[CDD 3ANS THESE] - Hybrid Approaches to Fault Prognostics in Complex and Uncertain Industrial Systems

Scientific Fields: Fault Prognostic, Knowledge Management, Industry 5.0.

Research Work

Thesis Subject Summary

As the cutting edge in industrial evolution, Industry 5.0 seamlessly fuses human intelligence with advanced technologies to forge highly personalized and hyper-efficient production systems. Prognostics and health management (PHM) techniques stand at the heart of this transformative era, delivering indispensable tools for proactive maintenance and peak performance optimization [20]. In this dynamic landscape, hybrid methods that synergize data-driven approaches with expert knowledge are surfacing as powerful solutions to the intricate challenges of contemporary industrial environments [8, 14]. The capability to manage uncertainty and operate within distributed systems is pivotal to the triumph of these innovative approaches [9].

This thesis endeavors to pioneer and validate cutting-edge prognostic models that harness neuro-symbolic methods and data fusion, aligning with the ambitious vision of Industry 5.0. The primary objective is to design a robust and sustainable predictive maintenance solution that not only meets but exceeds optimization and efficiency standards, effectively addressing the myriad challenges inherent in industrial maintenance [7, 17].

Thesis Project

Scientific Context

Industry 5.0 represents the next evolution of the industrial sector, where human intelligence, advanced technologies, and artificial intelligence (AI) converge to create more flexible, efficient, and personalized production systems.

Unlike Industry 4.0, which focused primarily on automation and data exchange, Industry 5.0 emphasizes the collaboration between humans and machines, leveraging the strengths of both to achieve unprecedented levels of productivity and innovation [19].

In this advanced industrial landscape, prognostics play a crucial role. Prognostics involve predicting the future condition and performance of systems and components, which is essential for proactive maintenance, optimizing operational efficiency, and minimizing downtime [16]. Accurate prognostics enable industries to anticipate failures before they occur, schedule maintenance activities more effectively, and ensure the smooth functioning of production processes. However, the complexity of modern industrial systems presents significant challenges for traditional prognostic methods. The data generated by these systems are vast, diverse, and often noisy. Additionally, the integration of various subsystems and the interactions between them introduce layers of complexity that are difficult to model and predict using conventional techniques [22].

Thesis Subject:

To address the challenges of modern industrial prognostics, a hybrid approach that combines data-driven methods and expert knowledge is essential. Data-driven methods, such as machine learning and statistical analysis, excel at identifying patterns and making predictions based on large datasets. Conversely, expert knowledge encapsulates years of human experience and domain-specific insights, which are invaluable for understanding the underlying mechanisms of system behavior [21]. The fusion of these two approaches can lead to more accurate and robust prognostic models [18]. However, implementing such a hybrid approach in real-world industrial environments introduces additional challenges. These environments are often distributed, with various components located across different geographical locations. This distribution requires solutions that can process data and execute models in a decentralized manner while ensuring consistency and reliability [23]. Moreover, uncertainty is an inherent aspect of industrial prognostics. Uncertainty arises from various sources, including measurement noise, incomplete data, and the stochastic nature of many industrial processes.

Managing this uncertainty is critical for making reliable predictions and informed decisions [3]. To navigate these complexities, neuro-symbolic approaches offer a promising solution. These approaches combine the learning capabilities of neural networks with the logical reasoning and interpretability of symbolic systems [7].

By leveraging the strengths of both, neuro-symbolic methods can enhance the accuracy and explainability of prognostic models, making them more suitable for complex industrial applications [10].

In summary, this thesis is set against the backdrop of Industry 5.0, where the integration of human and machine intelligence is driving the next wave of industrial innovation. The development of advanced prognostic methods that combine data-driven techniques and expert knowledge, manage uncertainty, and operate in distributed environments is essential for realizing the full potential of Industry 5.0. This thesis aims to contribute to this vision by developing and validating hybrid prognostic approaches that address these challenges and advance the state of the art in industrial prognostics.

