AMRC Uses IBM Maximo Visual Inspection for AI-Driven Manufacturing Defect Detection

Overview

The UK's Advanced Manufacturing Research Centre (AMRC) deployed IBM Maximo Visual Inspection software on an IBM Power Systems AC922 server with FlashSystem storage at its Factory 2050 facility to automate detection of faulty manufactured components, replacing manual light-beam inspection methods. The AI-optimized platform delivers 5 GB/s image-transfer speeds (versus 70 MB/s on prior on-premises infrastructure), trained an accurate vision model from about 500 images in roughly 20 minutes, and delivered a proof-of-concept for a startup customer in 5 days.

The challenge

AMRC sought an AI-enhanced visual inspection system to detect faulty manufactured parts faster and with less waste. Traditional computer-vision approaches were thrown off by changes in lighting and item position and could take weeks to fine-tune per fault type, while the centre's existing on-premises infrastructure — CPU-based servers in a separate building linked by a 1 Gb/s network — could not keep pace with the roughly 20 GB of images generated per component.

The solution

AMRC deployed IBM Maximo Visual Inspection software on an IBM Power Systems AC922 server (pairing POWER9 CPUs with NVIDIA Tesla V100 GPUs) in its Factory 2050 facility, paired with a colocated IBM FlashSystem storage solution, using the software's data-labeling, model-training and deployment tools to build computer-vision models for defect detection.

Reported business value

The new platform delivers image-transfer speeds of 5 GB/s, versus 70 MB/s on the prior on-premises infrastructure, trained an accurate vision model from about 500 images in roughly 20 minutes, and let the team deliver a proof-of-concept visual-inspection solution for a startup customer in just 5 days.