Abstract:

Objective: To solve the problems of large measurement error, cumbersome operation, and difficult online detection in traditional contact roughness meters for measuring wood sanding surface roughness, this paper proposed a visual detection method based on local autocorrelation function entropy (LAEnt). Method: Firstly, the mechanism of detecting surface roughness using LAEnt was studied, and a local autocorrelation function entropy algorithm was established; Then, the orthogonal experimental method was used for wood sanding experiments to obtain sanding surface images and surface roughness values; Finally, the influence of factors such as the mesh number of sanding belts, sanding belt speed, and air drum feed rate on the surface roughness of wood sanding was studied. The correlation between LAEnt or autocorrelation function entropy (AEnt) and surface roughness was analyzed. Based on the local autocorrelation function entropy and autocorrelation function entropy data of the sanding surface images, support vector machine (SVM) was used to establish wood sanding surface roughness detection models SVM-LAEnt and SVM-AEnt, respectively. Result: The granularity of the sanding belt has the most significant impact on the surface roughness of wood sanding. There is a strong negative correlation between the granularity and the surface roughness. In contrast, the effects of belt speed and air drum feed rate on surface roughness are relatively minor. The local autocorrelation function entropy (LAEnt) shows a significant linear correlation with the surface roughness of wood sanding, with a correlation coefficient of 0.973 3. Furthermore, the feature extraction efficiency of LAEnt significantly outperforms that of AEnt (autocorrelation function entropy), with the per-image computational time reduced to 2.95% of AEnt’s processing time. SVM-based modeling results demonstrate that the SVM-LAEnt model achieves an average relative fitting error of 2.56% (maximum: 11.22%) and an average relative prediction error of 5.13% (maximum: 11.30%), both of which are superior to the SVM-AEnt model’s performance (average fitting error: 8.98%, maximum: 20.68%; average prediction error: 15.08%, maximum: 31.13%). Conclusion: The local autocorrelation function can describe the texture features and roughness of the wood sanding surface. When detecting the surface roughness, the local autocorrelation function entropy can better characterize the surface roughness. The results of this paper provide an efficient and accurate non-contact detection method for wood sanding surface roughness measurement.

Key words: wood sanding, surface roughness, local autocorrelation function, entropy, support vector machine (SVM)