Lithium battery surface defect detection based on the YOLOv3 detection algorithm

The technique for target detection based on a convolutional neural network has been widely implemented in the industry. However, the detection accuracy of X-ray images in security screening scenarios still requires improvement. This paper proposes a coupled multi-scale feature extraction and multi-scale attention architecture. We integrate this architecture into the Single Shot MultiBox Detector (SSD) algorithm and find that it can significantly improve the effectiveness of target detection. Firstly, ResNet is used as the backbone network to replace the original VGG network to improve the feature extraction capability of the convolutional neural network for images. Secondly, a multi-scale feature extraction (MSE) structure is designed to enrich the information contained in the multi-stage prediction feature layer. Finally, the multi-scale attention architecture (MSA) is fused onto the prediction feature layer to eliminate the redundant features’ interference and extract effective contextual information. In addition, a combination of Adaptive-NMS and Soft-NMS is used to output the final prediction anchor boxes when performing non-maximum suppression. The results of the experiments show that the improved method improves the mean average precision (mAP) value by 7.4% compared to the original approach. New modules make detection much more accurate while keeping the detection speed the same.

Surface defect detection aims to classify and locate a certain defect that exists in the target surface area. It is an important part of industrial quality inspection. Most of the research on surface defect detection are currently based on convolutional neural networks (CNNs), which are more concerned with local information and lack global perception. Thus, CNNs are unable to effectively extract the defect features. In this paper, a defect detection method based on the Swin transformer is proposed. The structure of the Swin transformer has been fine-tuned so that it has five scales of output, making it more suitable for defect detection tasks with large variations in target size. A bi-directional feature pyramid network is used as the feature fusion part to efficiently fuse to the extracted features. The focal loss is used as a loss function to weight the hard- and easy-to-distinguish samples, potentially making the model fit the surface defect data better. To reduce the number of parameters in the model, a shared detection head was chosen for result prediction. Experiments were conducted on the flange surface defect dataset and the steel surface defect dataset, respectively. Compared with the classical CNNs target detection algorithm, our method improves the mean average precision (mAP) by about 15.4%, while the model volume and detection speed are essentially the same as those of the CNNs-based method. The experimental results show that our proposed method is more competitive compared with CNNs-based methods and has some generality for different types of defects.

Mechanical parts often have surface defects such as scratches, cracks and pits during the production process, which seriously affect product quality and service life. To address this challenge, this paper proposes a method for detecting and quantifying surface defects of mechanical parts based on image processing. First, this paper uses high-resolution image acquisition technology to obtain surface images of mechanical parts, and improves image quality through preprocessing steps. Then, this paper uses edge detection, morphological operations, and image segmentation algorithms to extract the features of surface defects, thereby realizing automatic identification of defective areas. Finally, this paper combines the Faster R-CNN (Faster Region-Based Convolutional Neural Network) algorithm to further refine the classification and quantification of defects and obtain the specific location, size and type of defects. Experimental results show that among the 500 mechanical parts samples tested, the proposed method has a mean average precision (mAP) of 0.93 in surface defect detection. Compared with other image processing methods, this method has significantly improved defect detection accuracy and speed. Especially when processing higher-resolution images, the Faster R-CNN model can achieve efficient defect classification and quantification to meet the needs of real-time detection in industrial production.

As the high voltage circuit breaker works, the metal parts are prone to many types of mechanical defects, such as plastic deformation, metal loss, corrosion, and so on. The defects will inevitably decrease the performance, mechanical reliability, lifetime, and other critical parameters of the breaker. Therefore, the detection of mechanical defects is of great help to reduce the mechanical failure rate of circuit breakers. In the present paper, a novel method based on edge detection and deep learning is proposed to detect the mechanical defects of the high voltage circuit breaker. Firstly, the high resolution images of the circuit breaker are photographed, and the components in the images are segmented in the minimum range by the contour detection algorithm. After binarization and morphological processing of the segmented images, the edge is drawn with the improved edge detection algorithm. However, the components of the circuit breaker are complicated, which results that the edge features of the critical component are difficult to extract. Then, the specific contour of components is detected by the depth estimation and segmentation methods. Secondly, a deep learning platform base on Tensorflow is established, which is an optimization method based on Convolutional Neural Network. In order to enhance the feature extraction capacity, a convolution kernel is inputted into a sparse self-coding network to proceeds with the optimal pre-training. Based on the above method, the features of images are automatically learned, and each convolution carries for the equilibrium of image entropy of convolution kernel by introducing similarity constraint rule. Finally, 3 types of mechanical defects, including plastic deformation, metal loss, corrosion, images and non-defect images as input are exploited to validate the CNN with convolution kernel optimized. The experimental results show that the average accuracy of the approach is better than the traditional convolution neural network model and has good feature extraction ability and generalization ability. In conclusion, combined with the improved edge detection and deep learning algorithms, the detection of typical mechanical defects can be well performed. For the collected images of circuit breaker, so as to reduce the impact of noise on defect accuracy of mechanical, the proposed method provides a reference for the non-destructive mechanical defect detection of high voltage circuit breaker.

