“We were working on a project where we needed complex numerical simulations of river flow into hydropower plant to analyze if the structure is efficient for energy generation from the river profile, and also see what would happen if there was a potential flood. It took us 4 years to nail down and fine-tune the model in expensive commercial software, with many long simulation iterations for every river setup we wanted to analyze. We spent months in-between on complex geometry meshing.”

— Pavol Breza, Founder & Director of Engineering, RFB s.r.o.

Figure below shows an aerial shot of the hydropower plant (a); rendered inference from trained neural physics simulator of river flowing into hydropower plant, which took just 6 seconds to generate for each parameterized input variables setup configuration (b); and the plots from the training process within an Environment page, showcasing few of the loss functions that are created by constraining the newtork with physics, showing how well the neural network is able to capture each physical quantity (c):

Hydropower plant neural simulator

Simulator, parameterized for multiple river speeds and viscosities, was trained for 17 hours and 1M steps on a single A6000 GPU, which is available within Tier 3 of computing resources in Siml.ai platform. Below is a visual representation of the simulator in SIml.ai's node-based editor:

“In contrast, we managed to create, train and test using 6 AI-driven simulators that leveraged PINNs and FNOs for flow analysis in just 3 weeks, from which oututs were comparable to the ones from expensive simulation software we used, with 96% cost reduction. Each simulation iteration for the analysis of different conditions went from 100 compute hours to 6 compute seconds!”

— Pavol Breza, Founder & Director of Engineering, RFB s.r.o.

In Siml.ai, we provide ways to quickly check uploaded CAD geometry (a). For the visualization and post-processing part, It's possible to opt-out of using our Simulation Studio, and instead do quick inference runs through integrated, web-based Inferencer app, form which you can then download the generated outputs in .VTI format and use visualization software of your choice, like Paraview (b).

Hydropower plant simulation

Want to use Siml.ai for your next project, but don't know where to start? Make sure to check out our documentation, get up-to-speed by going through videos in Siml.ai University, and make sure to try the app for free. No installation, special hardware, or credit card is required.