FEMM Hub presents at ICMR Conference

FEMM Hub at ICMR 2025: Advancing Digital Manufacturing for Electric Machines

The Future Electrical Machines Manufacturing (FEMM) Hub was pleased to contribute to the International Conference on Manufacturing Research (ICMR) 2025, held in Birmingham this September. ICMR is a leading UK forum for showcasing cutting-edge manufacturing research and fostering dialogue between academia, industry and government. This year, FEMM Hub researchers presented two papers that highlight our work at the intersection of digital manufacturing, machine learning and the circular economy.

Our contributions focused on two key challenges for sustainable manufacturing. The first explored the modelling and inspection of deformable linear objects such as wires, cables and coils in electrical machine production. The second examined the recognition of human actions in assembly tasks to support Re-X pathways such as repair, remanufacture and recycling. Together, these strands demonstrate the Hub’s mission to deliver digitally enabled and data-driven approaches that improve productivity, flexibility and sustainability in UK manufacturing.

⸻

Deformable Linear Object Modelling Through the Use of Synthetic Image Data

Presented by Dr Michael Farnsworth

Electrical machines rely heavily on the precise handling of wires, cables and coils, yet these deformable linear objects remain notoriously difficult to model, monitor and automate. Much of this work is still human led, where operators guide, place and route cables through assemblies, often under tight spatial and geometric constraints. Small deviations in winding tension or cable routing can result in faults that compromise performance, while the random nature of wire motion during these assembly tasks makes automation a major challenge. A further obstacle is the lack of large labelled datasets for training machine learning models. Unlike fields such as natural image recognition, manufacturing rarely produces the scale of data required for state-of-the-art deep learning.

This work addresses the dual challenges of physical complexity and data scarcity by developing a framework to generate synthetic image datasets for wire and cable manipulation tasks. Using 3D modelling and simulation tools, realistic training data was produced to support segmentation and detection of cables in manufacturing scenarios. Benchmark experiments with leading segmentation models demonstrated that incorporating synthetic data can improve accuracy and robustness by over 20 percent, particularly when dealing with out-of-distribution tasks. Beyond technical results, this research provides a practical pathway for data-efficient digital twins of manufacturing processes, enabling improved inspection, monitoring and ultimately automation of complex coil winding and wiring operations across the product lifecycle, including at end of life where Re-X decisions must be made.

⸻

Taxonomy Enhanced Human Action Recognition in Manufacturing Assembly

Presented by Mr Junxi Zhang (PhD researcher)

In addition to modelling materials and processes, the FEMM Hub is also addressing the challenge of understanding human activity in manufacturing environments. Assembly and disassembly tasks are central to both production and end-of-life processes, and they are often complex and human led. The same action may be performed with different tools or postures, and some actions are subtle, overlapping or ambiguous to machine vision systems. This variability makes it difficult to reliably track human actions, limiting the potential for intelligent support systems and human-robot collaboration.

Junxi Zhang’s paper introduces a domain-specific taxonomy for manufacturing assembly, encompassing over forty classes of actions, relationships and attributes. By embedding this structured knowledge into multimodal action recognition models, the research demonstrates that models can focus more effectively on the features that truly define an action, improving recognition accuracy even under difficult conditions. This work also emphasises the need for richer datasets in manufacturing, datasets that go beyond simple action labels to capture context, tool use, motion dynamics and operator variability. Such enriched models are directly relevant to Re-X pathways, since accurate tracking of human actions in disassembly, repair or remanufacturing tasks provides the data foundation for better decision making, digital product passports and adaptive automation in the circular economy.

⸻

The FEMM Hub’s participation in ICMR 2025 underscores our commitment to combining advanced modelling, synthetic data generation and multimodal AI to tackle the most pressing challenges in manufacturing sustainability. We welcome engagement from industry, academic and policy partners interested in collaborating on these themes. For further information, or to discuss potential partnerships, please contact the FEMM Hub team at l.j.farnsworth@sheffield.ac.uk