Quantitative Impact of Process Expert Knowledge on Developing Pumping Models for Single-Screw Extruders Using Symbolic Regression
Sprache des Vortragstitels:
Englisch
Original Tagungtitel:
39th International Conference of the Polymer Processing Society
Sprache des Tagungstitel:
Englisch
Original Kurzfassung:
Reliable process models are a valuable asset in polymer extrusion to reduce downtimes and rejects, to improve process efficiency, and to accelerate the development of new screw designs. With ongoing progress in computational power and capacities, increasing attention is paid to data-based modeling techniques that infer predictions directly from the process data. Out of these, symbolic regression is an attractive option for process engineers, since it provides information in the form of ready-to-use analytical mathematical expressions. However, extensive workload for both data curation and model generation impedes obtaining generalized regression models with a large scope of validity. In the field of polymer extrusion, integrating domain-specific expert knowledge into the regression procedure is already known to support the search for accurate prediction models. To assess this benefit systematically and quantitatively, we developed symbolic regression models for the pressure-throughput characteristics of single-screw extruders. These models are derived from the same source of flow simulation data within a hybrid modeling approach by successively adding several levels of process expert knowledge: Initially, models are created (i) without any a priori included expert knowledge, followed by models that (ii) accept derived physical parameters as alternative input features, and models that (iii-iv) contain theoretical reference solutions at two degrees of sophistication. For each stage of expert knowledge integration, the regression models are evaluated in terms of their interpolation and extrapolation capability on unseen data, training time, and structural complexity. This case study demonstrates that including domain-specific expertise significantly enhances the performance-to-cost ratio of symbolic regression analyses for solving polymer extrusion problems.
Forschungs-