The Use of Machine Learning and Surrogate Models in Optimization Methods for Milling Parameter Selection: A Structured Review and Conceptual Framework

Abstract. A structured narrative review of machine-learning models, surrogate models, and optimization approaches for milling-parameter selection is presented. The review covers 71 sources, including 66 publications from 2020–2026 and five earlier studies used for background. The reviewed literature is analyzed with respect to data sources, model roles, predicted process responses, optimization formulations, validation practices, and limitations. Particular attention is given to the distinction between predictive models and surrogate models used within iterative parameter-search procedures, as well as between optimization methods that generate new parameter combinations and multi-criteria decision-making methods that rank predefined alternatives. The reviewed evidence shows that many models and optimization procedures are validated only within narrow experimental ranges, while fewer studies demonstrate a complete workflow linking representative data, model validation, meaningful objectives and technological constraints, parameter search, independent confirmation, and controlled model updating. Based on the reviewed evidence, a role-based conceptual framework is proposed for constrained productivity maximization, Pareto trade-off analysis, sequential optimization when new evaluations are expensive, and controlled updating from production data. The proposed framework provides a basis for subsequent numerical, CAM-based, laboratory, and industrial validation.