Wheel Loader Simulation with romAI

Rajiv Dixit_21408
Rajiv Dixit_21408
Altair Employee
edited September 18 in Altair Exchange

Overview

System simulation can support the design process by estimating the loads and stresses on critical components ensuring the desired performance. Can Artificial Intelligence (AI) help us to speed-up the study by replacing the DEM system with a more time efficient but still accurate Reduced-Order Model?

In this demo problem, romAI is used to create a reduced order model of more expensive EDEM model. Once the ROM is available, MotionSolve simulations run faster, thus enabling users to run multiple iterations within a short time.

 

Pre-Requisite

The files used for this particular example is found in the attachments (Wheel_Loader_MS_EDEM_romAI_2024.zip).

Usage/Installation Instructions

Step 1: Create the EDEM model

Export the geometries interacting with the bulk material from MotionView, and import into EDEM. You can also do this by enabling the 'Coupling Server' on EDEM and adding the 'EDEM' subsystem in MotionView and selecting required graphics.

EDEMsubsystem.JPG

Generate the particles in EDEM and setup the motions.

If you are new to EDEM, check out EDEM Tutorials to learn more about how to setup an EDEM simulation.

edem_model.gif

 

Extract the data required from the simulation. Also make sure to extract the forces and the positions of the bucket. These data will be required for the romAI coupling. An example is shown here:

Step 2: Run multiple simulations for training data

Run multiple EDEM simulations with different velocities and rotation speeds at various times. This will be used as the training data for your reduced-order model in romAI.

In this example, 10 cases were run. You need to make sure that the cases cover a wide range of scenarios.

Xdisp.JPGZdisp.JPGAngle.JPG

Extract all the required data for all the simulations run.

Merge all the data into one csv file with the same one title row and all the data written one below the other without any spaces. You can use an Altair Compose script to achieve this.

Step 3: Create the Reduced Order Model

Altair TwinActivate is the tool used for creating the reduced order model. Go to Modeling->romAI Director.

Import the merged data file for training. Choose appropriate inputs for your reduced order model which will replace EDEM in an EDEM-MotionSolve coupled simulation, and instead use a romAI-MotionSolve coupling. Since the romAI model is replacing EDEM, it needs inputs like the positions and velocities of the equipments, and the romAI model will give the forces and torques as outputs. An example is shown here:

romAI.JPG

Choose appropriate number of hidden layers and the neurons in each layer along with the epochs for your reduced order model training. (This is not an easy choice, and trial-and-error is involved, or you can use the Auto-Exploration option for romAI to explore the right neural network parameters.)

Save your Reduced order Model.

Step 4: Create a ROM subsystem in MotionView

Create a ROM susbsystem, so that it can be used in TwinActivate. Make sure to have the outputs in this subsystem as the outputs from your reduced order model created in the above step.

To know more about creating ROM susbsystems in MotionView, please refer here: https://help.altair.com/hwsolvers/ms/topics/solvers/ms/couple_motionsolve_with_activate_r.htm

Step 5: Create a romAI MotionSolve coupled block in TwinActivate

In the palette browser of TwinActivate, go to Palettes->Activate->CoSimulation->MotionSolveSignals, and import the MotionView model created in Step 4.

Next, go to Palettes->romAI and import the romAI model created in Step 3.

Connect the two models in the appropriate manner. An example is shown here:

TwinActivate.JPG

Run the coupled simulation in TwinActivate by clicking Simulation->Run Simulation.

Step 6: Review Results

In the case of the wheel loader, the boom joint forces are of importance, and a comparison is drawn between the romAI-MotionSolve model and the EDEM-MotionSolve model. The difference observed in the maximum force was less than 7.4%.

Results_forces.gif

The VonMises stresses were also plotted for certain critical elements and compared between the two simulations. The difference in the maximum stress observed here was less than 1.3%.

Results_stress.gif

Post-Requisite

You can also find a detailed YouTube video on the workflow here: https://www.youtube.com/watch?v=aic1GX1YX-4