Defect inversion is a critical quantitative assessment technique to evaluate the integrity of energy transportation systems. In practice, non-stationary states and frequent situation switches in real-world industrial processes make it difficult to acquire ideal sensor signals with precise annotations, there by compromising accurate defect inversion. Simulation systems is a viable solution. However, the dramatic inter-domain disparity between simulated and real system poses challenges to existing studies. Moreover, intra-domain divergence within the real energy transportation systems is also unexplored. To address the above issues, an intra- and inter-domain interactive learning method is proposed for simulation-to-real (sim2real) defect inversion. Specifically, reality-aware feature stylization is proposed which diversifies simulated signals with real style variants, so as to alleviate the domain gap initially. Then, bi-level inter-domain alignment is proposed to not only penalize hard-to-align samples in abstract feature space, but also innovatively exploits intrinsic geometric relations of sensor signals to guide inter-domain gaps. Next, physical-informed intra-domain alignment is proposed where a similarity matrix is constructed to pursue the consistency between sensor signals and physical knowledge, and it is also served as pseudo-labels to constrain inter-domain adaptation, so that inter- and intra- domain adaptation interact with united strength to enhance information exchange. Finally, the proposed method is systematically validated on energy transportation experimental platform and competitive results are obtained, which shows its promise in real-world industrial processes.

Remarkable progress has been achieved in the detection and segmentation of the baseline; however, for high-level visual tasks in complex scenes (e.g., dense, occlusion, scale diversity, high background noise, etc.), existing frameworks often fail to provide satisfactory performance. To further improve the object recognition ability, this article introduces a leader-based multiexpert mechanism into the detection and segmentation tasks. In this work, we first design a leader-based attention learning layer to fully integrate multilevel features from the backbone network, which can effectively obtain global semantics and assign instructions to detection experts. Then, we propose multiple feature pyramids with dual fusion paths to replace the traditional single pipeline using semantic and spatial allocators. With this strategy, we can further establish deep supervision for multiple experts during training and sufficiently utilize the multiexpert detection results from leaders' assignments during reasoning, thereby comprehensively improving the performance of the model in complex scenarios. In the experiment, we established ablation studies and performance comparisons on COCO 2017 detection and segmentation tasks. Finally, we demonstrated the model's performance in three complex application scenarios (remote sensing, autonomous driving, and industrial fields), and the results showed our advantages.

Pipeline system, as a vital industrial infrastructure, inevitably suffers from corrosion due to long-term service, which poses great threat to safe and stable energy transport. Many intelligent quantitative assessment technologies have emerged to evaluate the pipeline corrosion condition. However, they heavily rely on the quality and quantity of corrosion data. For industrial scenarios, the acquisition of pipeline corrosion data is costly with precision equipment for excavation works, and the annotation task is also labor-intensive, which severely limits their applicability. To address these issues, a mechanism-aided bilevel knowledge transfer framework is proposed to achieve effective quantitative assessment of pipeline corrosion condition. First, a high-fidelity mechanism model is designed to simulate the pipeline corrosion process and generates sufficient simulation data to alleviate the dependence on real-field resources. Then, a bilevel distribution-aware idea is proposed to transfer knowledge from simulated to real corrosion data, which is our original effort to focus on both feature complexity and label finesse, so as to accommodate the real-world corrosion for sensible domain adaptation. Next, we propose the idea of divide-and-conquer based on variational integration embedding (VIE) to maximize model performance improvement with minimal expert workload, where unreliable corrosion data are asked to expert annotation, while others are utilized via pseudolabels obtained from VIE. In the experiments, the performance of our method based on both hardware test platform and practical application case is systematically validated, and the competitive results indicate that our method has great potential in intelligent pipeline systems.

Fault transfer diagnosis is a key technology to ensure the reliability and safety of industrial systems, the core of which is to identify the health status of the equipment among different working conditions with multiclassification methods. However, most of them are based on a closed-set assumption that the label space among different working conditions is consistent, which is hard to satisfy in a practical industrial environment as unknown faults would inevitably occur during operation, i.e., the open-set fault transfer diagnosis (OSFTD) problem. Moreover, during the transfer process, unnecessary source-specific knowledge tends to be adapted, which brings about biased diagnostics on both domain and category. Aiming at this issue, an OSFTD framework, coined as knowledge transfer and reinforcement based on biunbiased neural network (KTR-BUNN), is proposed. First, a domain-unbiased knowledge transfer subnet is proposed, including an uncertainty-aware fault transferability evaluator (FTE) that estimates the transferability of target-domain samples unbiasedly to guide distribution alignment of known faults and a triple-tier unknown fault separator (UFS) that takes transferability as the criterion to extrapolate unknown faults. Second, a class-unbiased knowledge reinforcement subnet is designed to promote the recognition of fault semantic features at the embedding space, where fault knowledge graphs (FKGs) are constructed to describe the relationships between fault types, and they are optimized by a contrastive fault correlation loss, so that fine-grained class-level fault features can be further aligned. The knowledge transfer and knowledge reinforcement mechanisms work jointly to facilitate the performance of OSFTD. Finally, extensive experimental results conducted on diverse diagnostic tasks illustrate the superiority of the proposed KTR-BUNN.

