TY - THES

T1 - Adaptive Heavy End Control Optimization

AU - Novak, Philipp

N1 - embargoed until 31-01-2029

PY - 2024

Y1 - 2024

N2 - The optimization of production processes using machine learning models is an important step towards more efficient and cost-effective production processes. The aim of this thesis is to formulate a strategy for the adaptive optimisation of parameters for Heavy End Control (HEC) based on input parameters for a specified minimum target length of the thickened tail of the tube. For this purpose, various machine learning regression models and a feed forward neural network are used, which are trained and tested with a provided data set. These models are tested for plausibility and evaluated by predicting the parameters for two different thickened tail lengths of two different materials. The Random Forest model achieves the best results in predicting the input parameters for small and medium lengths of the thickened tail based on known values. The feed forward neural network also shows a good prediction for some parameters, such as the number of scaffolds used and the stress factor. Subsequently, the median of the distribution of the values for temperature and mean wall thickness before the stretch reducing block (SRB) are selected, with which the models generate the optimised parameter values for a specified target length of the thickened tail. The random forest model and the feed forward neural network are used for this purpose. These results will be verified experimentally on the system in the future.

AB - The optimization of production processes using machine learning models is an important step towards more efficient and cost-effective production processes. The aim of this thesis is to formulate a strategy for the adaptive optimisation of parameters for Heavy End Control (HEC) based on input parameters for a specified minimum target length of the thickened tail of the tube. For this purpose, various machine learning regression models and a feed forward neural network are used, which are trained and tested with a provided data set. These models are tested for plausibility and evaluated by predicting the parameters for two different thickened tail lengths of two different materials. The Random Forest model achieves the best results in predicting the input parameters for small and medium lengths of the thickened tail based on known values. The feed forward neural network also shows a good prediction for some parameters, such as the number of scaffolds used and the stress factor. Subsequently, the median of the distribution of the values for temperature and mean wall thickness before the stretch reducing block (SRB) are selected, with which the models generate the optimised parameter values for a specified target length of the thickened tail. The random forest model and the feed forward neural network are used for this purpose. These results will be verified experimentally on the system in the future.

KW - Heavy End Control

KW - Streck-Reduzierwalzwerk

KW - Maschinelles Lernen

KW - Random Forest

KW - Neurales Netz

KW - K-Nearest Neighbor

KW - Polynominale Regression

KW - Regression

KW - Heavy End Control

KW - Stretch Reducing Block

KW - Machine Learning

KW - Random Forest

KW - Neural Network

KW - K-Nearest Neighbor

KW - Polynomial Regression

KW - Regression

M3 - Master's Thesis

ER -