Wearing gloves during machinery operation in workshops is essential for preventing accidental injuries, such as mechanical damage and burns. Ensuring that workers are wearing gloves is a key strategy for accident prevention. Consequently, this study proposes a glove detection algorithm called YOLOv8-AFPN-M-C2f based on YOLOv8, offering swifter detection speeds, lower computational demands, and enhanced accuracy for workshop scenarios. This research innovates by substituting the head of YOLOv8 with the AFPN-M-C2f network, amplifying the pathways for feature vector propagation, and mitigating semantic discrepancies between non-adjacent feature layers. Additionally, the introduction of a superficial feature layer enriches surface feature information, augmenting the model’s sensitivity to smaller objects. To assess the performance of the YOLOv8-AFPN-M-C2f model, this study conducted multiple experiments using a factory glove detection dataset compiled for this study. The results indicate that the enhanced YOLOv8 model surpasses other network models. Compared to the baseline YOLOv8 model, the refined version shows a 2.6% increase in mAP@50%, a 63.8% rise in FPS, and a 13% reduction in the number of parameters. This research contributes an effective solution for the detection of glove adherence.
Wearing gloves while operating machinery in workshops is an essential precaution to prevent mechanical injuries and burns from high temperatures, among other potential hazards. Ensuring workers are properly equipped with gloves, which is a crucial measure in accident prevention. Glove images often occupy a minimal proportion of the frame and are easily obscured by cluttered backgrounds, especially with limited edge computing resources. Consequently, this study proposes a glove detection algorithm called YOLOv8-AFPN-M-C2f based on YOLOv8, offering swifter detection speeds, lower computational demands, and enhanced accuracy for workshop scenarios. This research innovates by substituting the head of YOLOv8 with the AFPN-M-C2f network, amplifying the pathways for feature vector propagation, and mitigating semantic discrepancies between non-adjacent feature layers. Additionally, the introduction of a superficial feature layer enriches surface feature information, augmenting the model's sensitivity to smaller objects. To assess the performance of the YOLOv8-AFPN-M-C2f model, we conducted multiple experiments using a factory glove detection dataset compiled for this study. The results indicate that the enhanced YOLOv8 model surpasses other network models. Compared to the baseline YOLOv8 model, the refined version shows a 2.6% increase in mAP@50%, a 90.1% rise in FPS, and a 13% reduction in the number of parameters. This research contributes an effective solution for the detection of glove adherence.
Owing to the difficulty of collecting adequate fault data from hoisting systems under complex working conditions, data-driven diagnostic methods suffer from decreased accuracy due to inadequate fault information. To deal with this problem, we introduce an intelligent few-shot fault diagnosis approach using a deep graph convolutional generative adversarial networks (DGCGANs) algorithm. The goal of the proposed method is to acquire a more comprehensive few-shot fault feature information, thus facilitating the generation and augmentation of scarce data. Results of two cases including laboratory and real-world industrial datasets demonstrate the viability and efficacy of our approach for diagnosing few-shot faults. Specifically, the proposed DGCGAN method outperforms existing advanced diagnostics, thereby offering superior accuracy in identifying hoisting system faults with limited data. Finally, the practicality and adaptability of the proposed DGCGAN-based fault diagnosis method are further substantiated through comprehensive ablation studies.
25 December 2024 in IEEE Transactions on Instrumentation and Measurement
Due to the challenges in gathering fault data from storage stacking machinery operating under high-speed and heavy-load conditions, the diagnostic accuracy of data-driven fault diagnosis methods is often compromised by a lack of sufficient fault information. To mitigate this issue, we propose an intelligent fault diagnosis strategy utilizing a novel persistent homology deep meta-transfer learning (PHDMTL) algorithm. This method aims to develop a more extensive feature representation, enhancing fault identification accuracy with limited data availability. Initially, a barcode chart of vibration data is generated through the persistent homology mechanism, from which the corresponding fault topology features are derived based on the chart. Subsequently, the fault features identified are fed into a specially designed deep meta-transfer model for multi-task learning to derive a pre-trained model. This pre-training model is further refined both externally and internally to produce an optimal initialization model adept at rapidly adapting to few-shot tasks. Experimental results affirm the viability and efficacy of the proposed method in scenarios requiring few-shot fault diagnosis. Notably, when compared with some current and advanced diagnostic methods, the proposed PHDMTL method demonstrates superior diagnostic accuracy for fault diagnosis in storage stacking machinery operating under high-speed and heavy-load conditions.