With the development of data-driven models, deep learning has been increasingly applied in the field of defect detection. However, the performance of deep learning models is greatly restricted by costly labeling and sample scarcity. One of the best approaches to solve the data imbalance problem is increasing the quantity and diversity of defect samples. Meanwhile, the current models based on generative adversarial network (GAN) cannot readily control the category and shape of generated defect samples, which results in inefficient data augmentation. Thus, to simultaneously achieve the category and shape adjustability of defect samples, a novel model named diversifying tire-defect GAN (DTD-GAN) is proposed in this article by integrating the advantages of latent space decomposition and feature decoupling algorithm. In this model, an auxiliary classifier is designed to control the defect category by discrete latent codes, and the mutual information maximization approach is adopted to diversify the tire-defect shape via continuous latent codes that greatly improve the diversity of the generated tire-defect samples. To illustrate the effectiveness of the proposed DTD-GAN model, extensive experiments on image quality and diversity assessment are carried out on the images generated by this model and other state-of-the-art approaches, and the achieved improvement in deep learning model performance in a tire-defect detection task is also discussed.

Automatic defect detection of tire has become an essential issue in the tire industry. However, it is challenging to inspect the inner structure of tire by surface detection. Therefore, an X-ray image sensor is used for tire defect inspection. At present, detection of defective tires is inefficient because tire factories commonly conduct detection by manually checking X-ray images. With the development of deep learning, supervised learning has been introduced to replace human resources. However, in actual industrial scenes, defective samples are rare in comparison to defect-free samples. The quantity of defective samples is insufficient for supervised models to extract features and identify nonconforming products from qualified ones. To address these problems, we propose an unsupervised approach, using no labeled defect samples for training. Moreover, we introduce an augmented reconstruction method and a self-supervised training strategy. The approach is based on the idea of reconstruction. In the training phase, only defect-free samples are used for training the model and updating memory items in the memory module, so the reproduced images in the test phase are bound to resemble defect-free images. The reconstruction residual is utilized to detect defects. The introduction of self-supervised training strategy further strengthens the reconstruction residual to improve detection performance. The proposed method is experimentally proved to be effective. The Area Under Curve (AUC) on a tire X-ray dataset reaches 0.873, so the proposed method is promising for application.

In tire defect detection, an X-ray image sensor is utilized to collect samples, upon which automatic defect detection methods are constructed. The performance of existing defect detection methods is reduced significantly when working condition shifts. Fortunately, domain adaptation can address this domain shift problem effectively. However, previous methods only align the marginal distribution or conditional distribution, while ignoring the unified adaptation of the two distributions, and may result in an unsatisfactory transfer effect. To this end, this article proposes a dual-domain adaptation model, which harnesses the adaptation of global marginal distribution and fine-grained conditional distribution simultaneously. Specifically, to learn the feature with the consideration of intraclass compact and interclass discriminative, the relative importance of marginal distribution and conditional distribution is measured by a dynamic adaptive factor. In addition, a multihead self-attention (MSA) mechanism is designed in the feature extractor for global context modeling. To evaluate the effectiveness of the proposed method, the tire defect detection task on five kinds of familiar defects, namely slack, bend, overlap, impurity, and bubble, was constructed. Compared with other existing domain adaptation models, the proposed model can achieve 96.39% detect detection accuracy. The experimental results demonstrate the effectiveness of the proposed method in tire defect detection under the domain shift.

The defect detection of industrial products is an essential task in industrial production. In the field of tire defect detection, X-ray imaging sensors are employed by tire factories to investigate the inner structure of the tires and identify defects. In recent years, deep learning has become a good alternative to manual inspection, but one major challenge of utilizing deep learning models is the requirement of large datasets with precisely annotated labels. In industrial applications, the process of annotating labels is often dependent on domain experts, which introduces inter- and intra-observer variability. Consequently, industrial datasets often contain noisy labels, which will deteriorate the performance of convolutional neural networks (CNNs) during the training process. In this study, we propose an approach called Prototype-based Supervised Contrastive Learning (PSCL) to address label noise problems. First, we utilize a supervised contrastive learning (SCL) framework to pull together features of samples belonging to the same class and produce class prototypes based on these features. Subsequently, label correction is performed using the updated class prototypes. In return, these corrected labels will contribute to the training of a reliable network, which ensures the provision of better prototypes. To assess the effectiveness of our method, we conduct extensive experiments on both synthetic noisy datasets and industrial tire datasets. The results demonstrate the robustness of our approach even in datasets with large proportion of noisy labels. Therefore, our approach provides a promising solution for mitigating label noise problems in industrial datasets, offering great potential for practical industrial applications.

Tire defect detection, as an important application of automatic inspection techniques in the industrial field, remains a challenging task because of the diversity and complexity of defect types. Existing research mainly relies on X-ray images for the inspection of defects with clear characteristics. However, in actual production lines, the major threat to tire products comes from defects of low visual quality and ambiguous shape structures. Among them, bubbles, composing a major type of bulge-like defects, commonly exist yet are intrinsically difficult to detect in the manufacturing process. In this paper, we focused on the detection of more challenging defect types with low visibility on tire products. Unlike existing approaches, our method used laser scanning technology to establish a new three-dimensional (3D) dataset containing tire surface scans, which leads to a new detection framework for tire defects based on 3D point cloud analysis. Our method combined a novel 3D rendering strategy with the learning capacity of two-dimensional (2D) detection models. First, we extracted accurate depth distribution from raw point cloud data and converted it into a rendered 2D feature map to capture pixel-wise information about local surface orientation. Then, we applied a transformer-based detection pipeline to the rendered 2D images. Our method marks the first work on tire defect detection using 3D data and can effectively detect challenging defect types in X-ray-based methods. Extensive experimental results demonstrate that our method outperforms state-of-the-art approaches on 3D datasets in terms of detecting tire bubble defects according to six evaluation metrics. Specifically, our method achieved 35.6, 40.9, and 69.1 mAP on three proposed datasets, outperforming others based on bounding boxes or query vectors.

Continual learning, the capability to learn new knowledge from streaming data without forgetting the previous knowledge, is a critical requirement for dynamic learning systems, especially for emerging edge devices such as self-driving cars and drones. However, continual learning is still facing the catastrophic forgetting problem. Previous work illustrate that model performance on continual learning is not only related to the learning algorithms but also strongly dependent on the inherited model, i.e., the model where continual learning starts. The better stability of the inherited model, the less catastrophic forgetting and thus, the inherited model should be elaborately selected. Inspired by this finding, we develop an evolutionary neural architecture search (ENAS) algorithm that emphasizes the Stability of the inherited model, namely ENAS-S. ENAS-S aims to find optimal architectures for accurate continual learning on edge devices. On CIFAR-10 and CIFAR-100, we present that ENAS-S achieves competitive architectures with lower catastrophic forgetting and smaller model size when learning from a data stream, as compared with handcrafted DNNs.