[INTRAL] Interpretable and Interactive Transfer Learning in Process Analytical Technology

In May 2021 the project INTRAL was jointly started with our academic partner, the Software Competence Center Hagenberg GmbH and the industrial partners, Sandoz GmbH and Baxalta Innovations GmbH.

Within INTRAL we aim at developing and promoting transfer learning methods for Process Analytical Technology (PAT) applications in the (bio-) pharmaceutical and related domains.

In current PAT applications a significant amount of resources are required for data acquisition in order to derive process and sensor calibration models and to maintain them along the product lifecycle. Changing a probe or a bioreactor, or changing the production to a microorganism with slightly different genotype during process development usually requires development of new models. As a consequence, the full potential of PAT is hardly leveraged in the (bio-) pharmaceutical and related industries yet.


Transfer learning (TL) bears the potential to dramatically reduce the resources required for model development, maintenance and adaptation to new process conditions as it allows, in contrast to conventional modelling approaches, more efficient integration of historical data from related domains.

The INTRAL consortia will adapt existing transfer learning algorithms to fit the rigorous requirements of (bio-) pharmaceutical industry and will show their performance and usefulness in industrially relevant use cases for more transferable spectroscopic sensor calibrations and easy adaptable process models.