Multi-dimensional and multi-objective collaborative optimization of multi- blade centrifugal ventilator based on NGO-RBF

Abstract

The impeller is a crucial component in optimizing the aerodynamic performance of multi-blade centrifugal ventilation fans. However, incorporating variables into the parametric design of the impeller substantially elevates the complexity of the optimization problem. This paper proposes using Non-Uniform Rational B-Splines (NURBS) curves for blade parameterization and conducts a global multi-objective optimization to enhance its aerodynamic performance. In the optimization process, an effective refinement interval is determined through dimension reduction optimization, followed by the implementation of the Optimal Latin Hypercube Sampling (OLHS) design within this interval. Sample-based Computational Fluid Dynamics (CFD) results are utilized to train the Northern Goshawk Optimization Radial Basis Function (NGO-RBF) neural network. The optimization design objective is to maximize the flow rate (Q_v) and efficiency (η) of the ventilator. The neural network, in conjunction with the Non-dominated Sorting Genetic Algorithm II (NSGA-II), is utilized to derive the set of Pareto-optimal solutions within the optimization space. The aerodynamic performance parameters, η and Q_v, obtained from the CFD analysis of the equilibrium solution on the Pareto solution set, are 64.70% and 1.08 m³/s, respectively. The corresponding input parameters are K₁ = 5.7, K₂ = 10.64, K₃ = 8.36, and K₄ = 0.08. Compared to the original scheme, η is improved by 1.25% and Q_v by 8.7%, indicating a significant increase in efficiency and airflow, as well as a notable reduction in flow field losses inside the ventilator.