High-resolution imagery captured by drones can detect critical components on high-voltage transmission towers, providing inspection personnel with essential maintenance insights and improving the efficiency of power line inspections. The high-resolution imagery is particularly effective in enhancing the detection of fine details such as screws. The QYOLO algorithm, an enhancement of YOLOv8, incorporates context queries into the feature pyramid, effectively capturing long-range dependencies and improving the network’s ability to detect objects. To address the increased network depth and computational load introduced by query extraction, Ghost Separable Convolution (GSConv) is employed, reducing the computational expense by half and further improving the detection performance for small objects such as screws. The experimental validation using the Transmission Line Accessories Dataset (TLAD) developed for this project demonstrates that the proposed improvements increase the average precision (AP) for small objects by 5.5% and the F1-score by 3.5%. The method also enhances detection performance for overall targets, confirming its efficacy in practical applications.
Cross-domain fault diagnosis for rolling bearings under unseen working conditions is a challenging yet essential task, as the distribution bias significantly degrades the performance of data-driven methods. Causality-inspired domain generalization aims to address this challenge, with existing methods primarily focusing on alignment- or gradient-based operations from an entire signal perspective, which, however, overlooks the risk that biased alignments may induce spurious correlations and misguide learning. To this end, we reformulate the problem from a more physical and fine-grained perspective by treating fault-irrelevant frequency bands as confounders and aiming at localizing causal frequency bands encoded with robust and interpretable fault-related features to explicitly extract causal features. We propose a frequency band-aware method with a cyclostationarity-enhanced representation. Specifically, we first introduce a representation based on spectral correlation density (SCD) for wavelet packet-decomposed frequency bands. Then, treating cycle frequencies as channels, an adaptive feature extractor is designed based on a mixture-of-experts (MoE) block with a multi-view router, which integrates the views of spectrum, frequency band, and entire sample, to adaptively extract features across samples. In addition, prior knowledge guidance is introduced to enhance robust features. With cycle frequency-level features of frequency bands, a frequency band-aware attention module based on a tokenized Transformer, enhanced with an entropy-based sparsity regularization, is designed to model inter-band dependencies and localize fault-related frequency bands for diagnosis. Experiments are conducted on Case Western Reserve University (CWRU), Paderborn University (PU), and Harbin Institute of Technology (HIT) bearing datasets, and the proposed method shows effectiveness and interpretability across transfer tasks with different spans.
To address the problems of low detection accuracy of rolling bearings under different loads and the difficulty of effectively identifying the lack of labelled data, a rolling bearing fault diagnosis method combining GADF-DFT image coding and Multi-kernel domain coordinated adaptation network is proposed. Firstly, the vibration signal is converted into a two-dimensional image using GADF coding technology, and then the GADF image is converted into the frequency domain using discrete Fourier transform to extract deeper feature information. Combined with the multi-source domain adaptive method, the public feature extraction module is used to initially achieve feature mining of the image; the MK-MMD algorithm of the domain-specific adaptive module reduces the difference in feature distribution between the source and target domains; and the final classification difference minimization module reduces the problems caused by the classification errors that may be generated by the different domain classifiers due to the fact that the data samples are located near the category boundaries. The test uses the Case Western Reserve University dataset and divides the dataset with different operating conditions as the source and target domains, and the test results show that the proposed model demonstrates its effectiveness in responding to the complex operating condition changes in rolling bearing fault detection in multiple operating condition migration tasks, good adaptability and robustness, and is able to achieve accurate fault diagnosis under different operating conditions.
Cross-domain fault diagnosis under class imbalance conditions for rolling bearings is a challenging yet essential task, as the distribution bias and class imbalance significantly degrade the performance of data-driven methods. Recently, many boundary optimization algorithms have emerged to address this problem, but they have limitations in the adaptivity and controllability of boundary adjustment. In this work, we propose a feature-level decision boundary optimization strategy built upon the hyperspherical feature space with a mixture of von-Mises Fisher distribution and address the class-imbalance problem from three aspects: feature diversification, which diversifies features based on orthogonal information to overcome the insufficient feature extraction due to limited fault data; inter-fault margin sufficiency, which operates with feature-level class-sample-wise margins to enlarge fault boundary margins adaptively; normal-fault boundary compactness, which synthesizes near-in distribution fault samples to reinforce the normal-fault boundary. Experiments are conducted on Case Western Reserve University, Paderborn University, and the proposed method shows diagnostic effectiveness and boundary improvement across transfer tasks with different spans under different imbalance ratios.