Enabling automated model maintenance by application-oriented model diagnostic measures
|Title||Enabling automated model maintenance by application-oriented model diagnostic measures|
|Address||APACT 2018 - Advances in Process Analytics and Control Technology 2018 Conference, Newcastle, UK, April 25-27, 2018.|
A crucial part of using chemometric models in production is to ensure their ongoing predictive ability even after they have been deployed. Models can become unreliable over time due to variations of the analyzed medium or changes to the measurement hardware . Therefor we would like to achieve an efficient and automated approach for detecting that a calibrated model has become unreliable and requires recalibration.
We have developed an architecture for performing Computational Model Life-Cycle Management and implemented a prototype called ChemSaaS (Chemometric Software as a Service). ChemSaaS provides the infrastructure for computing predictions in real time (Inline System), storing huge amounts of chemometric data (Data Store) and models with all relevant information (Model Database). On top two separate user interfaces – a simplified one for semi-automatic model recalibration and monitoring purposes (Customer Tuning Environment) and a more flexible one which allows developing new workflows – have been implemented.