A Two-Stage Industrial Defect Detection Framework Based on Improved-YOLOv5 and Optimized-Inception-ResnetV2 Models

Aiming to address the currently low accuracy of domestic industrial defect detection, this paper proposes a Two-Stage Industrial Defect Detection Framework based on Improved-YOLOv5 and Optimized-Inception-ResnetV2, which completes positioning and classification tasks through two specific models. In order to make the first-stage recognition more effective at locating insignificant small defects with high similarity on the steel surface, we improve YOLOv5 from the backbone network, the feature scales of the feature fusion layer, and the multiscale detection layer. In order to enable second-stage recognition to better extract defect features and achieve accurate classification, we embed the convolutional block attention module (CBAM) attention mechanism module into the Inception-ResnetV2 model, then optimize the network architecture and loss function of the accurate model. Based on the Pascal Visual Object Classes 2007 (VOC2007) dataset, the public dataset NEU-DET, and the optimized dataset Enriched-NEU-DET, we conducted multiple sets of comparative experiments on the Improved-YOLOv5 and Inception-ResnetV2. The testing results show that the improvement is obvious. In order to verify the superiority and adaptability of the two-stage framework, we first test based on the Enriched-NEU-DET dataset, and further use AUBO-i5 robot, Intel RealSense D435 camera, and other industrial steel equipment to build actual industrial scenes. In experiments, a two-stage framework achieves the best performance of 83.3% mean average precision (mAP), evaluated on the Enriched-NEU-DET dataset, and 91.0% on our built industrial defect environment.

A surface defect detection method via fusing multi-level features

Surface defect detection is very important to ensure product quality, but most of the surface defects of industrial products are characterized by low contrast, big size difference and category similarity, which brings challenges to the automatic detection of defects. To solve these problems, we propose a defect detection method based on convolutional neural network. In this method, a backbone network with semantic supervision is applied to extract the features of different levels. While a multi-level feature fusion module is proposed to fuse adjacent feature maps into high-resolution feature maps successively, which significantly improves the prediction accuracy of the network. Finally, an Encoding module is used to obtain the global context information of the high-resolution feature map, which further improves the pixel classification accuracy. Experiments show that the proposed method is superior to other methods in NEU_SEG (mIoU of 85.27) and MT (mIoU of 77.82) datasets, and has the potential of real-time detection.

Simultaneous tailoring of pulsed thermography experimental and processing parameters for enhanced defect detection and sizing in adhesive bonds in carbon fibre composites

Thermographic inspection provides opportunity to tailor non-destructive evaluation to specific applications. The paper discusses the opportunities this presents through consideration of adhesive bonds between composites, such as those joining structural members and outer skins, where access is restricted to a single side. To date, literature focusses on the development of either an experimental procedure or data processing approach. This research aims to demonstrate the importance of tailoring both of these aspects to an application to obtain improved defect detection and robust quantification. Firstly, the heating stimulus is optimised to maximise the thermal contrast created between defect and non-defect regions using a development panel. Traditional flash heating is compared to longer square pulse heating, using a developed shutter system, compromising between experimental duration and heat input. A pulse duration of 4 seconds using two 130 W halogen bulbs was found double the detection depth from 1 mm to 2 mm, revealing all defects in the development panel. Temporal processing was maintained for all data using thermal signal reconstruction. Spatial defect detection routines were then implemented to provide robust defect/feature detection. Spatial defect detection encompassed a combination of image enhancement and edge detection algorithms. A two-stage kernel filter/binary enhancement method followed by the use of Canny edge detection was found most robust, providing a sizing error of 1.8 % on the development panel data. This process was then implemented on adhesive bonds with simulated bond line defects. The simulated defects are based on target detection threshold of 10 mm diameter void found at 1- 2 mm depth. All simulated void defects were detected in the representative bonded joint down to the minimum diameter tested of 5 mm. By considering the tailoring of multiple aspects of the inspection routine independently, an overall optimised approach for the application of interest has been defined.

Ultra-Compact Eddy Current Transducer for Corrosion Defect Search in Steel Pipes

This research is devoted to the application of non-destructive testing methods for detecting defects of the internal structure of the material in steel pipelines. Despite the use of modern approaches to the design and manufacture of pipelines, which make it possible to lay a significant margin of safety in the created system, the task of developing new approaches to measuring the technical and operational characteristics and parameters of steel parts using software and hardware complexes for non-destructive testing does not lose its relevance. The paper presents the results of the development of defect detection system aimed at detecting damage of the structure of the material with a diameter of 0.2 mm and located at a depth of up to 2 mm. The proposed system is based on the physical principles of the influence of the existing defect on the value of the transformer voltage, which is induced in the measurement circuit of the sensor built on eddy current effects. The focus of the research is the relationship between the linear dimensions of the defect, its location and the generated voltage indications of the developed sensor. Also, within the framework of the study, the results of processing and analysis of the data collected by the defect detection system are presented, the result of which was the determination of the parameters of the detected defects.