Interpolation-based virtual sample generation for surface roughness prediction

Surface roughness is an essential technical indicator for the surface quality of machined parts and significantly affects the service performance of the products. Accurate prediction of the surface roughness in the machining process can play an important role in reducing costs and increasing efficiency. However, data-based methods often require a large sample size for model training to improve prediction accuracy. Obtaining a sufficient number of training samples is challenging due to cost and efficiency constraints. To this end, an interpolation-based virtual sample generation scheme is proposed in this article, which utilizes a broad learning system (BLS) to generate virtual samples of the cutting groove surface roughness. Experimental verification was carried out in an ultra-precision machining center, where depth of cut and cutting speed were selected as inputs to the BLS. The results reveal that the proposed virtual sample generation approach can considerably improve the surface roughness prediction accuracy. Compared to other machine learning methods, BLS has the highest error reduction rate with and without virtual samples. © 2022, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.