Self-attention guided model for defect detection of aluminium alloy casting on X-ray image. (December 2020)

• Record Type: Journal Article
• Authors: Wang, Yongxiong; Hu, Chuanfei; Chen, Kai; Yin, Zhong
• Is Part Of: Computers & electrical engineering. Volume 88(2020)
• Publication Date: 2020-12
• Subjects: Convolutional neural network -- Defects of aluminium alloy casting -- Defect detection -- Self-attention mechanism -- X-Ray image; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2020.106821
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: A novel self-attention guided CNN based model for a DR based NDT system is proposed to detect defects of castings on X-ray images. The architecture of the gradual feature extraction for capturing the representations of subtle defects is demonstrated to be effective. The impact of image parameters, such as brightness and contrast, on the proposed model is investigated through experimental analysis. Graphical abstract: Abstract: An automated and reliable non-destructive testing (NDT) system for aluminium alloy casting is the key to guaranteeing the quality of products and reducing production costs. In order to achieve accurate detection with a digital radiography (DR) based NDT system, we propose a new deep model to detect the subtle defects of aluminium alloy casting on the X-ray image. Specifically, the proposed model is composed of two subnetworks: a general feature network (GFN), and a subtle feature network (SFN). Furthermore, the self-attention mechanism is modeled as a self-attention guided module (SGM) embedded in the SFN. SGM enhances the capability of the model to extract subtle features against a complex background. The effectiveness of the proposed model is validated on the actual X-ray images acquired in the casting process. The experimental results demonstrate that the proposed model has more satisfactory detection performances in comparison with classical deep learning models.