Research on turbine blade temperature field reconstruction by coupling proper orthogonal decomposition and artificial neural network methods

The temperature of turbine blades is a critical factor influencing their performance and lifespan. However, a high cost is required for the traditional experimental and computational fluid dynamics (CFD) methods to obtain an accurate temperature field of turbine blades. In this paper, an effective temperature field reconstruction method that combines proper orthogonal decomposition (POD) with an artificial neural network (ANN) is proposed. Initially, POD is employed to reduce the dimensionality of the turbine blade temperature field data by extracting the dominant spatial modes and corresponding mode coefficients, thereby significantly reducing data complexity. Subsequently, an ANN with a feedforward neural network as its core is developed to predict the mode coefficients, facilitating rapid reconstruction of the temperature field. Comparative results indicate that the POD-ANN approach not only maintains high prediction accuracy—with a maximum relative error of 2.61% for fluid and solid fields and only 0.10% for the solid domain—but also dramatically reduces computation time, achieving a speedup of 793 223.2 relative to conventional CFD methods. This study, therefore, presents a robust and feasible technical approach for the rapid prediction and optimization of turbine blade temperature fields.