Machine Learning-Based Prediction of Surface Roughness in FFF PLA and Hybrid CNC Machining

This dataset supports the study “Machine Learning-Based Prediction of Surface Roughness in FFF PLA and Hybrid CNC Machining” and provides comprehensive experimental, manufacturing, and machine learning data for predicting surface roughness (Ra) in Polylactic Acid (PLA) components produced using fused filament fabrication (FFF) and hybrid additive–subtractive manufacturing.

The study investigates the influence of surface inclination on Ra, an aspect rarely addressed in prior literature, along with the effects of CNC post-processing. A Taguchi L54 design of experiments was implemented across surface inclinations ranging from 10° to 80°, resulting in 54 printed parts. Ten supervised machine learning (ML) algorithms were evaluated, with the Explainable Boosting Machine (EBM) demonstrating the highest predictive performance (R² = 0.983 for AM and R² = 0.616 for hybrid machining). In the AM stage, inclination angle, layer height, nozzle diameter, and extrusion temperature were identified as the dominant parameters, with optimal Ra achieved at a 0.10 mm layer height, 0.25 mm nozzle diameter, and 210 °C extrusion temperature. In hybrid machining, the combination of AM parameters with machining depth of cut and infill density reduced Ra by 85–91%, with a 0.20 mm depth of cut consistently producing low Ra values. Verification experiments confirmed prediction errors within ±2 µm for AM (MAPE = 6.62%) and ±1 µm for hybrid machining (MAPE = 9.17%). Additionally, a preliminary coolant-assisted machining trial demonstrated a further Ra reduction of 15–21%.

The dataset enables full reproducibility of the experimental workflow, including additive manufacturing, hybrid CNC machining, surface roughness measurements, machine learning model development, and validation. It is intended for researchers and practitioners working in additive manufacturing, hybrid manufacturing, surface integrity, and data-driven process optimization.

The dataset is organized into eleven sub-folders containing AM G-codes for 54 printed parts, CNC G-codes for hybrid and validation machining experiments, surface roughness (Ra) measurement reports for AM, hybrid, and wet machining stages, Taguchi L54 design of experiments files, machine learning datasets, Python scripts for ten supervised ML algorithms, and verification datasets for model validation.