Grid conical shell milling gets a physics chain: predict cutting force first, then compute deflection, with accuracy above 93% on all four test sets
Synopsis
When a thin-walled grid conical shell is milled, the rib layout makes local stiffness jump from tool position to tool position, and the cutting force pushes the workpiece into an elastic retreat along the shell normal, so the actual wall thickness falls short of the theoretical depth of cut. The new route treats finite-element simulation as the source domain and real milling experiments as the target domain, uses transfer learning to predict the normal cutting force first, then feeds that load into an equivalent-stiffness analytical model to compute deflection. Across four test sets the method reached 93.05% to 95.27% prediction accuracy, above the CNN, PINN and TL comparison models.
Interpretation
Deformation prediction for grid conical shell milling now predicts the normal cutting force first and computes deflection with an analytical model, instead of letting a network map tool-position features straight to a deformation value. Earlier routes split in two: analytical methods mostly target thin plates, frame beams or general thin-walled parts and adapt poorly to the local stiffness jumps caused by periodic ribs, while finite-element methods carry high modeling cost and strong parameter dependence and struggle to cover many tool positions and working conditions quickly. In the grid conical shell milling experiment the method scored 93.05%, 95.27%, 94.98% and 94.12% under training patterns M1 to M4, against 89.85% to 91.56% for CNN, 91.47% to 92.55% for PINN and 90.69% to 92.35% for TL on the same test sets.
The deformation calculation splits in two: a transfer-learning network handles the cutting-force mapping under complex conditions, and an analytical model handles shell deflection under that load. The network extracts shared machining-state features with a shared encoder, aligns simulation and experimental sample distributions by maximum mean discrepancy, then fine-tunes on a small set of measured cutting-force samples; on the analytical side the periodic rib-skin structure is treated as an orthotropic conical shell, with direction-dependent bending stiffness from unit-cell homogenization. Validation ran on a five-axis gantry machine on a 6061 aluminum alloy integrally spun conical blank with a 1100 mm large end, 800 mm small end and 500 mm height, 48 grid cells, and seven milling rounds taking wall thickness from 6 mm to a 2.5 mm rib residual at 0.5 mm depth per round; seven experiments yielded 5600 samples split into target-domain sets C1 to C4 of 1400 each, with source-domain sets C5 to C16 from multiple simulation rounds.
Perspective
The demonstrated scope is a 6061 aluminum alloy integrally spun thin-walled grid conical shell milled in seven rounds on a five-axis gantry machine, with normal cutting force as the main intermediate variable and accuracy judged by predicted-versus-measured deflection within plus or minus 0.04 mm. The method targets milling error prediction and process parameter optimization for thin-walled stiffened curved components, so readers can judge how close their own geometry and process conditions are.
Real milling samples in the target domain are limited; the method relies on the finite-element source domain for wide working-condition coverage and on a small set of measured samples to correct real-condition deviation. Accuracy uses a plus or minus 0.04 mm deviation threshold, and the four test sets land between 93.05% and 95.27%, with no reported behavior under stricter tolerances.