| HU Jing-wen.Predictive Modeling of Surface Skewness and Kurtosis Based on BP Neural Network[J],46(2):235-239 |
| Predictive Modeling of Surface Skewness and Kurtosis Based on BP Neural Network |
| Received:August 14, 2016 Revised:February 20, 2017 |
| View Full Text View/Add Comment Download reader |
| DOI：10.16490/j.cnki.issn.1001-3660.2017.02.040 |
| KeyWord:surface skewness surface kurtosis grinding parameters neural network predictive modeling |
| Author | Institution |
| HU Jing-wen |
Foshan Radio & TV University, Foshan , China |
|
| Hits: |
| Download times: |
| Abstract: |
| The work aims to introduce surface skewness Ssk and surface kurtosis Sku to jointly characterize surface topography in a more precise and reliable manner when the symmetric degree and profile peak sharpness of surface texture structures and surface profile amplitude values tend to be subject to large difference, even though the surfaces obtained through various processing methods sometimes have the same Sa value regarding the situation that the surface topography is characterized mainly with the surface arithmetic average deviation Sa in the actual production. Orthogonal experiment and range analysis were applied to study the influence of grinding parameters on the change in Ssk and Sku. On this basis, BP neural network was introduced in the predictive modeling of Ssk and Sku. The complex problem of multi-input and nonlinearity for surface roughness modeling was effectively solved due to the property of self-learning. The effect laws of grinding parameters on Ssk and Sku were achieved. Ssk would reach the minimum when vs=20 m/s, vf=27 m/min, f=5 mm/min and ap=0.005 mm, and Sku was the minimum when vs=29 m/s, vf=23 m/min, f=25 mm/min and ap=0.002 mm. And then, the accurate neural network prediction models for Ssk and Sku based on grinding parameters were built respectively. vf and f have a significant impact on Ssk. Similarly, f and vs impact Sku the most. It is necessary to select suitable grinding parameters to obtain the surface with more valleys and less acute profile peaks. Moreover, the prediction models built can guide the optimization of grinding process effectively. |
| Close |
|
|
|