Innovative and Cognitive Production Technology and Systems, 16 - 18 July 2025
HomeQuality, Metrology & TestingPredicting the solidification time of low pressure die castings using geometric feature-based machine learning metamodels
Predicting the solidification time of low pressure die castings using geometric feature-based machine learning metamodels
Thank you for the presentation, Mr. Rosnitschek! Casting is one of the challenging manufacturing techniques to catch a pattern between consecutive products. What is your training time ? Additionally, why did not you shuffle the data? If training and test datasets would have included information about each design type, accuracy of model would probably increase.
Dear Gamze Kecibas, thank you very much for your question. The training times were about 80 mins. Indeed the data for the half torus and traverse link were shuffled.
As you mention, it is very difficult to catch patterns between consecutive products in casting. within the TO process, quite significant changes to the geometry can occur from one iteration to the other. In the worst case, this may lead to a “completely unknown geometry”. Accordingly, with the cantilever beam example, we wanted to test the transferability toward new or unknown structures.
Thank you for the presentation, Mr. Rosnitschek! Casting is one of the challenging manufacturing techniques to catch a pattern between consecutive products. What is your training time ? Additionally, why did not you shuffle the data? If training and test datasets would have included information about each design type, accuracy of model would probably increase.
Dear Gamze Kecibas, thank you very much for your question. The training times were about 80 mins. Indeed the data for the half torus and traverse link were shuffled.
As you mention, it is very difficult to catch patterns between consecutive products in casting. within the TO process, quite significant changes to the geometry can occur from one iteration to the other. In the worst case, this may lead to a “completely unknown geometry”. Accordingly, with the cantilever beam example, we wanted to test the transferability toward new or unknown structures.