Glossary · IIoT, data and AI
Machine learning operations (MLOps)
Also known as: MLOps, ML operations
German: Machine Learning Operations (MLOps)
In AI engineering, machine learning operations (MLOps) is the set of practices, roles and tools for reliably developing, deploying, monitoring and maintaining machine learning models in production, combining machine learning, software engineering (DevOps) and data engineering. It covers data and model versioning, automated pipelines, testing, deployment and monitoring.
- IIoT
- AI
In one sentence
MLOps is the set of practices and tools for reliably deploying, monitoring and maintaining machine learning models in production.
Example
A quality inspection model is retrained monthly in an automated pipeline, validated against a fixed test set and rolled out to edge devices only after approval.
How it applies
- Engineering: MLOps treats models, data and code as versioned artifacts and automates training, testing and packaging, so that every deployed model can be reproduced.
- Operation: Deployed models are monitored for Data drift, Model drift and technical health; retraining and rollback follow defined procedures.
- Change management: In automation, a new model version can change machine behavior. Integrate model releases into Change control and the plant's approval process, especially for AI functions that affect product quality or safety.
- Documentation: Maintain a model record for each version: purpose, training data, validation results, limitations and approval, for example as a Model card.
MLOps vs. DevOps
DevOps automates the delivery of software. MLOps adds what is specific to machine learning: dependence on data, statistical validation, drift, and the need to retrain models even when the code has not changed.