Machine Learning-Based Surface Roughness Prediction in Turning of Hardened AISI 4340 Steels: Incorporating Tool Wear via Cutting Length
This study explores the predictive modeling of surface roughness in the hard turning of AISI 4340 steel using machine learning techniques, specifically Random Forest (RF) and Gaussian Process Regression (GPR). The uncoated carbide was used and the experimental design considered varying cutting speed...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Tamkang University Press
2025-06-01
|
| Series: | Journal of Applied Science and Engineering |
| Subjects: | |
| Online Access: | http://jase.tku.edu.tw/articles/jase-202601-29-01-0003 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study explores the predictive modeling of surface roughness in the hard turning of AISI 4340 steel using machine learning techniques, specifically Random Forest (RF) and Gaussian Process Regression (GPR). The uncoated carbide was used and the experimental design considered varying cutting speeds ( 70, 90, 110 m/min ), feeds ( 0.14, 0.16, 0.22 mm/rev ), depths of cut ( 0.25, 0.5 mm ), and cutting length ( 80, 160, 240 mm ) to account for tool wear as an uncontrollable factor. The RF model achieved an RMSE of 0.1627µ m and an R^2 value of
0.9718 , while the GPR model had an RMSE of 0.1676µ m and an R^2 value of 0.8333 . The novelty of this study lies in considering the influence of tool wear via cutting length, significantly impacting the RMSE of the GPR model. Using K-fold cross-validation (K=7) on a 50% training dataset resulted in the lowest RMSE values for both models. Despite the GPR model’s slightly lower performance, it demonstrated robustness and consistency across different cross-validation splits and random states, making it a reliable option for predicting surface roughness. This research provides insights into the application of machine learning for process optimization in hard turning operations, highlighting the importance of tool wear and training dataset size. Future work could extend these findings to other machining processes and material types to validate the models’ robustness and generalizability. |
|---|---|
| ISSN: | 2708-9967 2708-9975 |