For the detection of wood surface defects, a convolutional neural network has a low detection efficiency and insufficient generalization ability, so it does not meet the requirements of online detection. Aiming to solve the above problems, the YOLOv3 baseline model, which has the advantage of multi-objective dynamic detection, was improved and applied to the online detection of wood surface defects. To solve the problem of the poor generalization ability of the network, GridMask was used to enhance the data and improve the robustness of the network. In order to solve the problem of the considerable amount of network parameter calculations and insufficient real-time performance, the residual block of the backbone network was changed to a Ghost block structure to achieve a lightweight model. Finally, the confidence loss function of the network was improved to reduce the influence of simple samples and negative samples on model convergence. The experimental results showed that, compared with the original network, the improved algorithm increased the mean average precision by 5.73% and the detection speed was increased to 28 frames per second (an increase of 11), which met the requirements for real-time industrial detection.

The complex interaction between laser and material in Laser Wire Direct Energy Deposition (LW-DED) Additive Manufacturing (AM) benefits from process monitoring methods to ensure process stability and final part quality. Understanding the relationship between process parameters and melt pool geometrical characteristics can be used to effectively monitor and in-process control the process, as the melt pool geometrical characteristics serve as crucial indicators of process stability and quality.This study presents a novel in-situ monitoring approach for LW-DED, utilizing process images for melt pool segmentation, melt pool geometrical characteristics estimation, process stability assessment, and bead geometry prediction. The segmentation of melt pool objects was successfully accomplished using Convolutional Neural Networks (CNN)-based models, enabling the prediction of essential parameters such as melt pool area, height, width, center of area, and the center point of the bounding box enclosing the melt pool. Multiple models were compared regarding the accuracy and processing speed using a controlled central composite design and random experiments. We used an Inconel alloy 625 wire and two distinct substrate materials for deposition, a coaxial laser welding head with a 3 kW fiber laser, and an off-axis welding camera for monitoring.Among the CNN architectures evaluated, YOLOv8l demonstrated superior accuracy with mean Average Precision (mAP) values of 0.925 and 0.853 for Stainless Steel (SS) and low carbon steel (S355) substrates, respectively. Additionally, YOLOv8s exhibited a notable processing speed of over 114 frames per second, which indicates its suitability for real-time process control. Furthermore, the results indicate a significant correlation between process parameters and melt pool geometry variables. Notably, a clear correlation was established between melt pool characteristics and bead geometries obtained through microscopic examinations, including penetration depth and heat-affected zone. The analysis revealed a significant correlation for the bead area and width parameters. In relation to the bead height, while the correlation exhibited a lower magnitude compared to bead area and width, it remained responsive. In addition, the tensor masks derived from the developed models have proven to be highly effective in accurately predicting bead geometries.The results demonstrate the effectiveness of YOLO-based algorithms for detecting and segmenting the melt pool. Statistical analysis confirms the significance of stabilized process data and the accuracy of melt pool geometric models. We demonstrate that integrating advanced monitoring and control techniques using artificial intelligence methods like CNN can facilitate process stability and quality control.

Investigation of the use of extended surfaces in paraffin wax phase change material in thermal management of a cylindrical lithium-ion battery: Applicable in the aerospace industry

