Fei Li

With the continuous increase in power demand in aerospace, shipping, electricity, and other industries, a series of manufacturing requirements such as high precision, complex structure, and thin wall have been put forward for gas turbines. Gas turbine blades are the key parts of the gas turbine. Their manufacturing accuracy directly affects the fuel economy of the gas turbine. Thus, how to improve the manufacturing accuracy of gas turbine blades has always been a hot research topic. In this study, we perform a quantitative study on the correlation between process parameters and the overall wax pattern shrinkage of gas turbine blades in the wax injection process. A prediction model based on a generalized regression neural network (GRNN) is developed with the newly defined cross-sectional features consisting of area, area ratio, and some discrete point deviations. In the qualitative analysis of the cross-sectional features, it is concluded that the highest accuracy of the wax pattern is obtained for the fourth group of experiments, which corresponds to a holding pressure of 18 bars, a holding time of 180 s, and an injection temperature of 62 °C. The prediction model is trained and tested based on small experimental data, resulting in an average RE of 1.5% for the area, an average RE of 0.58% for the area ratio, and a maximum MSE of less than 0.06 ${\mathrm{mm}}^2$for discrete point deviations. Experiments show that the GRNN prediction model constructed in this study is relatively accurate, which means that the shrinkage of the remaining major investment casting procedures can also be modeled and controlled separately to obtain turbine blades with higher accuracy.