Yingyi Liu

The widespread use of complex-structured printed circuit boards (PCBs) in high-tech equipment has increased reliability and fault detection demand. Low resolution and significant differences in infrared thermal imaging feature scales limit the application in PCB fault identification using existing deep learning models. A PCB fault identification framework based on multimodal infrared thermal imaging multiscale feature fusion and heat source attention is proposed to address this challenge. The framework significantly enhances the accuracy and robustness of fault detection by encoding multiscale features from multimodal infrared thermal images. In the proposed framework, the input images are extended into RST multimodal infrared thermal images, which include the spatiotemporal variation rate of the temperature field, enriching the thermal image feature. Additionally, the framework integrates feature pyramid networks (FPN) and heat source attention (HSA) modules into the encoding network, improving the ability to express fault-related features. Experimental validation on a two-phase drive circuit demonstrates that the proposed framework improves classification accuracy by 4.0% compared to existing deep convolutional neural networks, showing high robustness against focus blur and pixel failures. More importantly, the framework accurately detects faults at early stages. This study improves the accuracy and efficiency of PCB fault identification and provides technical support for the reliability monitoring of complex electronic equipment.

Infrared sensors are critical equipment for monitoring substations, but they are subjected to complex multi-physical field stresses in substation applications, leading to a significantly higher failure rate of infrared sensors in substations compared to other applications. Effective fault diagnosis of infrared sensors is of great importance for improving the safety of substations. This article proposes a method for fault diagnosis of infrared sensors. First, the network weights are trained based on the ImageNet dataset and kept fixed. Then, a transfer learning approach is used to fine-tune the model based on a dataset of infrared sensor failures, resulting in the classification results. To verify the effectiveness of the proposed method, the improved transfer learning model is compared to the model before improvement. Experimental results show that the proposed method greatly reduces training time and improves classification accuracy compared to the previous model. In conclusion, the method presented in this article offers a new approach for fault diagnosis of infrared sensors, which can contribute to enhancing the reliability of infrared sensors in substations.

High-voltage direct current (HVdc) transmission technology plays a crucial role in modern electrical power industries. However, with the increasing voltage of transmission lines, audible noise has become a significant factor limiting the planning and construction. To research audible noise’s generation mechanisms and prediction method, this article models corona discharge on a rod–plate structure and calculates the generation and movement of particles based on 23 gas-phase reactions and eight surface reactions. We integrate the model with an acoustic source framework, which employs the electroacoustic conversion theory, solving the acoustic fluctuation equation to compute audible noise. The simulation results show a 7.5% discrepancy in the time-domain amplitude of audible noise compared to experimental measurements. The average and distribution of audible noise amplitudes from simulations and measurements at different voltages and electrode distances show a high consistency, demonstrating the accuracy of the proposed method. This approach contributes to understanding the mechanisms underlying audible noise and facilitates predictions through numerical calculations, providing a reference for designing HVdc transmission lines.

Non-contact printed circuit board (PCB) fault diagnosis has been widely applied in PCB detection and maintenance. Due to objective factors such as visual blind spots, low-loss circuit structure design, and frequency insensitivity, traditional algorithms based on visual and temperature features are limited in practice. Therefore, the algorithm based on electromagnetic features containing rich physical connotations and prominent frequency features has received attention in PCB fault diagnosis. Based on the basic principles of electromagnetic physics and PCB fault relationship, this article proposes a scalar magnetic field source feature and further improves feature performance by adding topological relationships of multi-faults to generate the fusion feature. The backbone of the PCB diagnosis model is established on the Transformer architecture, effectively utilizing self-attention and parallel computing mechanisms to explore the inner correlation between each group feature. The paper provides a new non-contact PCB fault diagnosis solution that enriches existing methods. Besides, through actual experiments setting up multi-fault PCBs, the feasibility of our process is proved based on the proposed features and models. The specific multiple indicators Overall Precision (OP), Per Class Precision (CP), Overall Recall (OR), Per Class Recall (CR), Overall F1 Measure (OF1), Per Class F1 Measure (CF1), Accuracy (ACC), Mean Average Precision (mAP) are 98.55%, 94.89%, 98.55%, 95.11%, 98.55%, 95.32%, 96.01%, and 97.27%.

