In order to make the pump as turbine (PAT) run efficiently and safely, a multidisciplinary optimization design method for PAT blade, which gives consideration to both the hydraulic and intensity performances, is proposed based on multidisciplinary feasibility (MDF) optimization strategy. This method includes blade parametric design, Latin Hypercube Sampling (LHS) experimental design, CFD technology, FEA technology, GA-BP neural network and NSGA-II algorithm. Specifically, a parameterized PAT blade with cubic non-uniform B-spline curve is adopted, and the control point of blade geometry is taken as the design variable. The LHS experimental design method obtains the sample points of training GA-BP neural network in the design space of variables. The hydraulic performance of each sample point (including the hydraulic pressure load on the blade surface) and the strength performance analysis of blades are completed by CFD and FEA technology respectively. In order to save calculation time of the whole optimization design, the multi-disciplinary performance analysis of each sample in the optimization process is completed by single-coupling method. Then, GA-BP neural network is trained. Finally, the multi-disciplinary optimization design problem of PAT blade is solved by the optimization technology combining GA-BP neural network and NSGA-II algorithm. Based on this optimization method, the PAT blade is optimized and improved. The efficiency of the optimized PAT is improved by 1.71% and the maximum static stress on the blade is reduced by 7.98%, which shows that this method is feasible.

为了使泵透机（PAT）高效、安全运行，基于多学科可行性（MDF）优化策略，提出了一种兼顾水力性能和强度性能的PAT叶片多学科优化设计方法。该方法包括叶片参数化设计、拉丁超立方采样（LHS）实验设计、CFD技术、FEA技术、GA-BP神经网络和NSGA-II算法。具体来说，采用三次非均匀B样条曲线的参数化PAT叶片，并以叶片几何控制点作为设计变量。LHS实验设计方法在变量的设计空间中获取训练GA-BP神经网络的样本点。分别通过CFD和FEA技术完成各采样点的水力性能（包括叶片表面的水压载荷）和叶片的强度性能分析。为了节省整个优化设计的计算时间，优化过程中各样本的多学科性能分析采用单耦合方法完成。然后，训练GA-BP神经网络。最后采用GA-BP神经网络与NSGA-II算法相结合的优化技术解决了PAT叶片的多学科优化设计问题。基于该优化方法，对PAT叶片进行了优化和改进。优化后的PAT效率提高了1.71%，叶片最大静应力降低了7.98%，表明该方法是可行的。