A centrifugal compressor is a mechanical machine with purpose to convert kineticenergy from a rotating impeller wheel into the fluid medium by compressingit. One application involves supplying boost air pressure to downsized internalcombustion engines (ICE). This allows, for a given combustion chamber volume,more oxygen to the combustion process, which is key for an elevated energeticefficiency and reducing emissions. However, the centrifugal compressor is limitedat off-design operating conditions by the inception of flow instabilities causingrotating stall and/or surge. These instabilities appear at low flow rates andtypically leads to large vibrations and stress levels. Such instabilities affectthe operating life-time of the machine and are associated with significant noiselevels.The flow in centrifugal compressors is complex due to the presence of a widerange of temporal- and spatial-scales and flow instabilities. The success fromconverting basic technology into a working design depends on understandingthe flow instabilities at off-design operating conditions, which limit significantlythe performance of the compressor. Therefore, the thesis aims to elucidate theunderlying flow mechanisms leading to rotating stall and/or surge by means ofnumerical analysis. Such knowledge may allow improved centrifugal compressordesigns enabling them to operate more silent over a broader operating range.Centrifugal compressors may have complex shapes with a rotating partthat generate turbulent flow separation, shear-layers and wakes. These flowfeatures must be assessed if one wants to understand the interactions among theflow structures at different locations within the compressor. For high fidelityprediction of the complex flow field, the Large Eddy Simulation (LES) approachis employed, which enables capturing relevant flow-driven instabilities underoff-design conditions. The LES solution sensitivity to the grid resolution usedand to the time-step employed has been assessed. Available experimentaldata in terms of compressor performance parameters, time-averaged velocity,pressure data (time-averaged and spectra) were used for validation purposes.LES produces a substantial amount of temporal and spatial flow data. Thisnecessitates efficient post-processing and introduction of statistical averagingin order to extract useful information from the instantaneous chaotic data. Inthe thesis, flow mode decomposition techniques and statistical methods, suchas Fourier spectra analysis, Dynamic Mode Decomposition (DMD), ProperOrthogonal Decomposition (POD) and two-point correlations, respectively, areemployed. These methods allow quantifying large coherent flow structures atvfrequencies of interest. Among the main findings a dominant mode was foundassociated with surge, which is categorized as a filling and emptying processof the system as a whole. The computed LES data suggest that it is causedby substantial periodic oscillation of the impeller blade incidence flow angleleading to complete system flow reversal. The rotating stall flow mode occurringprior to surge and co-existing with it, was also captured. It shows rotating flowfeatures upstream of the impeller as well as in the diffuser. / <p>QC 20171117</p>