navROM

The navROM application calculates a Reduced Order Model using one of the methods POD or Isomap based on a set of parameterized input data (“snapshots”).

A ROM is the model of a complex process that has been simplified by applying mathematical methods in such a way that it is (much) easier to handle. A good ROM is characterized by the fact that the relevant properties of the simplified model match those of the complex process very well. To train the models, we need data sets, for example from CFD, computed for different parameter combinations. We then perform the model reduction using the POD or Isomap methods. The former uses a singular value decomposition, the latter identifies low-dimensional structures in high-dimensional data sets. In everyday industrial applications, a ROM is often practical: the reduced model can be used to compute predictions for new parameter combinations very quickly – spatially and for all variables. So instead of having to compute a new solution using numerical simulation, which can take hours or days, the solution is estimated within seconds by evaluating a ROM.

• field data (cell- or point-based) from simulation or experiment
• Parameterized problem, i.e. a specific parameter combination can be assigned to each snapshot
• Topologically equivalent networks, i.e. the networks on which the field data are available all have the same number of points and possess the identical connectivity – if necessary, this must be established as part of preprocessing (link à navROM, navConvert, navMap)

• Surfaces (.obj, .stl)
• VTK (.vtk, .vtp, .vtm)
• EnSight (.case)
• Femzip (.femzip) – requires license for Femzip*, which is not included in navROM.

• navROM is a command line application for a Linux operating system.
• Since for a lossless ROM calculation all input data must be loaded into the working memory at the same time, its size in everyday life usually limits the number of grid cells and variables that are available in ROM at the same time. You can estimate the memory requirement in bytes (exclusively for the ROM calculation) as follows: 4 / 8 bytes * K * N * (S + 3*V), where K corresponds to the number of snapshots, N to the number of grid cells, S to the number of scalar variables and V to the number of vector variables. Thus, a ROM with double precision (8 bytes), 100 snapshots, 2,000,000 grid cells, 3 scalar variables and one vector variable requires at least 9 GB for the calculation.

navROM generates a Reduced Order Model of the input data with the selected method, which can be evaluated with:

• navDesign
• NAVPACK-API