TY - JOUR
T1 - Surface roughness online prediction using parallel ensemble learning in robotic side milling for aluminum alloy
AU - Song, Ci
AU - Liu, Zhibing
AU - Wang, Xibin
AU - Qiu, Tianyang
AU - Liang, Zhiqiang
AU - Shen, Wenhua
AU - Gao, Yuhang
AU - Ma, Senjie
N1 - Publisher Copyright:
© 2025 Elsevier Ltd
PY - 2025/7/15
Y1 - 2025/7/15
N2 - Robotic machining has the advantages of large workspace and high flexibility, and the acquisition of high surface quality parts has become the research focus. To realize surface roughness online prediction in robotic side milling for aluminum alloy, an intelligent model driven by cutting vibration was adopted. Combined with variational modal decomposition (VMD) and fast fourier transformation (FFT), a rough and fine two-layer decomposition strategy of stable cutting vibration signals was proposed, which can effectively avoid the adverse effects of modal aliasing and endpoint action on the extraction of high-sensitivity features. Based on support vector machine (SVM), random forest (RF) and extreme learning machine (ELM), weighted reconstruction and voting selection were introduced to form a parallel ensemble learning model. Considering the posture influence in the sample construction, the machining system vibration state within common range was fully reflected by orthogonal cutting experiments at the corresponding postures of envelope zone vertices. For the setting process parameters, the obtained surface roughness was within 1.62–2.90 μm. Experiments under the testing postures showed the root mean square error (RMSE) and mean absolute error (MAE) of the model were 0.087 μm and 0.076 μm. The average relative error of 3.57 % and maximum relative error of 8.65 % had shown that the model obtained better surface roughness prediction ability. Finally, the influence significance and the factor action trend between cutting parameters and surface roughness were explored by range analysis, which has been proved by experiments that they can provide guidance for the dynamic regulation of cutting parameters.
AB - Robotic machining has the advantages of large workspace and high flexibility, and the acquisition of high surface quality parts has become the research focus. To realize surface roughness online prediction in robotic side milling for aluminum alloy, an intelligent model driven by cutting vibration was adopted. Combined with variational modal decomposition (VMD) and fast fourier transformation (FFT), a rough and fine two-layer decomposition strategy of stable cutting vibration signals was proposed, which can effectively avoid the adverse effects of modal aliasing and endpoint action on the extraction of high-sensitivity features. Based on support vector machine (SVM), random forest (RF) and extreme learning machine (ELM), weighted reconstruction and voting selection were introduced to form a parallel ensemble learning model. Considering the posture influence in the sample construction, the machining system vibration state within common range was fully reflected by orthogonal cutting experiments at the corresponding postures of envelope zone vertices. For the setting process parameters, the obtained surface roughness was within 1.62–2.90 μm. Experiments under the testing postures showed the root mean square error (RMSE) and mean absolute error (MAE) of the model were 0.087 μm and 0.076 μm. The average relative error of 3.57 % and maximum relative error of 8.65 % had shown that the model obtained better surface roughness prediction ability. Finally, the influence significance and the factor action trend between cutting parameters and surface roughness were explored by range analysis, which has been proved by experiments that they can provide guidance for the dynamic regulation of cutting parameters.
KW - Ensemble learning
KW - Robotic side milling
KW - Surface roughness prediction
UR - http://www.scopus.com/inward/record.url?scp=105007024189&partnerID=8YFLogxK
U2 - 10.1016/j.ymssp.2025.112932
DO - 10.1016/j.ymssp.2025.112932
M3 - Article
AN - SCOPUS:105007024189
SN - 0888-3270
VL - 235
JO - Mechanical Systems and Signal Processing
JF - Mechanical Systems and Signal Processing
M1 - 112932
ER -