STOCHOS Flow Use Cases

Engineering intelligence applied to real thermal, CFD, and coupled thermo-structural problems.

E-Motor Cooling

Thermal Usecases

Long simulation times: Each thermal CFD simulation of the cooling channel takes 6–10 hours, severely limiting the number of geometries that can be tested during design.
Complex thermal behavior: Small changes in channel geometry strongly affect local flow and heat transfer, making it difficult to identify optimal designs through traditional simulation alone.

34 different simulations with changing cooling channel geometries have been created to train a DIM-GP model to predict temperature field
CADFEM AI model accurately predicts the temperature field in less than 20 seconds for a new design
The mean absolute error is 0.86 for 4 test designs with an R² of 0.99
Training time 1h 20 minutes on GPU

Massive time savings: The AI model predicts the full temperature distribution in about 20 seconds, reducing simulation time by several orders of magnitude.
Faster design optimization: Rapid predictions enable exploring and optimizing hundreds of cooling channel geometries efficiently, accelerating thermal management design for the e-motor.

CFD Usecases

CFD-based multiphase mixing tank simulations are computationally expensive and time-consuming due to transient flow and VOF calculations.
Predicting pressure, velocity, and phase distribution for multiple operating conditions requires significant computational resources.

3D transient multiphase CFD simulations were performed in ANSYS Fluent for different combinations of stirrer height, tank radius, rotational speed, and mass flow rate.
Pressure, velocity, and VoF data from 98 design combinations were extracted and converted into .vtu datasets for training.
A CADFEM AI graph regression model was trained using geometry coordinates, node features, and global operating parameters as inputs.
The trained model rapidly predicted node-wise pressure, velocity, and VoF fields for unseen mixing tank configurations and achieved MAE 0.086.

Reduces the need for repeated transient multiphase CFD simulations.
Accelerates mixing tank design evaluation and flow analysis.
Enables fast prediction of pressure, velocity, and phase distribution with limited training data.
Significantly reduces computational cost while maintaining acceptable prediction accuracy.

Thermal Usecases

Conventional thermo-structural analysis requires sequential thermal and structural FEA simulations, resulting in high computational cost and long turnaround times.
Evaluating multiple operating conditions requires repeated thermal analyses followed by stress calculations.
Design exploration and optimization become time-consuming due to the large number of coupled simulations.

Developed a thermal AI model that predicts the oven inner-case temperature field directly from JSON inputs containing the outer surface film coefficient and inner wall temperature.
Generated thermal simulation datasets across multiple operating conditions and trained the model across 19 datasets to predict node-wise temperature distributions, achieving an R² score of 0.91 on unseen test cases.
Used the AI-predicted temperature field as input to DIM-GP graph neural network model to predict node-wise von Mises stress distributions, achieving an R² score of 0.99 on unseen test cases.
Established a fully AI-driven thermo-structural workflow capable of rapidly predicting both temperature and stress fields without performing repeated thermal and structural FEA simulations.

Significantly reduces computational cost and simulation turnaround time.
Enables rapid prediction of temperature and stress fields directly from operating boundary conditions.
Accelerates design evaluation and optimization while maintaining high prediction accuracy.