CAE Proxy Solution that speeds up simulations by 1000 times

basic information

■ Strengths of CAE Proxy Solutions - Ultra-fast computation: Proxy calculations using deep learning predict phenomena at a speed 1000 times faster than simulations. - Generation of distributions: Field quantities such as stress and flow velocity can be obtained as distributions, allowing results to be evaluated at arbitrary locations. - Learning from measured data: If sufficient measured data is available, a proxy model can be constructed without simulations. ■ Service Content We provide the know-how of deep learning necessary for building proxy models and answer fundamental questions that serve as hurdles to data utilization, such as: - How to use data for predictions - What features should be extracted for predictions - What models should be used to learn the data features ■ Reasons for Choosing ASTOM Among the plethora of AI services available, ASTOM is the only technology that specializes in CAE and can directly generate simulation results. Unlike just variations in numerical parameters such as material properties and boundary conditions, ASTOM's CAE proxy solution uniquely allows predictions based on shape changes.

Price information

You can approach the operation of the CAE agent solution in stages. Starting with the question, "Will the CAE agent solution be helpful for this issue?", we will respond to your individual requests regarding budget and duration. - Issue analysis: Free - Data consulting: From 500,000 yen - Construction of agent models: From 1,500,000 yen - System development: From 2,000,000 yen - Support: From 500,000 yen

Delivery Time

Applications/Examples of results

■ Surrogate model for predicting steady-state stress in structural analysis Replaces the calculation of steady-state stress under tensile load on a perforated aluminum plate, predicting the distribution and maximum value of von Mises stress. ■ Surrogate model for predicting unsteady vortices in fluid analysis Predicts the time-varying distribution of Kármán vortices within piping. ■ Surrogate model for predicting rainfall from measured data Trains on satellite global precipitation map GSMaP to predict rainfall one hour later across all of Japan.

Detailed information

• 

  ■ Use Cases for CAE Traditional simulations often involve significant time and effort in mesh creation and numerical calculations when using finite element methods or finite volume methods, which can be a source of frustration. With a CAE proxy solution, calculations can be completed in no time by directly connecting CAD input and result output. The trial and error of a large number of parameters becomes manageable, paving the way for real-time use of CAE that was previously given up on.

• 

  ■ Why Deep Learning? In CAE, humans create rules that connect data (input shapes and material properties) to answers (stress distribution and flow fields) for simulations. However, creating these rules is the biggest challenge. The CAE surrogate solution using deep learning not only automates this rule-making process but also demonstrates high expressiveness for nonlinear responses compared to conventional surrogate modeling methods such as response surface methodology and model order reduction, excelling in multivariate predictions of complex phenomena.

• 

  ■ Surrogate model for predicting steady-state stress in structural analysis Predicts the distribution and maximum value of von Mises stress when a perforated aluminum plate is pulled. Compared to the finite element method, the CAE surrogate solution achieves a speedup of 1400 times while predicting von Mises stress with an error of only a few MPa.

• 

  ■ Surrogate model for predicting unsteady vortices in fluid analysis It predicts the distribution of Kármán vortices in piping as a time-varying fluid problem. The CAE surrogate solution reproduces the generation of vortex streets in the pipe at a speed 660 times faster than simulations using the finite volume method.

• 

  ■ Surrogate model for predicting rainfall from measured data Using measured data directly, we predict the trends of rainfall around Japan. The surrogate model can predict the rainfall distribution one hour later in 0.1 seconds based on rainfall images from the past five hours and monthly information, reproducing the behavior of rain clouds without data on temperature, pressure, or wind direction.