Aiming at the existing fabric defect detection algorithms with their suboptimal detection accuracy, high model complexity, difficult deployment to the edge of the device, and insufficient ability to meet the problem of real-time detection of fabric defects, we propose a lightweight fabric defect detection model (LWFDD-YOLO) based on the improved YOLOv8n. First, a generalized efficient layer attention network with selective kernel attention (GELAN_SKA) is proposed to replace the C2f module, and selective kernel attention is added to the module to adjust the weights of the convolution kernel according to the different features of different scales, reduce the use of computational resources, thus improving the model’s detection performance and efficiency. Second, a cascaded group attention (CGA) mechanism is added to provide a different input segmentation for each head to enhance the feature diversity of the input attention heads and improve the model’s computational efficiency. An ultra-lightweight dynamic up-sampling operator (Dy_sample) is introduced, which employs a point-sampling-based approach to reduce the consumption of computational resources and improve model performance. Finally, the YOLOv8l framework is utilized to construct a complex model teacher network and the features learned from the teacher network are transferred to the lightweight network proposed in this paper, thus further improving the model performance of the algorithm. The experimental results show that on the self-built fabric defect dataset, the accuracy, recall, and mean average precision ( mAP ) of our algorithm reach 89.4%, 85.2%, and 87.9%, which are 8.9%, 7.2%, and 4.5% higher than the original model, respectively, the number of parameters of the model decreases by 23.4%, the amount of GFLOPs decreases by 25.6%, and the file size is only 9.2 MB. The detection speed can reach 163.4 FPS on GPU. On the AliCloud Tian Chi dataset, the accuracy, recall, and mAP were also improved by 7.4%, 2%, and 4.4%, respectively. The LWFDD-YOLO algorithm proposed in this study can realize real-time detection of fabric defects with relatively obvious improvement in accuracy, and the model requires less memory and is easier to deploy to edge devices, so can be used as a reference for real-time detection of fabric defects.

We propose a lightweight detection algorithm based on the Single Shot MultiBox Detector (SSD) algorithm in order to facilitate sheep management and to realize sheep facial identification, and we take the self-constructed dataset as the research object. First, the SSD replaces the VGG16 backbone network with MobileNetv3, a lightweight neural network, to create a hybrid model that is much smaller. Second, the ECA attention mechanism is incorporated into the backend of the 72 × 160 bottleneck layer. Finally, the SmoothL1 loss function is substituted with the BalancedL1 loss function. The optimized model’s size decreases significantly from the original SSD’s 132 MB to just 22.4 MB. It achieves a mean average precision of 83.47% and maintains an average frame rate of 68.53 frames per second. Compared to the basic SSD model, the mean average precision has increased by 3.25 percentage points, the model size has decreased by 109.6 MB, and the detection speed has improved by 9.55 frames per second. In comparative experiments using the same dataset with different object detection models, the proposed model outperforms the SSD, Faster R-CNN, Retinanet, and CenterNet in terms of mean average precision, with improvements of 3.25 percentage points, 4.71 percentage points, 2.38 percentage points, and 8.13 percentage points, respectively. The detection speed has shown significant improvements, increasing by 9.55, 58.55, 53.1, and 12.37 frames per second, respectively. The improved model presented in this paper significantly reduces the model’s size and computational requirements while maintaining an excellent performance. This provides a valuable reference for the digitalization of animal husbandry and livestock farming.

The detection of defects on automotive coated surfaces is of paramount importance to ensure the quality of automotive appearance. However, due to the challenges posed by the lack of sufficient samples of product-specific coated defects, the uneven number of species, and the difficulty of detecting defects in small targets due to the complex background of the body surface. The detection of defects on automotive coated surfaces still relies on experienced manual labor to a significant extent. To address these issues, Poisson fusion is employed to enhance the data for generating defect images, thereby mitigating the limitations of insufficient coated defect samples and an imbalanced distribution of defect species. Based on this, a new YOLOv8-SC defect detection model is proposed, in which the C2f-Star is used to replace the original C2f structure. This aims to improve the model’s ability to extract defects of varying sizes in complex backgrounds, while reducing the number of model parameters and the amount of computation. The content-guided attention fusion (CGAFusion) module, situated before each detection head of the model, has been designed to enhance the model’s capacity to extract defects of varying sizes and complexity. This is achieved through the adaptive fusion of low-level and high-level features. The proposed model’s superiority in terms of detection metrics has been demonstrated through the use of example validation. The experiments were conducted on the enterprise dataset CPD-DET and the public dataset NEU-DET. The results demonstrated that the defective image generation and data enhancement method could significantly improve detection performance and have good generalization. The proposed YOLOv8-SC reduces the model parameters by 12.2% compared to the normal model. Additionally, the mean average precision (mAP) at 50% and 50%–95% thresholds are enhanced by 4.7% and 2.3%, respectively. Furthermore, the accuracy of the model exhibits a discernible positive growth trend with an increase in the sample set size. The design of an automotive coated surface defect detection system is presented at last. When deployed in industrial settings, this system can facilitate the intelligent detection of coated defects.

