Tiny Defects Research Articles

Improving PCB defect detection using selective feature attention and pixel shuffle pyramid

– Due to the ongoing miniaturization of electronic products and the use of miniature printed circuit boards (PCBs), existing AI-based defect detection methods have exhibited poor performance in detecting tiny PCB defects. This issue can potentially compromise safety, degrade manufacturing quality, and increase production costs. To tackle this problem, we propose two novel techniques for PCB defect detection, namely Selective Feature Attention (SF attention) and Pixel Shuffle Pyramid (PSPyramid). SF attention identifies important features from a pyramid feature map to fuse the semantic and spatial information, while PSPyramid effectively fuses semantic features to detect various types of defects on PCBs, especially tiny defects. Moreover, a customized training strategy, specifically for PCB defect detection, is devised. To evaluate the performance of our proposed algorithms, extensive experiments have been conducted on two well-known PCB datasets containing tiny defects: the DeepPCB and TDD datasets. Our proposed non-referential method achieves performance comparable to existing referential methods on the DeepPCB dataset, making it more feasible for industrial applications. Compared to state-of-the-art methods, our method reduces the error by 16%, in terms of AP50, on the TDD dataset. The experimental results demonstrate the effectiveness of our proposed method in improving the quality assurance process for PCBs in the electronics industry.

TD-Net:tiny defect detection network for industrial products

AbstractThe detection of tiny defects in industrial products is important for improving the quality of industrial products and maintaining production safety. Currently, image-based defect detection methods are ineffective in detecting tiny and variously shaped defects. Therefore, this paper proposes a tiny defect detection network (TD-Net) for industrial products to improve the effectiveness of tiny defect detection. TD-Net improves the overall defect detection effect, especially the detection effect of tiny defects, by solving the problems of downsampling of tiny defects, pre-filtering of conflicting deep and shallow semantic information, and cascading fusion of multi-scale information. Specifically, this paper proposes the Defect Downsampling (DD) module to realize the defect information supplementation during the backbone downsampling process and improve the problem that the stepwise convolution easily misses the detection of tiny defects. Meanwhile, the Semantic Information Interaction Module (SIIM) is proposed, which fuses deep and shallow semantic features, and is designed to interact the fused features with shallow features to optimize the detection of tiny defects. Finally, the Scale Information Fusion Module (SIFM) is proposed to improve the Path Aggregation Network (PANet) for cascading fusion and information focus on different scale information, which enables further improvement of defect detection performance of TD-Net. Extensive experimental results on the NEU–DET data set (76.8$$\%$$ % mAP), the Peking University PCB defect data set (96.2$$\%$$ % mAP) and the GC10-DET data set (71.5$$\%$$ % mAP) show that the proposed TD-Net achieves competitive results compared with SOTA methods with the equivalent parameter quantity.

Study on the detection technology for inner-wall outer surface defects of the automotive ABS brake master cylinder based on BM-YOLOv8

A defect detection approach based on the BiFormer + MPDIoU’s YOLOv8 (BM-YOLOv8) model is proposed which addresses the challenges of low accuracy and low efficiency in detecting tiny defects on the inner-wall outer surface of automotive Anti-lock Brake Systems (ABS) brake master cylinder. This method constructs an imaging model based on process parameters such as speed and inspection accuracy required during the production of automotive ABS brake master cylinder. On this basis, it employs the dynamic sparse self-attention mechanism of the BiFormer to build a network for self-attention feature extraction and fusion. It also utilizes the Minimum Point Distance Intersection over Union (MPDIoU) to optimize the bounding box regression loss function, allowing for precise detection of defects on the inner-wall outer surface of automotive ABS brake master cylinder. Both qualitative and quantitative studies demonstrated that the BM-YOLOv8 method achieves a defect identification rate of 98.8% for the inner-wall outer surface defects of automotive ABS brake master cylinder. More than 25 images per second can be detected in this process. The performance of this method meets the accuracy and real-time requirements for defect detection on the inner-wall outer surface of automotive ABS brake master cylinder.

Acoustic field simulation-based ultrasonic water immersion non-destructive testing for titanium alloy plate

ABSTRACT Titanium alloy plate is widely used in automotive, aerospace, medical and other fields. Ultrasonic testing is a highly efficient non-destructive testing method for the inspection of titanium alloy components. However, there are strong distortion and attenuation of sound wave propagation in titanium alloys due to the presence of anisotropy and inhomogeneities, making it difficult to detect tiny defects in such components. In addition, some configuration parameters, such as the water path depth also affect the distribution of the acoustic field in the detected components. It is necessary to predict the acoustic field distribution to achieve a better testing accuracy. An acoustic field simulation method based on a multi-Gaussian beam is proposed, which can model the focused acoustic field in a multilayer anisotropy medium. The relationship between the acoustic focused area and the water path depth was explored and optimised comprehensively. C-scan imaging of specimen with flat bottom holes was performed at different water path depths. The results show that the proposed method can optimise the configuration parameters to improve the accuracy of the flaw sizing. This study provides an effective method for testing titanium alloy plate.

