| description abstract | This paper developed a process for turbine tip clearance prediction and control considering performance degradation to address the contradiction between computational efficiency and computational accuracy. The developed process consists of an offline high-accuracy database establishment for tip clearance with performance degradation and an online fast tip clearance prediction and control using machine learning. For the former, the steady-state tip clearance is obtained by the calculations for the two-dimensional axisymmetric casing and disk deformations using the finite element method and the one-dimensional blade deformation using the engineering calculation method. The effects of performance degradation, including blade creep and turbine inlet temperature degradation are introduced to update the boundary conditions in gas path and initial clearance. For the latter, the multilayer perceptron is used to realize the fast tip clearance prediction. Considering the independence of component deformations, the tip clearance prediction is achieved by the component deformation predictions, which also reduces the dimension of input parameters for each prediction model and improves the prediction accuracy. Combining the above two parts, the tip clearance with performance degradation can be obtained within 0.00025 s/time, and the maximum absolute error is only 0.012 mm. In addition, with the help of the process, the optimized tip clearance control strategy can be obtained for the performance degradation states, which restores the tip clearance with a 17.66% increment to the initial state without performance degradation. This paper will provide a reference for the tip clearance prediction and control with small computation and high accuracy. | |
