Research on modeling of machining processes and optimization of process parameters for intelligent manufacturing

Abstract

Traditional machining process models often suffer from limited accuracy, while parameter optimization still depends heavily on operator experience and is difficult to adapt to complex working conditions with conflicting objectives. This paper proposes an algorithm-oriented framework for smart machining that integrates finite-element process physics, a CNN-LSTM surrogate, and a digital-twin-enabled multi-objective optimizer. The physical model captures thermo-mechanical coupling, whereas the data-driven surrogate learns temporal process signatures from multi-source sensor streams. A physics-guided fusion mechanism is then used to correct prediction bias and quantify uncertainty for cutting force, cutting temperature, and surface roughness. On top of the hybrid predictor, an improved NSGA-III algorithm is embedded into a digital twin for closed-loop parameter adjustment with objectives of machining accuracy, efficiency, energy consumption, and tool life. Two representative cases, i.e., titanium-alloy blade milling for aero-engine components and gear turning for automotive transmission parts, are used for algorithmic verification. The proposed framework reduces the average prediction error to 2.1%, improves overall prediction accuracy by about 45% relative to single-model baselines, and keeps the decision deviation rate below 8%. Compared with conventional genetic and particle-swarm optimizers, the proposed method yields faster convergence and a better-distributed Pareto set, demonstrating its value for complex-part machining under intelligent manufacturing environments.