Industrial OneLoop-AI Lifecycle Management for the Industrial Edge

Custom trained AI models used in industrial environments can lose performance as equipment, processes, and operating conditions change. Maintaining these models across devices and sites often requires repeated data collection, retraining, validation, and deployment.

Industrial OneLoop provides a closed-loop process for maintaining and improving industrial AI after deployment.

Edge devices such as cameras, robots, sensors, autonomous mobile robots, and inspection systems provide operational and telemetry data. When performance gaps or edge cases are identified, relevant scenarios can be reproduced within NVIDIA Omniverse to generate new synthetic training data.

Amazon SageMaker is used to train, evaluate, and manage candidate model versions. AWS IoT Greengrass deploys selected candidates to NVIDIA Jetson devices, where they can run in shadow mode alongside the current production model to validate model performance in the field.

Performance is compared under real operating conditions using AWS monitoring and IoT services. Candidates that meet defined performance KPIs can be released to production. If further improvement is needed, additional data and simulation can be used for another training cycle.

Industrial OneLoop helps organizations:

• Reduce manual model maintenance efforts and associated operational costs
• Improve and maintain model performance as operating conditions change
• Validate model changes before production rollout
• Apply model improvements across multiple devices and sites
• Reduce downtime

The platform supports use cases including quality inspection, inventory management, industrial computer vision, robotics, predictive maintenance, warehouse automation, supply chain optimization, and industrial automation.

Industrial OneLoop uses NVIDIA Jetson, Omniverse, and OpenUSD together with AWS services including Amazon SageMaker, Amazon S3, AWS IoT Greengrass, AWS IoT Core, Amazon CloudWatch, and Amazon Timestream.