Prediction of surface roughness using fuzzy broad learning system based on feature selection

Abstract

Surface roughness is the most important indicators for product quality. Many uncertain factors affect the surface roughness of the workpiece in the machining processes. Fuzzy broad learning system (FBLS) has shown considerable advantages in nonlinear and uncertain modeling. Thus it is a promising candidate for predicting surface roughness in machining processes. However, there are some irrelevant or redundant features in the model training process. In this regard, a novel FBLS based on feature selection is proposed, in which three binarization mechanisms of the binary grey wolf optimization (BGWO-I, BGWO-II, and BGWO-III) are used to select features on the feature layer, enhancement layer, and hidden layer of the FBLS, respectively. Then the performance of the proposed method is evaluated by predicting the surface roughness in an actual slot milling process, wherein multi-sourced fusion process parameters and different signal features are analyzed. The obtained experimental results demonstrate that the proposed methods outperform conventional methods in terms of prediction accuracy.