TY - JOUR
T1 - Prediction of surface roughness with hybrid kernel extreme learning machine considering tool life
AU - Yang, Peiyi
AU - Duan, Zhongzhou
AU - Wang, Sanmin
AU - Deng, Liangfeng
AU - Tang, Xiangnan
AU - Liu, Zhibing
AU - Qiu, Tianyang
N1 - Publisher Copyright:
© The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2026.
PY - 2026/3
Y1 - 2026/3
N2 - Machined surface roughness is a significant indicator of product quality and the machining process; however, in complex production processes, surface roughness can be influenced by uncontrollable factors such as tool wear, which is challenging to predict with accuracy. In order to address the above problems, a novel hybrid kernel extreme learning machine (RBF_Arc_HKELM) with Gaussian and arc-cosine kernel functions was proposed, being used to predict the surface roughness in milling and machining of TC4 alloy. In addition, mesh searching was introduced to optimize the parameters of the kernel function in order to improve the prediction accuracy. Firstly, the TC4 alloy milling test was conducted in order to calculate the tool life based on the tool wear under the single-factor test with the variation of cutting speed, depth of cut, and feed per tooth. Finally, the empirical formula for tool life prediction including cutting speed, feed per tooth, and depth of cut was calculated and verified with orthogonal experiments. Secondly, the empirical formulas for tool life prediction, cutting parameters, and feature sets of cutting signal features were combined as inputs to train RBF_Arc_HKELM. Finally, to verify the validity and prediction performance of the model, it was compared with RBF_KELM and ArcCos_KELM. For the MAPE, RBF_Arc_HKELM decreased by 35.3% and 31.7%, respectively, and the mean absolute error (MAE) and the root mean square error (RMSE) were substantially reduced, which resulted in a better overall performance. Furthermore, the prediction results of the trained model without the incorporation of the empirical formula for tool life prediction as inputs were compared. It was found that the addition of the empirical formula for tool life prediction as inputs resulted in a 40.5% decrease in root mean square error (RMSE), a 43.7% decrease in mean absolute percentage error (MAPE), and a 48.1% decrease in mean absolute percentage error (MAPE). These results substantiated the validity and accuracy of the proposed model for online monitoring of surface roughness in TC4 alloy milling.
AB - Machined surface roughness is a significant indicator of product quality and the machining process; however, in complex production processes, surface roughness can be influenced by uncontrollable factors such as tool wear, which is challenging to predict with accuracy. In order to address the above problems, a novel hybrid kernel extreme learning machine (RBF_Arc_HKELM) with Gaussian and arc-cosine kernel functions was proposed, being used to predict the surface roughness in milling and machining of TC4 alloy. In addition, mesh searching was introduced to optimize the parameters of the kernel function in order to improve the prediction accuracy. Firstly, the TC4 alloy milling test was conducted in order to calculate the tool life based on the tool wear under the single-factor test with the variation of cutting speed, depth of cut, and feed per tooth. Finally, the empirical formula for tool life prediction including cutting speed, feed per tooth, and depth of cut was calculated and verified with orthogonal experiments. Secondly, the empirical formulas for tool life prediction, cutting parameters, and feature sets of cutting signal features were combined as inputs to train RBF_Arc_HKELM. Finally, to verify the validity and prediction performance of the model, it was compared with RBF_KELM and ArcCos_KELM. For the MAPE, RBF_Arc_HKELM decreased by 35.3% and 31.7%, respectively, and the mean absolute error (MAE) and the root mean square error (RMSE) were substantially reduced, which resulted in a better overall performance. Furthermore, the prediction results of the trained model without the incorporation of the empirical formula for tool life prediction as inputs were compared. It was found that the addition of the empirical formula for tool life prediction as inputs resulted in a 40.5% decrease in root mean square error (RMSE), a 43.7% decrease in mean absolute percentage error (MAPE), and a 48.1% decrease in mean absolute percentage error (MAPE). These results substantiated the validity and accuracy of the proposed model for online monitoring of surface roughness in TC4 alloy milling.
KW - Hybrid kernel extreme learning machine
KW - Surface roughness prediction
KW - Tool life prediction
UR - https://www.scopus.com/pages/publications/105031371951
U2 - 10.1007/s00170-026-17745-y
DO - 10.1007/s00170-026-17745-y
M3 - Article
AN - SCOPUS:105031371951
SN - 0268-3768
VL - 143
SP - 3301
EP - 3316
JO - International Journal of Advanced Manufacturing Technology
JF - International Journal of Advanced Manufacturing Technology
IS - 5-6
ER -