Development of a Comprehensive Model of Assessment of Surface Roughness after Surface Plastic Deformation

The article discusses the methodology for constructing a mathematical model based on machine learning methods for predicting roughness of the machined surface after smoothing and rolling. The proposed approach enables adjustment for various machining process conditions (workpiece material, tool type and parameters, machining modes). The technique also provides the option to retrain the model on new data for different machining conditions, which ensures the maximum degree of generalization of the predictive model. A technique for preprocessing input parameters depending on their type is presented. Two approaches to compilation of the training dataset are considered: (1) based on experimental data and (2) based on theoretical dependencies. Several different approaches to generation of the generalized model are investigated: individual models, including bootstrap-based models (linear regression, support vector machines, decision trees, Gaussian process regression), and ensemble methods based on bagging (boosted trees). The results confirm the applicability of the proposed approach to creation of generalized models that simplify design planning of technological processes in the conditions of multi-product manufacturing, which entails high variety of both input process conditions and machined workpieces.