Application of Artificial Intelligence to Tool Path Optimisation in CNC Milling and Turning

Toolpath optimisation is an important determinant of CNC machining performance because it affects machining time, surface quality, tool life, energy consumption and production cost. This structured review examines how artificial intelligence (AI) is used to generate and optimise toolpaths in CNC milling and turning. Applied methods are classified as direct when they modify toolpath geometry, toolpath patterns, operation sequences, non-cutting moves or machining strategies, and as indirect when they support toolpath decisions through workpiece feature recognition, machining outcome prediction, adaptive control or digital twins. The review also distinguishes toolpath variables from cutting parameters and tool-selection variables, which are included when they directly constrain path feasibility or are jointly optimised with the toolpath. Recent studies illustrate the potential of reinforcement learning for complex-cavity milling and meta-reinforcement learning for five-axis flank milling, while machine-learning-based strategy prediction supports the selection among candidate toolpaths. Reported improvements in machining efficiency are study-specific. For example, multi-objective genetic optimisation of non-cutting moves shortened the non-cutting toolpath by about 30% and the machining cycle time by about 7% relative to the original toolpath. Such results cannot be generalised without considering workpiece geometry, baseline strategies and machining conditions. Evidence remains limited for integrated optimisation of path geometry, tool orientation, tool dimensions, cutting conditions, energy consumption and surface quality under industrial conditions. Key barriers include data availability, computational cost, manufacturability and collision constraints, explainability, and the transfer of results from simulation to physical production. The review concludes that AI can improve selected toolpath decisions, but systematic comparative validation against well-tuned conventional CAM strategies is neceall R2.1 essary before broader performance benefits can be established.