Development of an intelligent multi-scale defect detection system for aspheric lenses with high accuracy and efficiency

Existing methods struggle to fulfill the demand for fast and precise detection of multi-scale faults in aspheric lenses. A high accuracy and efficiency intelligent detection system for aspheric lenses is developed, which utilizes four-step phase-shifted grating optical imaging and Fourier low-pass filtering as image acquisition and preprocessing systems. Accurate and efficient defect feature extraction methods are designed based on regression loss and lightweight feature selection networks, and the model's ability to learn multi-scale defect information is improved. Finally, experimental tests were conducted based on the self-constructed multi-scale defect datasets of aspheric lenses. The results show that the detection system achieves a global accuracy of 96.3 %, an accuracy of 63.8 % for tiny defects detection, as well as a single-image detection time of 24.9 ms, which is a significant improvement in terms of accuracy and efficiency compared with existing methods. Therefore, the intelligent detection system can be widely applied to defect detection in aspheric lenses.

Smart grid line fault detection based on deep learning image recognition algorithm

Abstract Amid the rapid evolution of smart grids, stringent demands for reliability, safety, and efficiency have escalated for transmission lines. Nevertheless, given their extensive coverage and the complexity of their environments, traditional inspection techniques struggle to meet the rigorous standards of real-time monitoring and precision. This paper introduces a novel fault detection approach for smart grid transmission lines, leveraging advanced deep-learning image recognition algorithms. This method improves the YOLOv5 series models by combining the Convolution Attention Module (CBAM), Bidirectional Feature Pyramid Network (BiFPN), and MobileNet-V3 as the feature extraction network of YOLOv5 to achieve defect detection of hardware on transmission lines. CBAM improves the model’s sensitivity to tiny defects by focusing on key areas and features. BiFPN improves detection accuracy and robustness through efficient fusion of multi-scale features. Furthermore, the integration of MobileNet-V3 enhances both the efficiency and precision of feature extraction, ultimately elevating the overall model’s performance. Experimental results show that compared with the original YOLOv5 model, the improved algorithm can achieve 92.89% accuracy and 84.10% recall, which is better than other mainstream target detection algorithms. Especially in the detection of defects in small targets and complex backgrounds, the improved model shows stronger robustness and adaptability. This provides strong technical support for the efficient operation and maintenance of smart grids.

X-ray detection of ceramic packaging chip solder defects based on improved YOLOv5

It is of great significance to detect solder defects of ceramic packaging chips by X-ray detection to improve the quality of electronic products. However, the scale of the solder defects varies violently and there are a lot of tiny, dense, and long narrow defects, making it difficult to identify timely and accurately. To solve the issues mentioned above, an improved object detection model based on the YOLOv5 network, namely YOLO-STPN, is proposed in this paper. We add another prediction head to the original model to detect tiny solder defects effectively. In addition, the swin transformer block and convolutional block attention module are integrated into the path aggregation network, which improve the networks’ ability to identify dense and long narrow defects. The positive sample matching mechanism and the complete intersection over union (CIoU) loss are modified to make the model focus on long narrow defects. Experimental results of 10 types of ceramic packaging chips show that the mean average precision at 50 % IoU (mAP50) of YOLO-STPN is 10.05 % higher than the original YOLOv5 on average, while the decrease of frames per second (FPS) is acceptable. The proposed method has high accuracy and speed. Therefore, it is applicable to the real-time detection of chip packaging.

HDTNet: PCB defect detection algorithm for tiny objects of high-density regions

PCB defect detection aims to identify the presence of gaps, open circuits, short circuits, and other defects in the PCB boards produced in the industry. Designing effective deep learning algorithms is crucial to finding a solution. Previously proposed PCB defect detection algorithms are limited in detecting tiny objects in high-density. Directly applying previous models to tackle PCB defect detection tasks will cause serious issues, such as missed detection and false detection. In this paper, we present a detection algorithm for tiny PCB defect targets in high-density regions to solve the above-mentioned problems. We firstly propose a detection head to detect tiny objects. Then, we design a four-channel feature fusion mechanism to fuse four different scale features and add an attention mechanism to find the attention region in scenarios with dense objects. Finally, we achieved accurate detection of tiny targets in high-density areas. Experiments were performed on the publicly available PCB defect dataset from Peking University. Our mAP@.5:.95 achieves 48.6%, while mAP@0.5 exceeds 90%. Compared with YOLOX and YOLOv5, our improved model can better localize tiny objects in high-density scenes. The experimental results certify that our model can obtain higher performance in comparison with the baseline and the state of the art.

