A Neural Network-Based Method for Solving the Mass Flow Related Forward and Inverse Problems of Laminate Cooling Structures

Laminate cooling structures are extensively used in the design of turbine blades for aircraft engines. There are two types of mass flow related problems in the design of laminate cooling structures: the forward problem of predicting the mass flow rate of cooling air, and the inverse problem of designing the geometric parameters given the target mass flow rate. The present study constructed a dataset for neural network training and testing using numerical simulation methods. A neural network model was established to solve the forward problem of predicting the mass flow rate, with four hidden layers and 64 neurons per layer. The average relative error of the predicted mass flow rate was 1.46%. A solution method for the inverse problem based on predicting the diameter of the film cooling holes was proposed. Using this method, six sets of geometric parameters for the laminate cooling structure were designed under known target mass flow rate conditions, with a design time of 0.014 s. Numerical simulation verification showed that the maximum relative error and average relative error between the designed mass flow rate and the target value were 2.9% and 1.54%, respectively, demonstrating the practicality of the proposed method.