Overhead transmission lines, which are directly related to the safe and stable operation of the power grid, are important channels for the transmission of electrical energy. Accurate and efficient detection of electrical components in transmission lines is crucial for maintaining power grid stability and safety. Existing methods often encounter difficulties with issues such as small targets, complex backgrounds, and varying scales. In this paper, we propose an enhanced version of YOLOv8 specifically designed for multi-object detection in transmission line inspections. Secondly, with the aim of enhancing the accuracy and robustness of electrical component identification as well as defect detection, the Faster-Block module, GhostNet, SPD-Conv, and the triplet attention module are incorporated into the original YOLOv8. Finally, improved YOLOv8 and the mainstream deep learning algorithms are trained and tested on the self-built dataset. The mean average precision (mAP) value of improved YOLOv8 reaches 86.73%, which is nearly 36%, 14% and 10% higher than that of Faster R-CNN, DETR, YOLOv8, respectively. In addition, the inference speed of the improved YOLOv8 reaches 38.1 frames per second (FPS), which is 4.4% faster than that of YOLOv8. Ablation studies and comparative experiments demonstrate the effectiveness of our approach in detecting insulators, dampers, clamps, and defects, especially for small targets and complex scenes.

Internal defects detection and classification in hollow cylindrical surfaces using single shot detection and MobileNet

With the wide application of lithium batteries in new energy vehicles, consumer electronics, energy storage systems and other fields, their quality and safety not only directly affect the performance and life of the product, but also relate to the safety of users' lives and properties. During manufacturing and transportation, various defects will be produced on lithium batteries, such as scratches, stains, and pits. Thus, defect detection is crucial for ensuring battery quality. However, the complex background texture on the bottom surface of lithium batteries, coupled with the small size of some defects and the low contrast between defects and background, poses a great challenge for traditional detection methods. To address these issues, an improved YOLO11-based lithium battery bottom surface defect detection model (called DFCL-YOLO11) is proposed in this paper. In the proposed model, a new downsampling module is presented to retain more spatial information and enhance the feature representation capability. Then, the C3k2 module of the traditional YOLO11 model is improved by combining FasterBlock and Convolutional Gated Linear Unit (CGLU), achieving higher detection accuracy with reduced computational complexity. Finally, the Large Separable Kernel Attention (LSKA) module is introduced into the proposed defect detection, to reduce information loss caused by the spatial pyramid pooling operations. In addition, a Lithium Battery Defect (LBD) dataset is constructed by collecting the images of the bottom surface defects of Tesla 21700 lithium batteries and labelling the defects in this study. Experimental results demonstrate that the proposed DFCL-YOLO11 model outperforms baseline YOLO11 and other mainstream defect detection models.

Lithium batteries represent a pivotal technology in the advancement of renewable energy, and their enhanced performance and safety are vital to the attainment of sustainable development goals. To solve the issue of the high missed detection rate of minimal defects on end face of lithium battery shells, a novel YOLO-based Minimal Defect Detection algorithm, named YOLO-MDD, is proposed. Firstly, aimed at the problem of insufficient data, a dataset of defects on the end face of lithium battery shells is constructed and annotated. Secondly, a YOLO-MDD network which includes a feature extraction module and a four-scale detection head for detecting defects of various scales is improved. Here, deformable convolution and an attention module are ingeniously embedded into the backbone of YOLO to capture more detailed and accurate information on object defects, and the four-scale head is used to handle the significant differences in the size and shape of defects on lithium battery shells. Finally, a hybrid loss including localization loss with normalized Wasserstein distance (NWD), classification loss, and confidence loss is designed to optimize our model to further enhance its sensitivity to minimal defects. The experimental results show that the proposed YOLO-MDD has a mean average precision of 80% for the defect detection of the lithium battery shells, especially with a minimal defect rust spots mean average precision of 74.1% and a recall rate of 71.5%, which is superior compared with other mainstream detection algorithms and provides the technical support necessary to achieve the goals of energy and environmental sustainability.

As people have higher requirements for object detection accuracy and adaptability, this research proposes a single-stage algorithm for thermal image target detection based on Yolo v4. The algorithm combines the visible light image and the thermal image to complement the image information, and uses the TMDF (Two Modol Differential Fusion) module to perform feature fusion on the visible light image and the thermal image and then enter the network for training. Adding the DCT(Discrete Cosine Transform) adjustment network to adjust the final output can finally increase the mAP (mean Average Precision) value of the model on the KAIST data set by 3.7%, and MR (missed detection rate) value decreased by 3.5%.