Research on the Construction of an Efficient and Lightweight Online Detection Method for Tiny Surface Defects through Model Compression and Knowledge Distillation

In response to the current issues of poor real-time performance, high computational costs, and excessive memory usage of object detection algorithms based on deep convolutional neural networks in embedded devices, a method for improving deep convolutional neural networks based on model compression and knowledge distillation is proposed. Firstly, data augmentation is employed in the preprocessing stage to increase the diversity of training samples, thereby improving the model’s robustness and generalization capability. The K-means++ clustering algorithm generates candidate bounding boxes, adapting to defects of different sizes and selecting finer features earlier. Secondly, the cross stage partial (CSP) Darknet53 network and spatial pyramid pooling (SPP) module extract features from the input raw images, enhancing the accuracy of defect location detection and recognition in YOLO. Finally, the concept of model compression is integrated, utilizing scaling factors in the batch normalization (BN) layer, and introducing sparse factors to perform sparse training on the network. Channel pruning and layer pruning are applied to the sparse model, and post-processing methods using knowledge distillation are used to effectively reduce the model size and forward inference time while maintaining model accuracy. The improved model size decreases from 244 M to 4.19 M, the detection speed increases from 32.8 f/s to 68 f/s, and mAP reaches 97.41. Experimental results demonstrate that this method is conducive to deploying network models on embedded devices with limited GPU computing and storage resources. It can be applied in distributed service architectures for edge computing, providing new technological references for deploying deep learning models in the industrial sector.

Learning-based defect inspection of laptop covers with a laser profile sensor

Quality inspection is a crucial process in the manufacturing industry, which significantly impacts the productivity and profits of a factory. To reduce costs and enhance quality control, companies are increasingly driving automation of this process due to the limitations of manual labour. However, the lack of flexible solutions that can adapt to the changing environment in factories has been a challenge. Computer-vision-based inspection methods often require hefty mechanisms and strict lighting conditions, which limit the investment in such solutions. Furthermore, with increasing demands for precision, traditional algorithms are struggling to cope with the problem, while data-driven solutions like deep learning require significant effort in acquiring and labelling data. To address these problems, this paper proposes an eye-in-hand system, i.e., a sensor attached to a robotic manipulator, integrated with unsupervised learning algorithms. One typical product, laptop bottom covers, is tested and discussed with regard to the system to validate our method since it has tiny defects and light-absorbing coatings. With only nine samples of the product, our system solved the case and demonstrated high-precision results and a flexible workflow. In this paper, our main contributions are: 1) we propose a flexible quality inspection system and workflow that needs extremely small volumes of data with unsupervised learning algorithms; and 2) we first propose using a laser profile sensor for collecting 2D intensity images, that are resilient to ambient light disturbance, of matt laptop cover data without lighting equipment.

Enhanced SSD framework for detecting defects in cigarette appearance using variational Bayesian inference under limited sample conditions.

In high-speed cigarette manufacturing industries, occasional minor cosmetic cigarette defects and a scarcity of samples significantly hinder the rapid and accurate detection of defects. To tackle this challenge, we propose an enhanced single-shot multibox detector (SSD) model that uses variational Bayesian inference for improved detection of tiny defects given sporadic occurrences and limited samples. The enhanced SSD model incorporates a bounded intersection over union (BIoU) loss function to reduce sensitivity to minor deviations and uses exponential linear unit (ELU) and leaky rectified linear unit (ReLU) activation functions to mitigate vanishing gradients and neuron death in deep neural networks. Empirical results show that the enhanced SSD300 and SSD512 models increase the model's detection accuracy mean average precision (mAP) by up to 1.2% for small defects. Ablation studies further reveal that the model's mAP increases by 1.5%, which reduces the computational requirements by 5.92 GFLOPs. The model also shows improved inference in scenarios with limited samples, thus highlighting its effectiveness and applicability in high-speed, precision-oriented cigarette manufacturing industries.

