Surface defect detection

Our group aims at developing the technique based on transformer architecture for surface defect detection in industrial field. 

(1) Transformer with block division and mask mechanism

[Publication] Qian Liu, Xiaohua Huang, Xiuyan Shao, and Fei Hao. Industrial Cylinder Liner Defect Detection by Transformer with Block Division and Mask Mechanism. Scientific Reports, vol. 12, 10698, 2022.

[Abstract] In the field of artificial intelligence, a large number of promising tools, such as condition-based maintenance, are available for large internal combustion engines. The cylinder liner, which is a key engine component, is subject to defects due to the manufacturing process. In addition, the cylinder liner straightforwardly affects the usage and safety of the internal combustion engine. Currently, the detection of cylinder liner quality mainly depends on manual human detection. However, this type of detection is destructive, time-consuming, and expensive. In this paper, a new cylinder liner defect database is proposed. The goal of this research is to develop a nondestructive yet reliable method for quantifying the surface condition of the cylinder liner. For this purpose, we propose a transformer method with a block division and mask mechanism on our newly collected cylinder liner defect database to automatically detect defects. Specifically, we first use a local defect dataset to train the transformer network. With a hierarchical-level architecture and attention mechanism, multi-level and discriminative feature are obtained. Then, we combine the transformer network with the block division method to detect defects in 64 local regions, and merge their results for the high-resolution image. The block division method can be used to resolve the difficulty of the in detecting the small defect. Finally, we design a mask to suppress the influence of noise. All methods allow us to achieve higher accuracy results than state-of-the-art algorithms. Additionally, we show the baseline results on the new database.

Framework of cylinder liner defect detection: It consists of two stages: a transformer network stage and a block division and mask mechanism stage. The transformer network focuses on training a deep network based on a local defect dataset, in which each image contains one defect, while the block division method separates an image into 64 blocks and then uses the network to detect each block, subsequently mapping the detected results into an original image, and the mask mechanism suppresses the noise in the background. In this section, we will detail each stage.

Visual analysis of two detected results (a) before and (b) after adding the mask mechanism. Best viewed in zoom and color.

In the first column, there are too many falsely detected results in the background. In fact, this background does not contain information about the cylinder liner. The mask mechanism removes the detected errors and increases the performance of the block division method. 


In the second column, the mask mechanism can suppress the influence of the background in our proposed method.