Micro Artificial Bee Colony Optimization Based On Surrogate Modeling And Its Application Study In Turbomachinery

Now the expense of numerical simulation and physical experiments remain expensive, especially in the optimal design of turbo-machinery based on CFD. To deal with this complex, time-consuming and error-prone process, surrogate models are often used to improve calculation efficiency. There are universal commercial software such as NUMECA, ISIGHT, etc. providing surrogate model optimization functions. However for specific engineering design problems, we still need to study open, customizable, modular and efficient surrogate optimization techniques to support the further overall optimal design. In this paper, we combined surrogate modeling techniques and bee colony intelligence, thus proposed a micro artificial bee colony (MABC) optimization method based on surrogate modeling for solving the optimal design problem of a mixed flow pump. The objective is to improve the hydraulic efficiency of the mixed flow pump.The main works of this paper are as follows:(1) In the aspect of optimization methods, we proposed a global micro artificial bee colony (MABC) algorithm. The MABC decreases the population size and re-initializes after last iteration based on ABC. This is because that solving optimization problems based on surrogate models requires fast iterations, and the samples obtained by the true model are limited which means the population size is small when the algoritnm is initializing. To verify the iteration speed and solving accuracy of MABC, We tested five benchmark functions using standard genetic algorithm, ABC and MABC. The testing results indicate the effectiveness of the MABC.(2) In the aspect of surrogate modeling, we proposed a surrogate modeling method based on radial basis function artificial neural network (RBF-ANN), and studied the problem of selecting kernel functions. Since the original real model is explicit, it’s hard to express it with a deterministic functional expression. We firstly proposed the general process of surrogate modeling, including experiment designs, choosing of surrogate models and evaluation strategies. We chose radial basis function artificial neural network (RBF-ANN) as our surrogate model, then selected a better model among RBF-ANNs with five different kernel functions and compared prediction results via an engineering example. The predictions suggest that the RBF-ANN performs better when multi-quadrics kernel function is applied. (3) Based on all above, aiming at a mixed flow pump, as an example, we optimized its flow components by using the methods mentioned above. First, we extracted parameters (the vane wrap angle and the deflection angle of the vane outlet relative to the volute inlet) from the mixed flow pump model, and investigated those factors’influence on hydraulic performance, namely hydraulic efficiency and head. Then after samples are obtained by engineering analysis, we applied the above RBF-ANN to approximate the original true model. At last, we used MABC to optimize the built RBF-ANN surrogate model. The calculation results indicate the MABC optimization method is applicable to the optimization design of the mixed flow pump.The proposed MABC optimization based on surrogate modelling will help in the development of computational intelligence theories, and it can provide reference for solving turbomachinery or other complex engineering problems.