A machine learning-driven formation design framework for variable-width blade tip repair process

Ensuring formation consistency at different positions is critical for single-track conformal repair on variable-width thin-walled blade tips. The novelty of this study lies in developing a machine learning-driven framework for formation design across sections of varying widths to maintain constant deposition height and melting depth. The framework comprises two core modules: a formation boundary exploration module to find the limiting values of deposition dimensions for a specific target width in the formation space using Particle Swarm Boundary-Exploration (PSBE) algorithm; and a parameter-geometry mapping module to establish the transformation relationship for solver particles in PSBE between parameters and the formation using a Deep Neural Network (DNN) model. The global formation design boundary is finally determined by the intersection of the boundaries corresponding to various target widths derived from the blade tip geometry. The framework was experimentally validated and the results indicated that formation combinations within the boundary showed low average errors of 0.058 mm (2.90%) in widths, 0.020 mm (8.45%) in heights and 0.025 mm (9.05%) in melting depth, respectively. Based on this framework, a variable-width blade tip repair process workflow from blade geometry to formation design and ultimately to process parameters was built, and successfully achieved conformal and defect-free sample repair.