Deep learning-based boundary effect filtering for nanometric roughness of diamond-turned micro-structured surface

Abstract

The functional performance of micro-structured surfaces manufactured by diamond turning is closely related to their nanometric surface roughness. Evaluating the surface roughness is crucial for determining the workpiece’s functionalities. However, conventional filters like wavelet and Gaussian filters can result in a boundary effect that affects the evaluation result when extracting surface roughness from micro-structured surface. To eliminate this boundary effect, a filter that combines deep learning with spectrum analysis for extracting nanometric surface roughness from micro-structured surfaces is proposed. First, the theoretical modeling of the method used for micro-structured surface filtering which contains spectrum analysis and deep learning structure is described in detail. Second, diamond turning experiment based on the designed micro-structured surface is performed for verifying the proposed method. Finally, the nanometric surface roughness without boundary effect is obtained by the proposed method and the filtering result is compared with the one obtained by the traditional filter. The result not only shows that the proposed method can achieve effective extraction for nanometric surface roughness, but also the values of the evaluation parameters are accurate and reliable compared with those obtained by the Gaussian regression filter which has the boundary effect. Significantly, the proposed method could fundamentally eliminate the boundary effect, thus improving the evaluation for nanometric surface roughness of micro-structured surface.