1. Increasing the prediction quality of artificial neural network–based optimisation approaches for the sheet moulding compound compression moulding process by means of targeted expansion of training data.
- Author
-
Hopmann, Christian, Hadenfeldt, Marie, Neuhaus, Jonas, and Foerges, Dominik
- Subjects
- *
STANDARD deviations , *PROCESS optimization , *ARTIFICIAL neural networks , *LARGE deviations (Mathematics) , *COMPUTER simulation - Abstract
The resulting fibre orientation when pressing Sheet Moulding Compound (SMC) is essentially dependent on the position and geometry of the resin mat in the mould. It is desirable to adapt the fibre orientation as best as possible to best withstand the applied load. The relationship between the position and geometry of the resin mat and the deflection of the final part under a specific load can be mapped using a numerical simulation; nevertheless, an optimisation is not feasible based on the simulation in an economically reasonable time frame due to the required computing time. For this reason, a metamodel based on an ensemble of artificial neural networks (ANN) was created on the basis of the simulation data, and an optimisation based on these ANN was set up. However, the investigations show that the minimal deflections found on the basis of the metamodel through optimisation have large deviations compared to the simulation. This leads to the hypothesis that the applied optimisation finds weak points of the metamodel instead of actual minima. From this hypothesis, an extension approach of the data set on which the metamodel is based by the weak points found is derived and implemented. This is done in an iterative process. The results show that the goal of increasing the prediction quality of the optimisation approach is achieved by the developed approach. The root mean square error (RMSE) could be reduced from the original deflection of 25 to 7 mm, which corresponds to a reduction of 72%. It is shown that the presented approach is successful in increasing prediction quality; nevertheless, further measures are needed to reach decent prediction quality. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF