STAMM — Soft sensor moniToring and mAintenance framework for Machine learning Models

As the number of models and simulations capable of integrating real-time data continues to grow — including machine learning–based soft sensors, dynamic models and hybrid models — so does the need to deploy them in real-world environments where they can effectively support decision-making in real time.

However, moving machine learning models into production is not trivial. Ensuring they remain reliable, reproducible, and robust in dynamic environments requires more than accurate predictions: it requires a structured approach. This is where Machine Learning Operations (MLOps) becomes essential. By combining principles from software engineering with data science workflows, MLOps enables the automation and management of the entire model lifecycle — from data preparation and training to deployment, monitoring, and continuous improvement.

In real-world applications, machine learning models face several challenges, including performance degradation due to data drift, reproducibility issues caused by evolving software dependencies, and silent failures resulting from inconsistencies in computational environments. Without proper monitoring and infrastructure, these issues can go unnoticed, leading to reduced model accuracy and reliability over time.

To address these challenges, we present STAMM: Soft sensor moniToring and mAintenance framework for Machine learning Models, an open-source MLOps framework specifically designed for soft sensors. STAMM enables the deployment of machine learning models in production environments while integrating real-time data from physical sensors and simulations into an interactive dashboard. It continuously monitors model performance and detects concept drift through a dedicated data pipeline, ensuring long-term reliability and transparency.

📄 STAMM Preprint

Suarez, Carlos; Astudillo, Alexander; Metcalfe, Brett; Crowther, Matthew; Koehorst, Jasper J.; Castillo, Esteban; Bize, Ariane and Corrales, David Camilo.
STAMM: Soft sensor moniToring and mAintenance framework for Machine learning Models.

Available at SSRN: https://ssrn.com/abstract=6054948
DOI: 10.2139/ssrn.6054948