AI-driven tool life prediction with LLM-based interpretation of vibration and surface roughness in Haynes 230 turning

Abstract

The machining process of Haynes 230 superalloy is challenging due to its high strength and thermal resistance, leading to tool wear and instability. This research presents a hybrid artificial intelligence (AI) framework for managing tool life through machine learning (ML), multiobjective optimization, and large language models (LLMs). Experiments captured data on vibration, surface roughness (Ra), cutting forces, and temperature, revealing that vibration over 65 Hz and surface roughness (Ra) exceeding 1.4 µm indicate rapid tool wear. Various regression and classification models provided accurate estimates of tool life and wear state. A Multi-Objective Particle Swarm Optimization (MOPSO) approach identified optimal machining conditions to improve tool longevity while minimizing vibration and surface roughness. SHapley Additive exPlanations (SHAP) analysis underscored the effects of these variables, and integration with the Falcon LLM translated finding Bidirectional Mean Absolute Error (Bi-long short-term memory)—AI into operator-friendly recommendations. This framework supports predictive maintenance and can be incorporated into robotic machining cells and digital twin environments, promoting intelligent manufacturing systems that enhance productivity and resilience in aerospace and power generation industries.