Interest in artificial intelligence (AI) has grown significantly in recent years, as reflected in the rapid development and widespread application of large language models, powerful image and pattern recognition techniques. Advances in algorithm development and accessibility, coupled with the availability of large amounts of data and the ever-increasing computing power of modern hardware, have driven the establishment of AI applications in numerous areas. These developments also present new opportunities and considerable potential in process engineering. In particular, machine learning methods offer the potential to overcome the limitations of purely mechanistic models.

While these models provide a physically sound description of processes, they are often only feasible with great effort in practice, as they require very detailed and often difficult-to-access process knowledge. In contrast, data-driven models can learn complex relationships and diverse interactions between input and output variables directly from measurement data. This enables highly complex processes to be mapped efficiently. However, as black box models, they have limitations, particularly with regard to their extrapolation capabilities.

This is where the interaction between machine learning and mechanistic models becomes important. Combining both approaches in hybrid models allows us to exploit their respective strengths and compensate for their weaknesses. Hybrid models are therefore a promising tool for designing, optimizing and automating process engineering processes. This work investigates the application of machine learning to filtration processes. It discusses the pros and cons of the presented approaches and highlights potential areas of application.