The importance of interlinking process knowledge and data
We utilize the power of hybrid modeling for our technology, delivering tools to enable better process understanding, faster process development, predictive modeling, and model predictive control.
What is Hybrid Modeling?
We believe that valuable process information, hidden in the data, and mechanistic process understanding can be combined in powerful hybrid models to estimate process variables of interest in real-time or even predict them multiple steps ahead. We develop these models by utilizing two different modeling approaches, powerful machine learning algorithms (called non-parametric or black-box modeling) and decades of global experience in process mechanistic (called parametric or white-box modeling).
Hereby, the advantages of each individual modeling technique are exploited, and the respective shortcomings are compensated, resulting in a superior hybrid model with outstanding generalization abilities. In addition, by utilizing this model structure, we facilitate the development of additional beneficial tools.
What is intensified DoE?
To reduce the experimental workload during the development and optimization steps of a bioprocess is of high value and interest. Therefore, we utilize a novel fermentation technique, in which process condition setpoints are shifted during the process. This enables the investigation of how the cells react to dynamic changes and to characterize multiple settings of the design space within one single bioprocess. Due to this bioprocess intensification, a certain design space is characterized by fewer experiments compared to solely utilizing bioprocesses with static process conditions.
Consecutively, the development of a hybrid model based on these intensified cultivations offers the additional advantage of incorporating enhanced process dynamics, improving the models' generalization ability. This emphasizes a combinatorial approach, using hybrid modeling and iDoE, to generate time-resolved process knowledge and simultaneously accelerate process characterization.
By utilizing a suitable hybrid model structure, using controllable modeling inputs, the application of a digital process twin is enabled. This powerful modeling tool can be used to simulate entire processes, based on the developed hybrid model, to understand your bioprocesses in the best possible way. Thereby, you can easily evaluate your process at any possible point in time, allowing you to investigate various process conditions and to identify process optima, which are not located at the end of the process but within.
The visualization of your process responses of interest via the digital process twin for the chosen process conditions in a time-resolved manner accelerates and simplifies optimization tasks and contributes to process understanding.
Hybrid models found their way to upstream and downstream processing. We successfully applied this approach to fed-batch cultivations, where a combination hybrid models, and intensified Design of Experiments led to increased process understand with less experiments. On the downstream part, investigations of tangential flow filtration were performed. Here, hybrid modeling yielded time-resolved descriptive and predictive models for different scales, membrane and protein types.