Pure physics modeling is not enough for particularly complex applications. Once the embeddable model is tested, an additional component can be added to the architecture if the required accuracy is not achieved: the discrepancy model.

Our proprietary neural network has several benefits, designed to guarantee stability:

1. The user defines 4 simple hyperparameters, and the network can be trained with a single mouse click.

2. It ensures high accuracy under real-time operating variability, due to nonlinear effects, time-varying boundary conditions, or unmodelled effects. 

3. Captures unmodelled dynamics and disturbances not configured in the FEM problem definition.

- Physics-based calibration alone does not meet the required accuracy→We utilize a hybrid model: physics + AI, that delivers accurate performance in real-time, even in the most dynamic environments.

- Difficulty in generating generalized models→Our technology ensures high accuracy under real-time operating variability, due to nonlinear effects, time-varying boundary conditions, and unmodelled dynamics and disturbances not configured in the FEM problem definition.