1DCNN-BiGRU network for surface roughness level detection. (1st December 2022)

• Record Type: Journal Article
• Authors: Pan, Zengren; Liu, Yanhui; Li, Zhiwei; Xun, Qiwen; Wu, Ying
• Is Part Of: Surface topography. Volume 10:Number 4(2022)
• Publication Date: 2022-12-01
• Subjects: surface roughness -- artificial intelligence -- 1-dimensional convolutional neural network -- bi-directional GRU network; Surfaces (Physics) -- Periodicals; Surfaces (Physics) -- Measurement -- Periodicals; 530.417
• DOI: 10.1088/2051-672X/aca10f
• Languages: English
• ISSNs: 2051-672X

Abstract

Abstract: Surface roughness was regarded as an essential indicator of the quality of machining. In machining demands, it was often necessary that the surface roughness of the workpiece lies in a specific range.For this reason, it was significant to detect the surface roughness level of the workpiece. For the traditional roughness detection methods with high manual involvement and unable to achieve automation, this paper proposed a new artificial intelligence detection approach. The approach consisted of a 1-Dimensional Convolutional Neural Network (1DCNN) and a Bi-directional Gated Recurrent Unit Network(BiGRU), called the 1DCNN-BiGRU model. 1DCNN-BiGRU accomplished the detection of roughness levels by classifying surface images directly, without extracting specific roughness features. First, 1DCNN was applied to automate the extraction of roughness-related features along the texture direction of the product surface image. Subsequently, the feature sequences extracted by 1DCNN were fed into BiGRU to learn the overall dependence of the roughness on the sequences. Experiments were performed on a 45steel workpiece roughness image dataset. The 1DCNN-BiGRU model gave 90.60% and 88.06% detection performance on the training and test sets, respectively.