Synthetic Data Generation Using Causal Models for Injection Molding Processes

The generation of synthetic data is particularly interesting for areas such as industrial production, where the generation of new data is associated with high effort, costs and possible production downtimes. By generating additional data, potential problems associated with the use of small sample sizes when applying machine learning algorithms can be avoided. This allows machine learning-based approaches to be better utilized and processes to be trained more efficiently. In order to address the problem of small sample sizes and the time-consuming generation of additional data, a process model based on causal mechanisms is generated in this work, which can be used to generate new synthetic data that is based on the available observational data. To make this possible, a minimal model is generated, which can be trained using existing real data and used for further data generation. The generated synthetic data are tested on a real-life application use-case of an injection molding machine and evaluated using various utility metrics.