As AI and machine learning become embedded in industrial biotechnology, more soft sensors are being built to monitor and predict process behaviour in real time. But building a model is only the first step keeping it accurate and reliable once deployed is a separate, ongoing challenge.
How do we make sure machine learning models keep working once they leave the lab?
A new open-access publication in SoftwareX (Suarez et al., 2026) tackles this question, introducing STAMM: the Soft sensor moniToring and mAintenance framework for Machine learning Models. Developed by researchers of the Bioindustry 4.0, the framework addresses a gap that often gets overlooked once a model is deployed: how to supervise it, keep it running, and catch it when it starts to drift.
STAMM brings together six components: data acquisition, a time-series database, workflow management, a model repository, real-time monitoring dashboards, and drift detection, into a single system that supports soft sensors throughout their working life, not just at deployment. The framework was built and validated using an industrial-scale fed-batch fermentation process as its guiding use case, but its design keeps it broadly applicable across other industrial process settings.
Read the whole publication:
Suarez, C., Astudillo, A., Metcalfe, B., Crowther, M., Koehorst, J. J., Castillo, E., Bize, A., & Corrales, D. C. (2026). STAMM: Soft sensor moniToring and mAintenance framework for Machine learning Models. SoftwareX. https://doi.org/10.1016/j.softx.2026.102783