Accurately predicting flow-induced vibrations (FIV) in high-precision lithography equipment is challenging, as standard Reynolds-Averaged Navier-Stokes (RANS) simulations fail to capture the driving turbulent fluctuations.
This work evaluates the accuracy and uncertainty of a Stochastic Noise Generation and Radiation (SNGR) pipeline that reconstructs time-resolved, divergence-free velocity fluctuation fields from RANS statistics to improve in FIV and aero-acoustic assessments. Therefore, helping to ensure nanometer-scale manufacturing precision. The MATLAB based implementation ingests CGNS/HDF5 solver output, assembles modal Fourier fields from a prescribed energy spectrum, enforces incompressibility, applies anisotropic tensor mapping to match Reynolds stresses, and offers both a time-marching (single time-loop) and an ensemble snapshot mode. Validation is performed against canonical
Direct Numerical Simulation (DNS) reference data for turbulent channel flow (Lee & Moser, up to Reτ ≈ 5200)
and a benchmark backward-facing step, and comparisons are made with RANS (StarCCM+) results. A modular MATLAB pipeline automates diagnostics and produces slice-wise statistics (RMSE, absolute/relative errors), spatial heatmaps, PSDs, and GIF visualizations. Results show that SNGR successfully reproduces spectral content and the
spatial distribution of turbulence, yielding instantaneous fields that correlate with RANS TKE maps. However, the analysis reveals that discrepancies - particularly concentrated in near-wall amplitudes, outer-layer overshoots and the smallest resolved scales - are primarily inherited from biases in the initial RANS model. The report quantifies how these RANS biases, domain and boundary conditions choices and numeric resolution propagate into SNGR
reconstructions, and it recommends practical diagnostics and sensitivity checks for the robust industrial application of the SNGR method.

In contemporary world, integrated chips(ICs) play a vital role in how various systems function. These are manufactured by lithography machines to print intricate features on the silicon wafers with nanometer precision. As these use high energy radiations(Deep Ultra-violet(DUV) or Extreme Ultra-violet(EUV)), they generate high amount of heat, which needs to be dissipated via cooling circuits. As the circuit can contain bends, change in cross-sectional area, junctions, orifices; these can cause turbulent flow fluctutations. These fluctutations can lead to Flow-Induced Vibrations(FIV), which can affect the accuracy of these machines. Hence, in this thesis, an efficient method to compute the turbulent flow fluctuations is dealt in detail.

Generally, at ASML, experimental methods or numerical methods like Direct Numerical Simulation(DNS) or Large Eddy Simulation(LES) are used to compute turbulent fluctuations. However, these are time consuming, complex and computationally expensive. So, in this thesis a Reynolds Averaged Navier Stokes(RANS) based stochastic method has been developed from existing methods to generate turbulent velocity fluctuations for ASML applications. This method can account for multiple turbulence characteristics like spatial correlation, velocity-time correlation and anisotropy that are essential for computing FIV. In addition, a pressure Poisson equation solver has been developed to compute turbulent pressure fluctuations from velocity fluctuations. Further, the developed method has been validated for two academic test-cases, homogeneous isotropic turbulence(experiment) and turbulent channel flow(DNS). The method was found to generate turbulent fluctuations adhering to the energy spectrum imposed; the exponential time correlation applied and account for anisotropy based on the Reynolds Stress Tensor given as input. Finally, the method has been applied to a test-case, sharp-bend and compared against LES results. It was found that the method agrees well to the energy spectrum imposed, however it deviates from the LES energy spectrum. Nonetheless, the capability of turbulent fluctuation generation from RANS at lesser computational effort than LES/DNS has been demonstrated, even though there can be improvements made to the energy spectrum given as input to the developed method.