Scientific Challenges

The development of advanced prognostic models for Industry 5.0 involves addressing several scientific challenges.

These challenges encompass the integration of diverse data sources, the creation of hybrid neuro-symbolic models, the management of uncertainty, the design of distributed environments, and the practical application of these approaches in real-world industrial settings. Below are the key scientific challenges to be tackled:

1. Data and Knowledge Fusion

• Develop methods to effectively integrate heterogeneous data from various sources and formalize expert knowledge into a unified framework.
• Address conflicts and inconsistencies between data and knowledge, and explore new methods for knowledge integration.

2. Neuro-Symbolic Approaches

• Design hybrid models: One major challenge is to effectively design hybrid models that integrate the deep learning capabilities of neural networks with symbolic systems for knowledge management. This requires finding a balance between the strengths of both approaches, ensuring seamless interaction, and overcoming the inherent differences in their methodologies.
• Ensure explainability and traceability: Another significant hurdle is to ensure the explainability and traceability of decisions made by these hybrid models. While deep learning models, particularly neural networks, are often seen as "black boxes" with little transparency, symbolic systems are known for their clarity and logic. The challenge lies in making the complex decision-making processes of these hybrid models comprehensible and ensuring that each step can be traced back and justified.

3. Uncertainty Management

• Develop techniques to quantify and manage uncertainty in prognostic predictions, using uncertainty management frameworks such as belief function theory.

4. Distributed Environments

• Design distributed architectures for data processing and the execution of prognostic models.

5. Application to Industry 5.0

• Demonstrate the effectiveness of the proposed approaches in real-world use cases within the industry.

Previous Work in the Laboratory:

This thesis aims to push the boundaries of prognostic technologies by integrating innovative hybrid approaches and addressing the challenges posed by uncertainty and distributed environments, thus contributing to the ambitious goals of Industry 5.0. Several research projects have been conducted at the CESI Lineact laboratory, particularly focusing on predictive maintenance in Industry 4.0 contexts [12, 13]. These include data-driven approaches for Remaining Useful Life (RUL) prediction applied to aircraft engines [5, 6, 15], as well as knowledge-based methods [11] and hybrid approaches explored in various industrial settings [1, 2]. Other work has focused on uncertainty management using belief functions [3, 4].

Work Program: Steps and Schedule

1. Literature Review and Initial Research (Months 1-3):

• Conduct an extensive review of existing literature on prognostics, neuro-symbolic methods, and Industry 5.0 technologies.
• Identify key challenges and gaps in current research such as uncertainty management.

2. Development of Hybrid Prognostic Models (Months 4-9):

• Design and implement hybrid models that integrate neural networks with symbolic reasoning systems.
• Develop algorithms for data fusion and uncertainty management.

3. Implementation and Testing (Months 10-18):

• Deploy the models in a simulated industrial environment using CESI LINEACT’s facilities.
• Conduct extensive testing to refine and validate the models.

4. Real-world Application and Evaluation (Months 19-24):

• Gather and analyze data to evaluate the performance and robustness of the models.
• Collaborate with industrial partners to apply the models in real-world settings.

5. Finalization and Documentation (Months 25-30):

• Compile the research findings and draft the thesis.
• Prepare publications for conferences and journals.
• Defend the thesis and complete the PhD requirements.

Expected Scientific/Technical Output: Publications and Deliverables

1. Journal Articles and Conference Papers:

• Publish findings in high-impact journals such as the IEEE Transactions on Industrial Informatics and the Journal of Process Control.
• Present research at international conferences such as the IEEE International Conference on Prognostics and Health Management and the International Conference on Applied Computing.

2. Software Tools:

• Develop and release software tools for prognostics and health management that incorporate neurosymbolic methods and data fusion techniques.
• Ensure the tools are documented and accessible to both academic and industrial users.

