this issue through the organization of data,
semantic annotation and integration (Karray et al., 2010). Moreover, as reported in many reviews (see for instance
Biggio & Kastanis, 2020; Fink et al., 2020; Nguyen et al., 2023), all PM&PHM steps will benefit from leveraging
data-driven AI algorithms. Indeed, data-driven AI technologies, such as machine learning and data mining, have
been used in the literature to detect and predict potential anomalies and, hence, improve production efficiency
within manufacturing processes (Cao et al., 2022). However, both the lack and the excess of heterogeneity of data
can impact the predictability of the algorithms (Dalzochio et al., 2020) and therefore their performance.
This highlights the opportunity to combine symbolic AI (as ontologies), reasoning (as SWRL rule-based reasoning),
and data-driven AI for predictive maintenance (Franciosi et al., 2024). In this respect, symbolic AI and data-driven
AI can support each other and be combined toward the development of predictive maintenance approaches. On
the one hand, machine learning algorithms can enable the extraction of concepts and patterns (as machine
degradation models) from data or the calculation of information (as specific indices) that can then be enriched
due to the querying performed by maintenance domain ontologies and rule-based reasoning on this input data.
On the other hand, data from various sources and systems within a facility could be integrated to gather accurate
information in order to improve data-driven approaches.
Therefore, the objectives of this Master Thesis are the following: (1) exploring the scientific literature on
ontologies and machine learning techniques in order to understand in which way they are concretely combined
and what are the applications in the production contexts; (2) proposing a learning and reasoning approach for
PM&PHM of production systems.
Keywords:
artificial intelligence, prognostics and health management, productions systems.
Department(s):
| Modelling and Control of Industrial Systems |
