Adaptation of Surface Roughness Models Based on AI in Changing Environments

Machining empirical models for surface roughness based on statistical or artificial intelligence (AI) approaches have been intensively studied during last decades. However, there is no industrial methodology to be applied in industry due to the time consuming and costly experimental procedures required. Furthermore, continuous changes in the cutting process such as cutting-tool replacements or changes in material work pieces make them unfeasible. This paper presents two methodologies to develop empirical models based on few experimental runs based on design of experiments, Taguchi's array, and AI techniques to learn the model from experimental data. The first methodology models the system from the scratch. The second methodology adapts a previous AI model with few experimental runs, defining two possible methods for adaption according to model change. A case study is presented, where a face milling operation with two cutting-tool changes is analysed. The results show a steady prediction error less than 20% along process changes.