3. PhD thesis:

• Complete and defend the PhD thesis, summarizing the research findings and contributions to the field of fault prognostics in industrial systems.

References

[1] Fidma Mohamed Abdelillah, Hamour Nora, Samir Ouchani, and Sidi Mohamed Benslimane. Predictive maintenance approaches in industry 4.0: A systematic literature review. In IEEE International Conference on Enabling Technologies: Infrastructure for Collaborative Enterprises, WETICE 2023, Paris, France, December 14-16, 2023, pages 1–6. IEEE, 2023.

[2] Fidma Mohamed Abdelillah, Hamour Nora, and Samir Ouchani and Sidi Mohamed Benslimane. Hybrid data-driven and knowledge-based predictive maintenance framework in the context of industry 4.0. In Model and Data Engineering - 12th International Conference, MEDI 2023, Sousse, Tunisia, November 2-4, 2023, Proceedings, volume 14396 of Lecture Notes in Computer Science, pages 319–337. Springer, 2023.

[3] Safa Ben Ayed, Roozbeh Sadeghian, and Rachid Hamza. Remaining useful life prediction with uncertainty quantification using evidential deep learning. [Reviewing in progress].

[4] Safa Ben Ayed, Malika Ben Khalifa, and Samir Ouchani. Modeling distributed and flexible phm framework based on the belief function theory. In IFIP International Conference on Artificial Intelligence Applications and Innovations, 2024.

[5] Ibrahima Barry and Meriem Hafsi. Towards hybrid predictive maintenance for aircraft engine: Embracing an ontological-data approach. In 20th ACS/IEEE International Conference On Computer Systems And Applications, December 2023. URL https://hal.science/hal-04369673.

[6] Mohamed-Amin Benatia, Meriem Hafsi, and Safa Ben Ayed. A continual learning approach for failure prediction under non-stationary conditions: Application to condition monitoring data streams. [Reviewing in progress].

[7] T. R. Besold, A. d’Avila Garcez, S. Bader, H. Bowman, P. Domingos, P. Hitzler, and G. Zaverucha. Neural-symbolic learning and reasoning: A survey and interpretation 1. In Neuro-Symbolic Artificial Intelligence: The State of the Art, pages 1–51. IOS press, 2021.

[8] Q. Cao, C. Zanni-Merk, A. Samet, C. Reich, F. Beuvron, A. Beckmann, and C. Giannetti. Kspmi: A knowledge-based system for predictive maintenance in industry 4.0. Robotics And Computer-Integrated Manufacturing, 74:102281, 2022.

[9] C. Chen, J. Shi, N. Lu, Z. Zhu, and B. Jiang. Data-driven predictive maintenance strategy considering the uncertainty in remaining useful life prediction. Neurocomputing, 494:79–88, 2022.

[10] A. d. Garcez, M. Gori, L. C. Lamb, L. Serafini, M. Spranger, and S. N. Tran. Neural-symbolic computing: An effective methodology for principled integration of machine learning and reasoning. ArXiv preprint arXiv:2102.12344, 2021.

[11] Meriem Hafsi. Exploring a knowledge-based approach for predictive maintenance of aircraft engines: Studying fault propagation through spatial and topological component relationships. In 8th European Conference Of The Prognostics And Health Management Society 2024, volume 8, July

2024. URL https://hal.science/hal-04649557.

[12] Meriem Hafsi, Nora Hamour, and Samir Ouchani. Advancing industry: A systematic literature review of recent developments and perspectives in predictive maintenance for cps. [Reviewing in progress].

[13] Meriem Hafsi, Nora Hamour, and Samir Ouchani. Predictive maintenance for smart industrial systems: A roadmap. In The 6th International

Conference On Emerging Data And Industry (EDI40), March 2023. URL https://hal.science/hal-04398272.

