Predictive Maintenance: Latest Trends, Deployment Challenges and Future Directions

Predictive Maintenance (PdM) is evolving from a data-driven approach for reducing unplanned downtime into a strategic capability for resilient, adaptive and cost-efficient industrial operations. This webinar will provide a Technology Overview by CRIT, focusing on the latest trends shaping modern PdM systems. The presentation will cover the evolution from reactive, preventive and condition-based maintenance to predictive and prescriptive models; the end-to-end PdM cycle from data acquisition and preprocessing to diagnostics, prognostics and decision support; and the main algorithmic families, including data-driven, physics-based, knowledge-based and hybrid AI approaches.

A central focus will be the current landscape of AI algorithms for PdM: machine learning and deep learning models for time-series and sensor analytics, hybrid approaches combining data-driven models with physics and domain knowledge, weakly supervised and transfer learning methods to address scarce fault labels, explainable AI for trustworthy deployment, digital-twin-enabled modelling and emerging roles for foundation models, LLMs, generative AI and reinforcement learning in maintenance decision-making. Particular attention will be given to practical deployment challenges, such as data scarcity, weak labels, sensor quality, model validation, explainability, human-in-the-loop workflows and integration with digital twins.

The webinar will also feature an in-depth contribution by Prof. Olga Fink, Assistant Professor of Intelligent Maintenance and Operations Systems at École Polytechnique Fédérale de Lausanne (EPFL), focused on scalable and trustworthy industrial AI for predictive maintenance. Prof. Fink talk will explore how four complementary paradigms, namely Graph Neural Networks (GNNs), Physics-Informed Machine Learning (PIML), Tabular Foundation Models (TFMs) and Agentic AI, applied to predictive maintenance, are driving the industry transition from isolated analytics toward autonomous, reasoning-capable systems combining predictive intelligence with rigorous physical understanding.