UHF partial discharge sensors are key equipment for substation monitoring, but they are subject to complex multi-physical field stresses in substation applications, which leads to a significantly higher failure rate among UHF partial discharge sensors used in substations compared to other applications. Effective fault diagnosis is of great significance for improving the safety of substations. In this paper, we propose an improved model based on ViT (Vision Transformer), which effectively identifies the local features of the data by designing a sliding window mechanism, and has a good feature extraction capability for the feature library formed by UHF partial discharge sensors. The experimental results show that the diagnostic accuracy of the improved model, based on the ViT model, can reach 97.6%, which can effectively improve classification accuracy and shorten training times compared with the ViT model.

As electronic sensors and sensor networks advance, perception data are increasingly characterized by mixed attributes. Traditional anomaly detection methods predominantly focus on numerical attributes. In this paper, we introduce a weighted neighborhood information network (WNIN)-enabled anomaly detection method tailored for mixed-attribute data from electronic sensors and sensor networks. Firstly, we employ the analytic hierarchy process (AHP) to analyze the security of sensor networks, leveraging a hierarchical electronic sensor network model to construct a hierarchical perception security architecture for anomaly detection. Subsequently, a neighborhood information system is established to ascertain the relationships between data objects with mixed attributes. We then develop the WNIN to encapsulate the relationships, and a state-transferring probability matrix based on data object similarity is derived. Ultimately, a random wandering process within the WNIN is executed, and the importance of data objects is evaluated using the steady-state distribution vector, thereby determining the anomaly data. Simulation outcomes reveal that our proposed method attains superior anomaly detection rates compared with existing methods.

An overview of the developmental history, characteristics, and applications of infrared sensors are firstly provided in the paper. Subsequently, it delves into the sensing mechanisms of commonly employed infrared sensors in substations, namely the thermistor sensors, pyroelectric sensors, and photoconductive sensors. Thermistor infrared sensors gauge infrared radiation by tracking changes in material resistance with temperature. Pyroelectric sensors detect thermal changes in objects based on the thermal effects induced by infrared radiation. Photoconductive-type sensors leverage the photoconductive effect of semiconductors to convert optical signals into electrical signals, adapting to various light conditions. In the process of measurements, variables like temperature, distance between the target object and the infrared sensor, and electromagnetic interference impact accuracy. In extreme temperatures, the cooling precision of infrared thermal imaging systems may decline, affecting measurement accuracy. The distance between the target and sensor also influences results due to energy attenuation in the infrared radiation path through the atmosphere. Faraday’s law of electromagnetic induction warns of potential electromagnetic interference signals in substations, causing device breakdowns, output image halo, or increased internal noise. Finally, the paper envisions the future development of infrared sensors.

Classification tasks are pivotal across diverse applications, yet the burgeoning amount of data, coupled with complicating factors such as noise, exacerbates the challenge of classifying complex data. For algorithms that require a large amount of data, the annotation work for datasets is also exceptionally complex and tedious. Drawing upon existing research, this paper first introduces a novel semi-supervised category dictionary model based on transfer learning (SSDT). This model is designed to construct a more representative category dictionary and to delineate the associations of information across different domains, utilizing the lens of conditional probability distribution. This approach is particularly apt for semi-supervised transfer learning scenarios. Subsequently, the proposed method is applied to the domain of bearing fault diagnosis. This model is suitable for transfer scenarios; moreover, its semi-supervised characteristic eliminates the need for labeling the entire input dataset, significantly reducing manual workload. Experimental results attest to the model’s practical utility. When benchmarked against other 6 models, the SSDT model demonstrates enhanced generalization performance.