基于RBF神经网络与NSGA-Ⅱ算法的渣浆泵多目标参数优化

    Multi-objective parameters optimization of centrifugal slurry pump based on RBF neural network and NSGA-Ⅱ genetic algorithm

    • 摘要: 由于渣浆泵普遍存在扬程低于设计扬程、效率低、磨损严重等问题,该文选取比转速为75的离心式渣浆泵为研究对象,运用商用CFD求解软件Flunet,选取RNG k-ε湍流模型与欧拉两相流模型对其内部流动进行计算。以离心式渣浆泵的效率、高效区作为优化目标,结合Plackeet-Burman筛选试验,将渣浆泵叶片的进口安放角、出口安放角与叶片包角作为优化变量。采用均匀试验设计安排样本空间,利用RBF神经网络拟合优化变量与优化目标间的映射关联,基于NSGA-Ⅱ遗传算法进行多目标寻优。针对优化所得的Pareto解集,选取其中效率最优个体和高效区最优个体与优化前初始模型进行对比:分析了上述3个个体的通过数值模拟得到的性能曲线之间的差异,得到效率最优与叶片进、出口安放角、叶片包角为21.76°、23.43°、145.56°,高效区最优时为19.38°、22.68°、116.71°。通过试验验证,优化后个体性能得到显著提升,效率最优个体的效率较初始个体的效率提高了3.81%,高效区最优个体较初始个体高效区范围提高了4.33%。给出并分析了上述3个个体在叶轮流道中间剖面上固相相对速度矢量及湍动能分布、叶片工作面、叶轮后盖板的固相浓度分布差异。优化结果表明,该优化方法使叶轮的水力特性得到改善,提高了离心式渣浆泵的性能。

       

      Abstract: With the rapid development of national industry and mining, slurry pump is widely applied in transportation of two-phase flow as a sort of practical and reliable fluid machinery. Due to that the medium of centrifugal slurry pumps transport is fluid-solid two-phase medium, there is great difference for internal flow field between centrifugal slurry pumps and clean water pumps. Therefore, the design of the former is more complex and its design theory and method are not perfect now. In terms of performance, the main criticisms are its working head lower than design head, low efficiency and severe wear. So, it is a kind of equipment whose performance should be promoted rapidly in modern safe and effective industrial production. Thus, the optimal design of centrifugal slurry pumps is very meaningful for improving its performance. A centrifugal slurry pump with a specific speed of 75 was chosen as the research object. With the commercial CFD (computational fluid dynamics) software Fluent, RNG k-ε turbulence model and Eulerian two-phase flow model were selected to calculate its internal flow. The efficiency and the high efficiency region of the centrifugal slurry pump were set as the optimization goal. Design Expert 8.0.5b was used to make Plackett-Burman screening experimental design to pick out 3 structural parameters from 12 structural parameters of model pump, which were slurry pump blade inlet angle, outlet angle and wrap angle, set as the optimization variables, because too many structural parameters may affect the high efficiency area and the highest efficiency of centrifugal slurry pump. The 37-level uniform experiment was finished and the training and testing samples of RBF neural network were established. RBF (radial basis function) neural network was used to fit the relationship between the variables and objectives, which could be applied in NSGA-Ⅱ algorithm to get the Pareto optimal solution. Aiming at the result of the optional Pareto solution set, the optimal efficiency individual and high efficiency region individual were selected to compare with the initial model that was not optimized: The difference of external characteristic curves, distribution of absolute pressure at middle section of impeller runner and volute runner, relative velocity vector of fluid and solid phase, distribution of turbulence kinetic energy, distribution of solid concentration of pressure blades, blades back, front shroud and back shroud at middle section of impeller runner were all compared between the 2 extreme value individuals and initial individual. According to the analysis results of the differences between the performance curves of the above 3 individuals, the obtained optimal efficiency blade inlet angle, outlet angle and wrap angle were 21.76°, 23.43°, and 145.56°, respectively, and the high efficiency region blade inlet angle, outlet angle and wrap angle were 19.38°, 22.68°, and 116.71°, respectively. The experiment proved that the efficiency of the optimal individual was improved by 3.81% and the high efficiency region individual was improved by 4.33% compared with the original individual. The optimization results show that this optimization method improves the hydraulic characteristics of the impeller and the performance of centrifugal slurry pump.

       

    /

    返回文章
    返回