This article proposes a practical and generalizable object detector, termed feature extraction-fusion-prediction network (FEFP-Net) for real-world application scenarios. The existing object detection methods have recently achieved excellent performance, however they still face three major challenges for real-world applications, i.e., feature similarity between classes, object size variability, and inconsistent localization and classification predictions. In order to effectively alleviate the current difficulties, the FEFP-Net with three key components is proposed, and the improved detection accuracy is proved in various applications: 1) Extraction Phase: an adaptive fine-grained feature extraction network is proposed to capture features of interest from coarse to fine details, which effectively avoids misclassification due to feature similarity; 2) Fusion Phase: a bidirectional neighbor connection network is designed to identify objects with different sizes by aggregating multilevel features and 3) Prediction Phase: in order to improve the accuracy of object localization and classification, a task specific prediction network is presented, which sufficiently exploits both the spatial and channel information of features. Compared with the State-of-the-Art methods, we achieved competitive results in the MS-COCO dataset. Further, we demonstrated the performance of FEFP-Net in different application fields, such as medical imaging, industry, agriculture, transportation, and remote sensing. These comprehensive experiments indicate that FEFP-Net has satisfactory accuracy and generalizability as a basic object detector.

Process fault diagnosis is of great importance to ensure the safe and stable operation of industrial systems. Many existing deep-learning-based process fault diagnosis methods assume that the samples are sufficient and obey the same distribution; however, it is almost impossible to achieve in practical industrial applications due to changing working conditions and the high cost of acquiring fault samples, which leads to a prominent performance degradation. In essence, those methods do not fully exploit the intrinsic and relevant knowledge under different working conditions. To address the above issue, a knowledge-mining-based semantic-aware network (KMSA-Net) is proposed in this article. First, a self-correlation knowledge mining subnet is proposed, where unshared attention mechanism is designed to extract knowledge inherent in each working condition so that the discriminative features can be captured. Second, a cross-correlation knowledge mining subnet is proposed, where we develop a fault relational knowledge graph so as to explicitly constrain the local consistency between the source domain, target domain, and cross-domain. Third, a semantic-aware knowledge transfer subnet is designed to impose a semantic constraint during knowledge transfer by encouraging the output of KMSA-Net to be consistent and distinguishable. These three subnets are jointly trained and then applied for cross-domain industrial process fault diagnosis. Finally, benchmark simulated experiments and real-world application experiments are conducted, and the experimental results validate the effectiveness and superiority of the proposed method.

Pipeline magnetic flux leakage (MFL) detection is an efficient and energy-saving nondestructive testing (NDT) method. However, under the high-speed detector, MFL signals become distorted with the influence of velocity effects, which adversely affects the accuracy of defect size inversion. The essential cause is the distorted signal multiplicity by velocity effects. In response to this issue, a high-precision defect size inversion method is proposed for the first time, which is called knowledge-guided contrastive fusion network (KCF-Net). First, MFL and eddy current (EC) mechanisms are analyzed, which are concluded that the sensitivity of EC signals to speed is much lower than that to defect sizes, so that EC and MFL abstract features are mined to improve the sensitivity of defect sizes. Moreover, MFL mechanism representations are mined to supervise neural networks to enhance the interpretability of the network. MFL and EC knowledge including abstract features and mechanism representations is fused to highlight the disparities between undistorted and distorted signals and enrich available information. Then, joint decision-making is proposed to eliminate the instability of fusion knowledge and enhance the universality and effectiveness of defect inversion. Finally, the experiments prove the effectiveness of KCF-Net. The length, width, and depth inversion MAEs of measured signals reach 2.2676, 1.6185, and 0.5664, respectively.

Defect detection aims to locate and classify defects in images, which is a necessary yet challenging task in industrial product quality monitoring. The current anchor-based detectors have weak generalization performance due to their inability to consider numerous scale priors. Moreover, the basic networks lack the ability to dynamically capture and utilize multiscale feature representations, resulting in low accuracy in industrial defect detection. To counter these challenges, an efficient anchor-free detector with dynamic receptive field assignment (DRFA) and task alignment is proposed. First, a feature pyramid structure with DRFA is innovatively designed to sufficiently extract multiscale feature representation and flexibly adjust the receptive field to detect diverse defects. Second, a task decoupling prediction mechanism is proposed to improve localization and classification prediction capabilities by introducing feature reassembly and task-specific information enhancers. Next, an anchor-free-based deep supervision with task-aligned is presented to encourage both to make accurate and consistent predictions, thereby effectively improving the overall detection performance. Finally, three industrial defect datasets (NEU-DET, PCB, WELD) are employed for experiments. The results show that the proposed method achieves 5.3% higher average AP than other state-of-the-art detectors.

Class-imbalanced weld defect recognition, which realizes defect recognition via learning features of class-imbalanced X-ray images, is an emerging but challenging task. Nevertheless, the existing studies on the class-imbalanced problem mainly focus on large-scale data, and it is difficult to extract high-quality features from the insufficient industrial data, resulting in weak recognition performance. To address the above issue, this article comprehensively learns features from the perspective of basic-class and cross-class, on this basis, a novel hybrid feature learning model for class-imbalanced weld defect recognition is proposed. First, an image acquisition method completed by photographing, scanning, and sampling is designed to collect the class-imbalanced X-ray images. Second, a hybrid feature learning model is proposed to learn the distinctive and effective features from acquired images, so that the class-imbalanced data are mapped to a balanced feature distribution. Third, with the distinguishable features learned by our hybrid feature learning model, an unbiased defect recognition model can be trained to recognize different types of defects. The practical weld data W-PPLN, W-MTL, and W-GDXray are adopted in the experiments, and the experimental results show that our method outperforms the state-of-the-art methods on the task of class-imbalanced weld defect recognition.