Invited Speakers

 

Biography: Pradeep Kundu is currently working as an assistant professor in the Department of Mechanical Engineering at KU Leuven, Belgium. Before joining KU Leuven, he worked as a Post-Doctoral Fellow and Research Associate at the University of Cincinnati, USA, and the University of Strathclyde, UK, respectively. His research focused on the broad domain of artificial intelligence for sustainable manufacturing, reliability engineering, predictive maintenance, prognostics and health management, and digital twins. His research helps industries in reducing unplanned outages, increasing productivity, automating quality control, and reducing operation and maintenance costs. He has published around 50 articles in reputed journals and conferences. He has delivered more than 20 keynote/invited talks. He has received several awards, including runner-up for PHM Europe 2022 Data Challenge, overseas visiting doctoral fellowship from SERB, etc. He serves as a guest/handling/associate editor for four journals, including Associate Editor for Measurement, Elsevier Journal. He has contributed to 10 conference committees, including his current role as General Chair for the 15th Prognostics and System Health Management Conference 2025.

Speech Title: Digital Twin Development for Rotating Machinery Condition Monitoring

Abstract: Pradeep Kundu's presentation addresses the critical issue of asset failure and degradation. These failures lead to unplanned outages, compromised product quality, decreased productivity, and increased operating and maintenance costs. To mitigate these challenges, machine learning (ML) based data-centric models are often employed for health assessment, encompassing anomaly detection, fault diagnostics, and prognostics for predicting remaining useful life. However, the scarcity of training data, especially for unreported damages, poses a significant limitation in the development of these ML models.
Physics-based models, which utilize the understanding of damage progression, offer an alternative by reducing the data requirements of data-driven models. Despite their potential, these models can exhibit high modeling errors due to assumptions and simplicity. The digital twin, a dynamic, virtual representation of a physical asset, presents an opportunity to overcome these limitations and enhance prediction accuracy.
The talk focuses on the development of a robust digital twin model that can address various challenges such as the unavailability of data for all failure modes, the black-box nature of ML models, high modeling uncertainty at the fleet level, and the assumptions inherent in physics-based models. By integrating both data-centric and physics-based approaches, the digital twin can provide a more accurate and reliable prediction framework.
The presentation is structured to first explain the impact of asset failure and degradation on productivity, quality, and costs in manufacturing environments. It then delves into the use of ML-based models for health assessment, the limitations posed by insufficient training data, and the role of physics-based models in mitigating these limitations. Finally, it discusses the potential of digital twins in creating robust predictive models and outlines strategies for their development and implementation.

 

Biography: TBA