The aim of the current research is to compare the surface roughness models based on the Box–Behnken Design and Levenberg–Marquardt Algorithm-based artificial neural networks. For prediction of the surface roughness of magnetic abrasive flow machining (MAFM) finished rare earth oxides (REOs) aluminium composites, Box–Behnken Design models were developed using three-level factorial design as magnetic flux density, number of cycles and extrusion pressure as process parameters. The artificial neural networks predictive models of surface roughness were developed using feed forward back propagation network procedures called the Levenberg-Marquardt Algorithm. Also, an attempt has been made to compare the Levenberg–Marquardt Algorithm-based artificial neural networks and Box–Behnken Design for the modeling of surface roughness results. The value of coefficient of determination for the Box–Behnken Design model is found to be high (R² = 0.9737), which is an indication of good fit for the model with high significance. The percentage error for the Box–Behnken Design model is observed to be more as compared to the Levenberg–Marquardt Algorithm-based artificial neural networks model. The comparison evidently indicates that the prediction capabilities of trained artificial neural networks models are far better than the Box–Behnken Design models. Further, specimens were examined using atomic force microscopy for three-dimensional surface profiles.

当前研究的目的是比较基于 Box-Behnken 设计和基于 Levenberg-Marquardt 算法的人工神经网络的表面粗糙度模型。为了预测磁磨粒流加工 (MAFM) 成品稀土氧化物 (REO) 铝复合材料的表面粗糙度，Box-Behnken Design 模型使用磁通密度、循环次数和挤压压力作为工艺的三级因子设计开发参数。表面粗糙度的人工神经网络预测模型是使用称为 Levenberg-Marquardt 算法的前馈反向传播网络程序开发的。此外，还尝试比较基于 Levenberg-Marquardt 算法的人工神经网络和 Box-Behnken 设计对表面粗糙度结果的建模。R ²  = 0.9737)，这表明该模型拟合良好，具有高显着性。与基于 Levenberg-Marquardt 算法的人工神经网络模型相比，观察到 Box-Behnken 设计模型的百分比误差更大。比较显然表明，经过训练的人工神经网络模型的预测能力远优于 Box-Behnken Design 模型。此外，使用原子力显微镜检查样品的三维表面轮廓。