[14] S. Hagmeyer, P. Zeiler, and M. Huber. On the integration of fundamental knowledge about degradation processes into data-driven diagnostics and prognostics using theory-guided data science. In PHM Society European Conference, volume 7, pages 156–165, 2022.

[15] Rachid Hamza and Safa Ben Ayed. Supervised learning based approach for turbofan engines failure prognosis within the belief function framework. In International Conference on Applied Computing, 2022.

[16] Y. Lei, N. Li, L. Guo, Z. Xu, N. Li, and T. Yan. Machinery health prognostics: A systematic review from data acquisition to rul prediction.

Mechanical Systems and Signal Processing, 104761, 2020.

[17] A. Liguori, S. Mungari, E. Ritacco, F. Ricca, G. Manco, and S. Iiritano. Neuro-symbolic techniques for predictive maintenance, 2023.

[18] Q. Liu, W. Zhang, B. Wang, and J. He. A hybrid prognostic model for predicting remaining useful life of mechanical systems. Reliability Engineering & System Safety, 208:107399, 2021.

[19] P. K. R. Maddikunta, Q. V. Pham, S. M. Basha, and T. R. Gadekallu. Industry 5.0: A survey on enabling technologies and potential applications. Journal of Industrial Information Integration, 26:100257, 2022.

[20] R. Oudenhoven and E. Demerouti. Predictive maintenance for industry 5.0: behavioural inquiries from a work system perspective. International Journal Of Production Research, 61:7846–7865, 2023. doi: 10.1080/00207543.2022.2154403.

[21] J. Z. Sikorska, M. Hodkiewicz, and L. Ma. Prognostic modelling options for remaining useful life estimation by industry. Mechanical Systems and Signal Processing, 144:106860, 2021.

[22] R. Zhao, R. Yan, Z. Chen, K. Mao, P. Wang, and R. X. Gao. Deep learning and its applications to machine health monitoring: A survey. IEEE Transactions on Neural Networks and Learning Systems, 33(3):1094–1113, 2021.

[23] K. Zhou, L. Zhang, and S. Yang. Cyber-physical-social system in intelligent manufacturing. Automation in Construction, 137:104164, 2022.

CESI LINEACT (UR 7527), Laboratory for Digital Innovation for Businesses and Learning to Support the

Competitiveness of Territories, anticipates and accompanies the technological mutations of sectors and services related to industry and construction. The historical proximity of CESI with companies is a determining element for our research activities, and has led us to concentrate our efforts on applied research close to the company and in partnership with them. A human-centered approach coupled with the use of technologies, as well as territorial networking and links with training, have enabled the construction of cross-cutting research; it puts humans, their needs and their uses, at the center of its issues and addresses the technological angle through these contributions.

Its research is organized according to two interdisciplinary scientific teams and several application areas.

• Team 1 "Learning and Innovating" mainly concerns Cognitive Sciences, Social Sciences and Management Sciences, Training Techniques and those of Innovation. The main scientific objectives are the understanding of the effects of the environment, and more particularly of situations instrumented by technical objects (platforms, prototyping workshops, immersive systems...) on learning, creativity and innovation processes.
• Team 2 "Engineering and Digital Tools" mainly concerns Digital Sciences and Engineering. The main scientific objectives focus on modeling, simulation, optimization and data analysis of cyber physical systems. Research work also focuses on decision support tools and on the study of human-system interactions in particular through digital twins coupled with virtual or augmented environments.

These two teams develop and cross their research in application areas such as: Industry 5.0, Construction 4.0 and City of the Future, Digital Services.

Scientific and Technical Skills:

• Artificial Intelligence: Experience in Machine Learning and Deep Learning,
• Programming: Proficiency in several languages: Object Oriented, Python and IA librairies,
• Modeling, Knowledge representation and Reasoning ( rules-based and Expert systems, ontologies).

Interpersonal Skills:

• Being autonomous, having initiative and curiosity,
• Ability to work in a team and have good interpersonal skills,
• Being rigorous.