Eventos Anais de eventos
ENCIT 2022
19th Brazilian Congress of Thermal Sciences and Engineering
LOW-COST IDENTIFICATION OF TWO-FACTOR INTERACTIONS THROUGH FRACTIONAL FACTORIAL METHOD APPLIED TO 1D MODEL OF A S-CO2 CENTRIFUGAL COMPRESSOR
Submission Author:
Elóy Esteves Gasparin , SP
Co-Authors:
Elóy Esteves Gasparin, Gabriel Inácio Rodrigues Silva, Daniel Dezan, Fabio Saltara, Paulo Eduardo Batista de Mello, Jurandir Itizo Yanagihara, Leandro Salviano
Presenter: Elóy Esteves Gasparin
doi://10.26678/ABCM.ENCIT2022.CIT22-0051
Abstract
The Sensitivity Analysis (SA) of high-dimensional and computational expensive models is able to provide important insights into model behavior. The so-called screening methods are known for their low-cost and good proxy for main effects identification, which is essential for SA of computationally expensive models, such as the Computational Fluid Dynamic (CFD) model of centrifugal compressors. If utilized with factor fixing purposes, it can significantly diminish computational effort for a subsequential optimization procedure. Morris’ screening method has been widely applied to engineering problems and provided excellent main effects prediction and robust factor ranking. Moreover, Morris’ Design of Experiment (DoE) demonstrated to provide good sample space coverage for Response Surface (RS) training. However, the two-factor interactions are not assessed by Morris’ method, since its one-factor-at-a-time nature is not fit for that purpose. Therefore, this work intended to assess if the Fractional Factorial (FF) screening method is able to provide a low-cost first assessment of two-factor interactions on computational expensive engineering models before the factor fixing is performed, which could avoid the fixing of an interacting factor. Thus, an already developed and computationally fast one-dimensional (1D) model of s-CO2 centrifugal compressor was considered herein, allowing the comparison with more expensive and robust SA methods for proper evaluation of this strategy. Axial height (ΔZ), number of blades (Zfb) and blade thickness (t) were considered as inputs. Overall, this combined new strategy was able to properly rank variables influence and its two-factor interactions, which reduces the computational cost of factor fixing SA of high-dimensional models (for 8 input variables this methodology is able to reduce the number of model runs in more than 80%).
Keywords
Sensitivity analysis, Fractional Factorial, Centrifugal Compressor, Supercritical CO2
