Surrogate models for predicting transient changes in temperature, von Mises stress and equivalent strain in a gas turbine blade were constructed based on conjugate heat transfer simulations and elastic/elastic-plastic-creep analyses by applying Gaussian Process regression with multiple kernel functions. The temperature, strain and stress predicted by the surrogate models agreed adequately with the computational simulations under arbitrary operating conditions. Despite initial errors or noisy inputs, the surrogate models predicted computational results with adequate accuracy.

View full abstract