Flow mechanisms and machine learning-based formation optimization on ultrasonic-assisted friction stir channeling of cast aluminum alloys

Friction stir channeling (FSC) forms internal flow channels within single pass by extracting plasticized materials to the surface with a profiled tool, enabling monolithic cold-plate structures for electric-vehicle battery thermal management. Cast aluminum alloys used in battery housings, however, exhibit limited flowability that leads to irregular geometry and rough inner walls. We reported an ultrasonic-assisted FSC (UaFSC) route for ZL114 cast aluminum alloys that leverages acoustic softening to reduce flow stress, promote upward material flow, and regularize the channel. Rather than exhaustive orthogonal trials, a compact “random seed→neural network→adversarial refinement” workflow learned the multi-parameter process window and yielded high-accuracy predictions of rectangularity, width, height, and a surface-quality index. The model across validation sets achieved 100% accuracy for cover-surface grades and > 80% for geometric metrics. Pareto analysis showed ultrasound increases mean channel height by ∼26% and expanded feasible windows. A coupled Eulerian-Lagrangian finite-element model ascribed these improvements to reduced stress, lower temperature rise, and higher void fractions. Tracer-based kinematics revealed a periodic, probe-entrained flow in UaFSC that recovered wall-normal displacements and smoothed the advancing side, cutting inner-wall roughness. The results clarified the formation mechanisms and provided a data-efficient pathway to robust process design for compact thermal devices.