Resumo:
Turning is one of the most widely used manufacturing processes in the industry. Its
extensive application means that turning processes are increasingly focused on producing
of high-quality parts, aiming to associate efficiency, precision, and productivity. The
challenges of achieving high-precision surface finishes are even greater when internal
turning is applied to modern materials such as polyetheretherketone (PEEK). To achieve
the best process conditions, predictive models must be estimated, and optimization must be
conducted. This work presents a statistical learning approach for modeling and optimizing
the internal turning process in PEEK tubes. Average roughness and roundness of the hole
were measured to quantify the hole quality. The cutting force, considered an important
indicator of machinability, was also measured. Cutting speed, feed rate, and fixture position
were considered as input parameters. For modeling, a learning procedure was proposed,
considering polynomial response surface regression, generalized additive methods, treebased
methods, support vector regression and extreme gradient boosting. Cross-validation
was used for learning and model selection, including k-fold and bootstrap approaches.
The results indicated that the extreme gradient boosting model was the best for all
predictors. For Ra the final prediction metrics results were RMSE = 0.1395, MAE =
0.1126, and R2 =1.0000, for Fc, RMSE =1.8609, MAE =0.9311, and R2 =0.9280, and for
Ront, RMSE = 21.3084, MAE = 17.8053, and R2 = 0.6562. Multi-objective evolutionary
optimization was performed, considering the extreme gradient boosting models for average
roughness, roundness, and cutting force, in addition to the deterministic model of material
removal rate. The NSGA-II method was selected considering the hypervolume for the
three-objective optimizations. The pseudo-weight approach is used to select high trade-off
solutions, facilitating selection in practical production scenarios. For optimization of Ra
vs Ront vs MRR, the balance between the three responses was achieved with a higher
vc, f = 0.12 mm/v, and fp = 15.14 mm. For optimization of Fc vs Ront vs MRR, the
balance between the three responses was achieved with vc = 378.78 m/min, f = 0.10
mm/v, and fp = 13.00 mm. The proposed learning and optimization approach enabled
the achievement of the best results for the internal turning process in PEEK and can be
applied to other intelligent manufacturing applications.
