USW researchers develop AI tool to transform wheelchair component identification

University of South Wales (USW) researchers are developing an innovative AI-driven software tool which could transform how individual machine components are identified and documented.

Working in collaboration with Cardiff and Vale University Health Board’s Rehabilitation Engineering Unit, the tool is specifically designed to identify parts of wheelchairs.

The AI-based object detection system can automatically recognise and catalogue various wheelchair components from images, including frames, wheels, footrests, cushions, brakes, and more.

Janusz Kulon, Professor of Biomedical Engineering and Applied Artificial Intelligence at USW and Principal Investigator, said the project could lead to major savings for the health service and beyond.

“With approximately 27,000 parts and around £1m worth of stock per typical posture and mobility service spread across numerous supplier catalogues, this tool’s ability to streamline identification is invaluable,” Professor Kulon said

“This is the first AI system of its kind, built on a bespoke dataset of wheelchair components that previously didn’t exist.

“To train the model, we captured thousands of video frames showing individual parts. Each image was manually labelled — a laborious but essential step — to ensure the model could accurately distinguish visually similar components under different lighting, angles, and conditions.

“The tool aims to eliminate one of the most time-consuming and error-prone tasks in posture and mobility services: the manual documentation and sourcing of wheelchair parts.”

Professor Colin Gibson, Head of the Rehabilitation Engineering Unit at Cardiff and Vale University Health Board CVUHB, highlighted the real-world challenges the tool addresses.

“Identifying the correct wheelchair component can be surprisingly complex and time-consuming,” he said.

“We manage a caseload of thousands of users across South Wales. Every wheelchair is unique — often custom-built — with many interchangeable parts from multiple suppliers.

“Often, parts must be matched by eye, cross-referenced against large paper catalogues, or tracked via serial numbers. This becomes especially problematic during urgent repairs or replacements.”

Professor Gibson added that the AI tool developed by USW researchers promises to significantly reduce that burden.

“Clinicians can visually capture a component using a mobile device (either directly or having received an image forwarded by a service user at a remote location) and instantly receive its identification, specifications, and location,” he said.

“This capability can dramatically reduce repair times and improve service delivery, particularly for users in rural or low-access areas.”

Professor Kulon noted that the proof of concept — which demonstrated high performance in part identification — has already been successfully showcased.

“Our next steps involve expanding the dataset and retraining the model for broader deployment,” he said.

“It’s not just about automation. It’s about accuracy, speed, and improving clinical outcomes. Quickly and correctly identifying parts reduces human error, streamlines the prescription and repair process, and frees up clinicians to spend more time with patients.

“We are also actively seeking further innovation funding to scale up the project and bring even greater efficiency to wheelchair component management.”