2D and steady analysis have been replaced by 3D and unsteady analysis because of dramatic improvements in computational environments. Analysis models that faithfully simulate actual products conventionally tend to be complicated and large scale. Therefore, the storage of analysis results has been increasing tremendously, and the dominant flow-field structure has been difficult to clarify. Data reduction by proper orthogonal decomposition (POD) is effective for simultaneously reducing the number of results of unsteady computational fluid dynamics (CFD) and for comprehending the dominant flow-field structure. However, only a few applications are used in the domain of industrial machinery. In this study, we applied the POD method to unsteady CFD results of a centrifugal blower in a vacuum cleaner and evaluated the benefits.
We extracted a time series of static pressure distribution in the diffuser from unsteady CFD results corresponding to one rotation of the impeller, applied the POD to these data, and compared the results of an experiment. The results were that the first six modes had a 99.4% contribution in terms of the L2 norm. In the scope of this research, the first six modes were revealed to surrogate the pressure fluctuation sufficiently. Also, the data storage was reduced to less than 2.0% of the original unsteady results. Next, frequency spectra were obtained by applying a discrete Fourier transform (DFT) to the expansion coefficients. The spectra of the expansion coefficients of the POD modes were found to have a peak near the blade passing frequencies (BPFs). The noise, the frequency of which is BPF, causes the majority of the noise that occurs in the diffuser. Therefore, we found by using both the POD and DFT that we could both reduce the dramatic data storage and extract the flow-field structures.
