Investigating Sand-Related Causes of Mold Collapse in the V-Process

Sammanfattning

This study, conducted in collaboration with a industrial steel foundry, investigates how variations in sand properties influence mold stability in the V‑process, a vacuum‑assisted molding technique used in steel casting. A mixed‑method approach was applied, combining analysis of historical data, laboratory permeability testing, personnel interviews, and data-driven permeability modeling to identify which sand characteristics are most strongly associated with collapse events. The historical data show that a marked reduction in fine‑particle content following a renovation of the sand‑handling system coincided with a substantial increase in collapse frequency. Although this correlation is clear at the process level, the limited resolution of the historical dataset and the variability of the measurements prevent definitive causal conclusions. Laboratory permeability tests revealed substantial inherent measurement variability, which in several cases was comparable to the differences observed between sand mixtures. This variability, together with specimen‑volume differences introduced by the constant‑mass preparation method, obscured the underlying relationship between particle size distribution and airflow behavior. To complement the experimental work, machine learning models were trained on the company’s historical sand data. These models showed only moderate predictive capability, largely constrained by the limited resolution and variability of the available measurements. Support vector machines showed the most stable performance, while k‑nearest neighbor and neural network models were more sensitive to data partitioning and exhibited higher variance. Overall, the findings suggest that maintaining stable a level of fine particles is important for process robustness, but that collapse behavior is influenced by multiple interacting factors. The results highlight the need for improved monitoring practices and more controlled data collection to support reliable process control in industrial V‑process production.