Tiny Defect-Oriented Single-View CT Reconstruction Based on a Hybrid Framework

Tiny Defect-Oriented Single-View CT Reconstruction Based on a Hybrid Framework

Research on the application of information technology modeling based concrete reinforcement construction for bridge projects

Abstract This paper discusses the application of concrete in bridge and house reinforcement, especially for detecting tiny defects during construction. Taking the optical principle of shear device and the principle of scattering interferometry as the theoretical basis, the article proposes a nondestructive testing technique applicable to concrete construction in bridge engineering. This technique can effectively detect the tiny defects in the structure by combining the time phase shift interferometry. The research and analysis show that the digital shear scattering interferometric detection system proposed in this paper performs well in defect detection, and its calculated defect values are highly similar to the actual data with an error of less than 6%. The system can also present different responses in time according to the different lengths or depths of the defects. Further applications showed that the vibration resistance of the bridges was significantly improved after the concrete reinforcement construction using the defects detected by the system, and the self-oscillation cycles were reduced by 8.1% and 11.8%, respectively. Overall, the study provides an efficient technical method for detecting and treating defects in concrete reinforcement works.

An Efficient Tiny Defect Detection Method for PCB With Improved YOLO Through a Compression Training Strategy

An Efficient Tiny Defect Detection Method for PCB With Improved YOLO Through a Compression Training Strategy

TD‐YOLO: A Lightweight Detection Algorithm for Tiny Defects in High‐Resolution PCBs

AbstractPrinting circuit board (PCB) defect inspection precisely and efficiently is an essential and challenging issue. Therefore, based on several improvements upon YOLOv5‐nano, a novel lightweight detector named TD‐YOLO is proposed to inspect tiny defects in PCBs. First, the lightweight ShuffleNet block is implemented into the backbone to effectively reduce the model weight. Second, novel anchors are designed using modified k‐means clustering to accelerate the model convergence and yield superior detection precision. Then, data augmentation strategy is recomposed by rejecting mosaic augmentation to suppress the emergence of extremely tiny targets. Finally, a mighty feature pyramid network namely MPANet, is newly proposed to boost the feature fusion capability of the model. The experiment results denote TD‐YOLO achieves the highest 99.5% mean average precision on our dataset, outperforming other state of the arts. Specially, the detection metrics for the smallest two defects, such as spur and mouse bite, are increased by 2.1% and 1.2%, respectively, compared with YOLOv5‐nano. Besides, TD‐YOLO has only 1.33 million parameters, decreased by 25% than the baseline. Using a mediocre processor, the detection speed is boosted by 20%, reaching 37 frames per second for the input size of 22402240 pixels.

Design of multi-objective guided filter infrared thermal image fusion for hypervelocity impact damages detection

Spacecraft hypervelocity impact damage assessment is becoming increasingly important as the space debris environment deteriorates. Common single-target guided filtering methods are hardly capable of coping with the complex damage infrared thermal image fusion requirements of hypervelocity impacts. In this paper, a two-layer infrared reconstructed image fusion algorithm based on Multi-objective guided filtering (MOGF) is proposed. The aim is to keep the filtered input and output images similar while reducing thermal image noise and retaining as much of the tiny defects as possible in the fusion process. MOGF combines a multi-objective evolutionary optimization algorithm to simultaneously optimize multiple guided filter cost functions in the form of vectors in order to obtain the optimal set of guided filter weight parameters. Given the problem of irregular frontier surfaces in practical multi-objective optimization problems, a weight vector adaptive adjustment strategy is introduced to obtain a more uniform population. Finally, the optimal weight set is passed to the full pixel fusion layer to complete the complex damage infrared reconstructed image fusion. The experiments demonstrate that the proposed fusion method considering multiple targets has higher performance in hypervelocity impact complex damage fusion compared to the general single target fusion method.

Surface defect detection of industrial components based on vision

Early and effective surface defect detection in industrial components can avoid the occurrence of serious safety hazards. Since most industrial component surfaces have tiny defects with high similarity to the detection background, there are often issues of missed or false detections when defects are detected, leading to low detection accuracy. To deal with the aforementioned issue, this essay suggests a high-precision detection model for surface defects in industrial components based on the YOLOv5 algorithm. First, the original spatial pyramid pooling (SPPF) is innovated by proposing the SPPFKCSPC module, which improves the network's capacity for feature extraction from targets at different scales and fuses multiscale features better. Then, C3 is combined with SPPFKCSPC and replaces the C3 module of the backbone network, which improves feature expression and enhances the receptive field of the network. Finally, the coordinate attention mechanism (CA) has been embedded into the YOLOv5 neck network, and the bounding box regression loss function of the algorithm is improved to EIOU, not only improving the precision of the target localization and recognition model but also enhancing the overall network performance. Based on the public datasets NEU-DET and PV-Multi-Defect, multiple sets of experiments were conducted using innovative algorithms. On the NEU-DET dataset, we got a mean average accuracy (mAP) of 88.3%, which is 7.2% greater than the original approach. On the PV-Multi-Defect dataset, the mAP value reached 97.5%, an improvement of 1.5%. As shown by the experimental data, the detection results significantly improved.

