MACHINE LEARNING-BASED PREDICTION OF SURFACE QUALITY AND TOOL DEGRADATION IN TURNING OPERATIONS

Abstract

Turning operations remain fundamental to modern manufacturing because they enable the production of cylindrical and rotational components with controlled dimensional accuracy, geometric integrity, and surface finish. However, machining performance is strongly affected by nonlinear interactions among cutting speed, feed rate, depth of cut, workpiece material, tool geometry, cutting temperature, vibration, force, acoustic behavior, spindle load, and progressive tool degradation. Conventional surface-quality assessment frequently depends on post-process inspection, while tool replacement is often scheduled according to fixed intervals or operator experience. These approaches can increase inspection delay, unnecessary tool replacement, defective production, and unplanned downtime. This paper proposes a machine learning-based framework for predicting surface quality and tool degradation in turning operations through integrated process parameters and multi-sensor industrial data. The proposed methodology combines cutting speed, feed rate, depth of cut, machining duration, vibration, acoustic emission, spindle current, temperature, cuttingload indicators, and historical tool-state information to develop predictive models for surface-condition categories and progressive tool degradation. The framework includes experimental data acquisition, sensor synchronization, quality validation, preprocessing, contextual segmentation, feature extraction, feature selection, training-data construction, model development, crossvalidation, uncertainty analysis, and real-time inference. Multiple machine-learning approaches are considered for classification and predictive assessment, including tree-based learning, ensemble models, support vector techniques, and neural architectures. The methodology also introduces joint interpretation of surface-quality deterioration and toolcondition progression so that the system can distinguish parameter-related quality variation from degradation-related instability. Representative evaluation demonstrates that integrated process and sensor features can improve surface-quality prediction, tooldegradation recognition, early warning capability, and false-alarm performance compared with parameter-only and single-sensor models. The proposed framework supports predictive quality control, condition-aware tool replacement, intelligent machining optimization, and data-driven smart manufacturing.