Simulation deployment

Simulation and Deployment in Twin Fabrica

The simulation and deployment phase is where the virtual thermal sensor is thoroughly tested and prepared for real-world use. This phase allows users to run simulations, fine-tune the model, and integrate it into external systems for final deployment.

Steps in Simulation and Deployment

Finalizing the Model:

The calibrated model is evaluated and finalized during the simulation step. This final version is ready to be exported and integrated into external systems.
Once the model is finalized, users can run simulations on new or existing test scenarios to validate its performance.

Simulating with Deliverable Models:

Simulations are carried out using deliverable models, which are based on the calibrated models. These deliverable models require minimal configuration before they are ready for simulation.
Twin Fabrica offers enhanced configurations for deliverable models, which may include state observers and discrepancy models. These elements provide additional flexibility and accuracy, allowing users to tailor the model for specific deployment needs.

Adding State Observers:

A state observer enhances the model by providing output feedback, improving the performance and robustness of the virtual thermal sensor.
The observer compares the model's output with real-world measurements. The difference, or residual, is used to adjust the model’s internal states in real time, ensuring that the sensor maintains accuracy even in dynamic or uncertain conditions.

Using Discrepancy Models:

Discrepancy models are AI-based techniques that correct the output of the virtual sensor when the physics-based model fails to capture certain dynamics.
Unlike state observers, discrepancy models rely on data-driven residuals to adapt to unpredictable or nonlinear system behaviors that are not fully described by physics.
These models are particularly useful when traditional physics-based modeling alone is not enough to capture the full complexity of the system.

Training the Discrepancy Model:

When a discrepancy model is added to the deliverable model, an additional training step is required.
This step involves fine-tuning the internal neural network that drives the discrepancy correction. Using experimental data, the discrepancy model learns to correct the residual error between the model’s output and the measured values, improving its accuracy and robustness.

With these steps, the simulation phase ensures that the virtual thermal sensor is accurate, robust, and ready for deployment in real-world applications.