Calibration workflow

Overview of Calibration Workflow in Twin Fabrica

Calibration is a crucial step in fine-tuning Reduced-Order Models (ROMs) with real-world data to ensure that the virtual sensor reflects the physical system's behavior accurately. This workflow ensures the virtual sensor aligns with physical measurements, improving its reliability and performance in simulations.

Steps in the Calibration Workflow

1. Understanding Calibration:
• Calibration fine-tunes ROMs using experimental data, aligning the model's parameters with real-world behavior.
• The calibration process involves key stages like sensor positioning, experimental measurements preprocessing, parameter range definition, and calibration instance assembly.
2. Sensor Positioning:
• In this step, virtual sensors are placed within the digital geometry, ensuring they are accurately aligned with the physical sensors used during experiments.
• Proper sensor placement is critical for achieving accurate calibration. Any misalignment between virtual and physical sensor positions can lead to inaccurate results and poor model calibration.
3. Experimental Measurements Preprocessing:
• After positioning the virtual sensors, experimental measurements are mapped to the corresponding virtual sensor locations.
• This preprocessing ensures that temperature readings and other signals collected from physical experiments are correctly linked to the virtual sensors.
4. Defining Parameter Ranges:
• Calibration recipes specify which physical parameters will be adjusted during the calibration process.
• Users can select specific parameters to calibrate and define their exploration ranges. These ranges guide Twin Fabrica in identifying the optimal parameter values that minimize errors when compared with experimental data.
• Properly configured recipes help speed up the calibration process and improve accuracy.
5. Calibration Instance Assembly:
• The calibration instance is the configuration that brings together all necessary elements: the exploration ranges, selected parameters, and reference test data.
• This instance acts as a container for the calibration process, allowing users to save and manage different calibration scenarios.
6. Hyperparameter Tuning:
• During calibration, users can adjust several hyperparameters related to the optimization algorithm. These settings help guide the calibration process and impact how quickly and accurately the parameters are tuned.
7. Calibration and Validation:
• Once the calibration instance is configured, the optimization process adjusts the model's parameters to align with experimental data.
• After calibration, it is crucial to run validation to assess the model's accuracy. The validation process tests the model against the entire dataset, not just the calibration data, and evaluates its generalization performance on unseen data.
• Statistical indicators such as mean error, standard deviation, and root mean square error are used to quantify the model's accuracy. The validation interface offers both visual and numerical insights to ensure the calibrated model reliably replicates experimental measurements before it is deployed.

By following these steps, you ensure that your virtual thermal sensor is calibrated accurately and performs reliably in real-world scenarios.