Adaptive acquisition and recognition system of blade surface defects during machining process

This study proposes an adaptive acquisition and recognition system of blade surface defects during machining process. The study has developed a hardware system consisting of motion platforms with four degrees of freedom (DOF) and industrial camera systems, along with an optimized workflow for the acquisition of blade surface defects. Subsequently, the hill-climbing algorithm, the energy gradient function and an adaptive evaluate image definition method are utilized to acquire clear images of the blade surface which contains tiny machined surface defects. An improved you only look once v5 (YOLOv5) algorithm is finally proposed to recognize the type and location of blade surface defects. The improved YOLOv5 algorithm uses K-means++ algorithm to cluster marking boxes, introduces a convolutional block attention module (CBAM) attention mechanism in the cross stage partial network with 3 convolutions (C3 module), and adopts the efficient intersection over union (EIoU) loss function instead of the complete intersection over union (CIoU) loss function to improve the recognition accuracy. The result shows that the proposed adaptive acquisition and recognition system can clearly collect the blade surface defects. The improved YOLOv5 algorithm can identify the type and location of blade surface defects, and the mean average precision (mAP) improved by 1.4 % compared to the original YOLOv5.

Cascaded detection method for surface defects of lead frame based on high-resolution detection images

In the field of semiconductor production and manufacturing, the detection of defects on lead frame surfaces is a vital process. This process plays a key role in ensuring the quality of the final product. Using high-resolution detection images to detect multi-scale tiny surface defects is necessary, but this amplifies the impact of environmental noise. Therefore, suppressing both the false negative rate and false positive rate in practical detection scenarios is a challenge that needs to be overcome. Current research on lead frame surface defect detection is mostly concentrated on the downloaded standard original images, which limits its application in actual production lines. This paper presents a cascaded detection method for surface defects of lead frame based on high-resolution detection images. Firstly, this study presents the unit cell extraction module to convert the detection object from high-resolution image to hundreds of unit cells. The proposed module can handle real-time detection images in the production pipeline, especially addressing situations such as lighting imbalances and tilted detection images. Subsequently, this study proposes a lead frame surface defect detection network (LDD-net), which takes unit cells as inputs and can effectively detect multi-scale defects. Compared to other models, LDD-net can effectively capture the features of subtle defects. Additionally, this paper introduces the deviation in the central width direction into the CIoU localization loss, enhancing the accuracy of defect localization in LDD-net. The data set is constructed using the machine vision detection system and conducts training and testing. Specifically, experiments of LDD-net on the data set obtained 85.01% mean average precision (mAP) and 37 ms of inference time, respectively. The detection accuracy exceeds 95%, and the false negative rate can be controlled below 6%. This approach will assist manual monitoring personnel in evaluating product quality.

Influence of surface roughness on laser ultrasonic detection for laser powder bed fusion manufactured 316L stainless steel

In view of the non-destructive and non-contact features, laser ultrasonic (LU) technology has long been the effective method to detect tiny defects for laser powder bed fusion (LPBF) additive manufactured specimens. Of larger concern is the variation and the corresponding mechanism on tested results of LU detection as the property of LPBF additive manufactured specimen is changed. Aiming at the property of surface roughness, this work investigated the propagation characteristics of excited ultrasonic waves in LPBF additive manufactured 316L stainless steel with different surface roughness, as well as the interaction between ultrasonic waves and artificial submillimeter holes. Both numerical simulated and experimental study were conducted. Simulated results revealed that the amplitudes of longitudinal wave (L wave) and its echo wave L1 at the holes exhibited a discernible increase as the surface was coarser. The increase in surface roughness was detrimental to the resolution of defect detection as was expected from the increased amount of noise. LPBF fabrication and the subsequent LU pulse-echo detection were conducted for 316L stainless steel. Both B-scan and C-scan were able to detect the holes with the diameter of 0.6 mm. The speckle phenomenon deriving from the increase in surface roughness emerged, corresponding to the increased ultrasonic signal energy but deteriorated resolution of detected images. It is feasible to optimize LU detected effect by minimize the surface roughness of tested specimens.