Surface roughness measurement method based on multi-parameter modeling learning. (December 2018)

• Record Type: Journal Article
• Authors: Chen, Suting; Feng, Rui; Zhang, Chuang; Zhang, Yanyan
• Is Part Of: Measurement. Volume 129(2018)
• Page Start: 664
• Page End: 676
• Publication Date: 2018-12
• Subjects: Feature extraction -- Random forest -- Mutual information -- Roughness learning; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2018.07.071
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A multi-feature fusion descriptor based on speckle feature, grayscale feature and texture feature is constructed. An identification reasoning method based on ACC-random forest is proposed to determine the work-piece classification. In order to realize surface roughness measurement effectively, a multi-parameter learning model was established. Abstract: To improve the accuracy and efficiency of the existing roughness measurement methods, we propose a new surface roughness measurement technique based on multi-parameter modelling learning. First, multi-feature descriptor is constructed through speckle feature, grey feature and Tamura texture feature. Then, an identification reasoning method based on ACC-random forest was proposed to determine the work-piece classification. Finally, to realize surface roughness measurement efficiently, a multi-parameter learning model is established. Through establishment and optimization of multi-parameter surface roughness modeling, the value of surface roughness can be measured accurately. Thus, not only the class of work-piece be classified, also the value of surface roughness can be measured. Our proposed method breaks through the limitations of existing methods, which are based on several roughness measurement models for different classes of work-pieces. The experimental results indicate that our proposed method significantly outperform the state-of-the-art methods in terms of classification accuracy and measurement error rate.