TY - GEN

T1 - Comparative study of data-driven and model-based real-time prediction during rubber curing process

AU - Frank, Tobias

AU - Bosselmann, Steffen

AU - Wielitzka, Mark

AU - Ortmaier, Tobias

N1 - Funding information: *This work was supported by the German Federal Ministry for Economic Affairs and Energy 1All Authors are with Institute of Mechatronic Systems, Gottfried Wilhelm Leibniz Universität Hannover, 30167 Hanover, Germany

PY - 2018/8/30

Y1 - 2018/8/30

N2 - In chemical processes model-based methods are commonly used to optimize and control distributed parameter systems. The curing process of rubber products has been modeled accurately during the last decades. Many studies have been carried out, optimizing process setpoints to achieve desired quality of the final product. However, these optimizations are performed offline concurrent with product design and thus, disturbances occurring during the heating process can not be considered properly. Therefore, we propose a realtime prediction during the heating process, for repetitively forecasting temperature curves inside the rubber until end of the cooling phase. Thus, mold and ambient temperature change can be taken into account, enabling adaption of heating duration. We use a twofold approach in which we compare a model-based prediction to a data-driven neural network time series forecasting. The evaluation is performed regarding computational effort and deviation from an accurate ground truth simulation. Both methods show promising results, but since the model has to be optimized regarding computation time, it lacks in accuracy. Contrary to the model, the neural network shows a significant shorter execution time and a better conformity.

AB - In chemical processes model-based methods are commonly used to optimize and control distributed parameter systems. The curing process of rubber products has been modeled accurately during the last decades. Many studies have been carried out, optimizing process setpoints to achieve desired quality of the final product. However, these optimizations are performed offline concurrent with product design and thus, disturbances occurring during the heating process can not be considered properly. Therefore, we propose a realtime prediction during the heating process, for repetitively forecasting temperature curves inside the rubber until end of the cooling phase. Thus, mold and ambient temperature change can be taken into account, enabling adaption of heating duration. We use a twofold approach in which we compare a model-based prediction to a data-driven neural network time series forecasting. The evaluation is performed regarding computational effort and deviation from an accurate ground truth simulation. Both methods show promising results, but since the model has to be optimized regarding computation time, it lacks in accuracy. Contrary to the model, the neural network shows a significant shorter execution time and a better conformity.

UR - http://www.scopus.com/inward/record.url?scp=85053877292&partnerID=8YFLogxK

U2 - 10.1109/aim.2018.8452261

DO - 10.1109/aim.2018.8452261

M3 - Conference contribution

AN - SCOPUS:85053877292

SN - 9781538618547

T3 - IEEE/ASME International Conference on Advanced Intelligent Mechatronics, AIM

SP - 164

EP - 169

BT - AIM 2018 - IEEE/ASME International Conference on Advanced Intelligent Mechatronics

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2018 IEEE/ASME International Conference on Advanced Intelligent Mechatronics, AIM 2018

Y2 - 9 July 2018 through 12 July 2018

ER -