.png){kind=icon}
.png){kind=icon}
Return
A systematic approach to modelling & optimisation of small-scale empirical data of robotic GMAW cladding using machine learning
.png){kind=icon}
PRE
AI
R
A
M
B
M
S
R
DOI：10.1007/s40964-026-01897-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Machine Learning (ML) applications require substantial data. Capturing bead geometry in cladding is expensive and time-consuming, requiring steps like sectioning, grinding, polishing, etc. It may restrict ML applications in the present use case. Hence, the current study has focused on developing a systematic approach for modelling and Optimisation using ML for limited and expensive experimental data. The experiments were conducted as per response surface methodology based on central composite design (RSM–CCD), and the obtained data was modelled with five ML regressors, i.e., Machine learning polynomial regression (MLPR), kernel ridge regression (KRR), support vector regression (SVR), Deep neural network (DNN) and adaptive boost regression (ABR), which was optimised with diligent hyperparameter tuning. A data-split method to cope with limited data size has been studied, and algorithms have been compared with the most rigorous form of cross-validation, i.e., leaving one out cross-validation (LOOCV). SVR was the best model for mean absolute error (MAE), mean absolute percentage error (MAPE) and LOOCV. For all studied responses, i.e., penetration (P), reinforcement height (RH), bead width (BW) and percentage dilution (%D), SVR was found to have 16%, 46%, 14%, and 69% lower mean of mean absolute error (MMAE), respectively, in comparison to MLPR, which is central to many traditional modelling techniques. Finally, a search algorithm inspired by grid search has been proposed to find optimal process parameters using machine learning, and a comparison has been drawn with a standard grey wolf optimiser. The framework proposed in the paper can be used to augment traditional modelling and optimisation techniques, enabling a better capture of non-linear variations in the response variables.
Keywords:
Gas metal arc welding (GMAW)
Leave one out cross-validation (LOOCV)
Kernel ridge regression (KRR)
Support vector regression (SVR)
Deep neural network (DNN)
Optimisation
Journal
P
IF:
5.4
Papers:
1.8K
Citations:
3.2K
Organization
Cited Papers