Automatic design of an efficient forward-curved centrifugal blower using gradient-free approaches

This study presents the design of a high-efficiency forward-curved centrifugal blower (FCCB) using a fully automated optimization framework based on gradient-free methods. The framework integrates three open-source tools, including Salome for computer-aided design and mesh generation, OpenFOAM 8 for computational fluid dynamics (CFD) simulations, and Dakota for optimization, coupled through several Python scripts. Several gradient-free approaches, such as artificial neural network, efficient global optimization, Gaussian process (GP), and multivariate adaptive regression spline, are employed for the FCCB design. Eleven design variables related to both the impeller and the housing are selected to design the FCCB for maximum total efficiency ϵtot. In addition to the design process, a quadratic response surface is constructed to evaluate the influence of individual design variables and their interactions on the objective function, providing further insight into the FCCB design space. Turbulent flow within the FCCB is modeled using the Reynolds-averaged Navier–Stokes equations, closed with the k–ω shear stress transport model. In addition, unsteady simulations are performed for the optimized configuration to investigate the transient flow structures and provide further physical insight into the aerodynamic behavior of the FCCB. Based on the results, the GP model identified the most efficient FCCB design with ϵtot=59.1%, showing a negligible difference of 0.2% compared to CFD validation and a difference of 3.13% in the pressure rise compared to the experimental results at the design point. The flow field analysis demonstrates that the improved aerodynamic performance is primarily associated with a more uniform impeller discharge flow, a smoother circumferential pressure distribution, and a more organized flow field within the impeller–volute region.