Research on neural network model of surface roughness in belt sanding process for Pinus koraiensis

Abstract

At present, researches about surface roughness mainly focus on the measurement methods and the relation between wood surface roughness and adhesion strength as well. It is well known that some sanding parameters are considerably key factors for analyzing the cause of surface roughness. However, only a few research studied the relations among surface roughness, sanding parameters, sanding pressure and wood texture direction. Back-propagation network (BP network) system was applied to simulate and predict the surface roughness value during the sanding process. The aim of this study was to determine the effects of sanding parameters on surface roughness of pinus koraiensis and the relation between sanding pressure and surface roughness, and finally establish the model of surface roughness combined with wood texture characteristic of pinus koraiensis through neural network system. The results showed that All values of surface roughness had a “vacuum belt” when λ = 0° and λ = 45° but it did not appear when λ = 90° (λ refers to the angle between the feeding direction and the wood grain). The confidence of the fitting surface roughness results was 97% by BP network model and the average error was 5.8%, which can simply and successfully predict surface roughness during sanding process.