Particle processes are used broadly in industry and are frequently used for removal of insolubles, product isolation, purification and polishing. These processes are challenging to control due to their complex dynamics and physical-chemical properties. With the developments in particle monitoring tools make it possible to gain real-time insights into some of these process dynamics. In this work, a systematic modelling framework is proposed for particle processes based on a hybrid modelling concept, which integrates first-principles with machine-learning approaches. Here, we utilize on-line/at-line sensor data to train a machine learning based soft-sensor that predicts particle phenomena kinetics by combining it with a mechanistic population balance model. This approach allows flexibility towards use of process sensors and the model predictions do not violate physical constraints. Application of the framework is demonstrated through a laboratory-scale lactose crystallization, a laboratory-scale flocculation, and an industrial-scale pharmaceutical crystallization, using only limited prior process knowledge.
- Hybrid modelling
- Modelling framework
- Machine learning based soft-sensor
- Real-time training