An intelligent decision framework for multi-stage milling parameters to sustainable manufacturing

Annals of Forest Science · Springer Nature · 2026-11-13 · DOI: 10.1007/s10845-026-03019-7

Аннотация

In mechanical surface machining, current investigation primarily focuses on single discrete stage in milling, grinding or planing. However, a workpiece that satisfies the standard typically requires continuous machining involving multi-stage, and research in this domain is relatively scarce. Additionally, there is a lack of research on applying intelligence in the field of multi-stage continuous machining. This necessitates an intelligent framework that can integrate multiple machining stages and the intelligent decision method for the entire machining process. Accordingly, this paper first proposes a machining method for multi-stage surface milling, and integrates it with the intelligent decision framework of process parameters including data acquisition, intelligent judgment and automatic decision modules. This framework obtains different parameter sets through investigation and experimentation channels, completed machine tool state recognition, established the correlation model between objective parameters and multi-stage milling parameters, and achieved balanced optimization considering energy consumption, efficiency, and quality. Specifically, we employ a bidirectional gated recurrent unit (BiGRU) hybrid network model combined with a self-attention mechanism and use the Crested Porcupine Optimizer (CPO) to optimize its hyperparameters. Subsequently, the interpretability analysis is conducted based on the above correlation model using SHApley Additive exPlanations (SHAP). Eventually, the decision process touched on is implemented through the combination of the PESA-II multi-objective optimization algorithm and the CRITIC-VIKOR evaluation model to identify the optimal solution. In this paper, 7075 aluminum alloy is taken as an example for plane milling and the energy consumption, machining time, and surface roughness are improved by 20.9%, 15.57%, and 63.7% in the experiment, respectively, validating the correctness and effectiveness of the proposed optimization method.

Читать у издателя →

Похожие статьи