Modern industrial systems increasingly need to bridge two worlds with fundamentally different constraints: deterministic, real-time control at the shop floor and data-intensive, AI-driven analytics in the cloud. This talk argues that real-time performance is not a property that scales uniformly through the system; it exists only in a narrow "waist" between two much larger, inherently non-real-time layers. At the bottom, sensors, PLCs, and SCADA systems are constrained by energy budgets, making continuous, deterministic transmission impractical, especially in IoT and IIoT deployments. At the top, big data, AI/ML pipelines, digital twins, and LLM-based agents operate on volumes of data that preclude hard real-time guarantees. Only in between – where standards such as OPC UA FX and the emerging IEC/IEEE 60802 Time-Sensitive Networking profile now converge to provide a common data model over deterministic Ethernet – can true real-time behavior be achieved. Drawing on the presenter's work in industrial communication and Service-Oriented Architecture, as well as recent work on vision-based diagnostic systems (including the MEDUSA project for ultrasound-based synovitis detection), the talk traces how data flows from the physical layer, through this real-time core, up to signal and image processing, edge AI, and cloud-based analytics. The talk situates this "hourglass" architecture within current standardization efforts (ISA-95:2025, RAMI 4.0/Asset Administration Shell, ISO 23247, OPC UA FX/TSN) and discusses the practical implications for designing systems that combine deterministic control with modern AI-based decision support.
Co-sponsored by: Poznan University of Technology
Speaker(s): Professor Marcin Andrzej Fojcik,
Room: room 201, Bldg: CENTER FOR MECHATRONICS, BIOMECHANICS, AND NANOENGINEERING, POZNAŃ UNIVERSITY OF TECHNOLOGY, ul. Jana Pawła II 24, 61-131 Poznań, Poland, Poznań, Wielkopolskie, Poland, 61-131, Virtual: https://events.vtools.ieee.org/m/575270
