Qihong Zhou

Short-horizon forecasts of output sliver weight provide advance information for high-frequency quality monitoring in cotton drawing. The measured signals are non-stationary and quasi-periodic, so the lag structure that supports one input window may weaken in the next. We propose the Evidence-Guided Adaptive Period Network (EAP-Net), whose Evidence-Guided Adaptive Period Bank retains multiple candidate-period representations and estimates sample-specific support by combining contextual statistics with lag-specific autocorrelation evidence. A lightweight patch-based head decodes the weighted representation, while reversible instance normalization and multi-resolution fusion stabilize the forecast. We evaluate EAP-Net on 9, 629 consecutive 10-Hz records from one JWF1316 drawing frame. Given a 16-record multivariate input window that includes historical output-sliver-weight measurements, the model predicts the next two values at 0.1 and 0.2 s ahead. On the physical scale of output sliver weight, EAP-Net achieved an root mean squared error (RMSE) of 0.894 ± 0.006 mg and an MAE of 0.631 ± 0.006 mg. Relative to MFF-Mixer, the strongest neural baseline by mean RMSE, these errors decreased by 6.58% and 8.02%, respectively. EAP-Net also produced the lowest median absolute error and the highest coverage within ±1% of the measured output sliver weight. Ablation results showed that dual-evidence, sample-specific weighting improved upon global and uniform weighting. The experimental results demonstrate that EAP-Net can effectively exploit window-dependent periodic information to improve short-horizon output-sliver-weight forecasting, providing useful predictive support for high-frequency quality monitoring in cotton drawing.

To accurately detect and differentiate complex bearing degradation states (e.g. normal lubrication, dry friction, and corrosion) in the guide bar rocker arm of double needle bed warp knitting machines under high-speed, large-amplitude conditions, this paper proposes a residual network model based on cross-domain attention fusion (CDF-ResNet18). The model takes as input two-dimensional feature images obtained by converting the same vibration signal into temporal, spectral, and time–frequency domain representations. By introducing a cross-domain attention fusion mechanism, it adaptively explores and performs weighted fusion of complementary information from different feature domains, thereby improving the classification performance for the tested bearing states under the current controlled industrial scenario. To validate the effectiveness of the proposed method, vibration data was collected from an actual operating double needle bed warp knitting machine under three conditions: normal lubrication, dry friction operation, and corrosion. Once preprocessed and converted into images, the dataset served to optimize, validate, and assess the proposed model. The experimental outcomes establish that CDF-ResNet18 significantly outperforms both single-domain models and one-dimensional convolutional neural networks in terms of classification accuracy, convergence speed, and stability, validating the superiority of the multi-domain feature fusion strategy and attention mechanism in bearing condition classification. The scope of the present validation is limited to the tested machine, speed, sensor position, and three bearing states; broader cross-machine, variable-speed, and long-term degradation generalization requires further experimental verification.

Dual linear motor-driven systems (DLMDS) are widely used in industrial manufacturing due to their high dynamic stability and robust performance, typically featuring a symmetric Y1–Y2 axis structure. High-precision synchronization control of the motion platform is crucial for overall system performance. However, in practice, such systems are inevitably affected by mechanical installation errors, load disturbances, and nonlinear friction, which lead to the asymmetry of the Y1–Y2, severely degrading the synchronization accuracy between the two symmetric axes. To address these challenges, this paper proposes an EtherCAT-enabled active disturbance rejection control (ADRC) strategy for high-performance gantry synchronization systems. To cope with strong coupling effects, external disturbances, and high-speed operation, a master–slave synchronization architecture is developed based on ADRC and the EtherCAT cyclic synchronous torque (CST) mode. An extended state observer (ESO) is employed to estimate and compensate for lumped disturbances in real time, enabling precise synchronization without relying on an accurate mechanical model. Experimental results under both low-speed and high-speed operating conditions show that the proposed method significantly improves the synchronization stability and robustness compared with conventional cross-coupling control and master–slave control strategies. Specifically, the ADRC-based approach reduces synchronization errors by more than 20% under disturbance-free conditions and suppresses approximately 80% of disturbance-induced errors during high-speed operation. These results confirm the effectiveness and practical applicability of the proposed control strategy for high-precision gantry motion systems. Unlike conventional torque-mode implementations that merely replace the position loop with torque regulation, the proposed method introduces a disturbance-estimation-driven synchronization architecture co-designed with deterministic EtherCAT cyclic timing, which enables distributed real-time compensation beyond classical torque feedforward strategies.

Sliver quality in the spinning process can be enhanced significantly by using an autoleveling control system (ACS) of the drawing frame, which improves the uniformity of sliver linear density. Most existing research on closed-loop control systems focuses on theoretical analysis, and fixed-time open-loop control systems cannot accurately detect and control slivers. This paper develops and experimentally evaluates an open-loop fixed-length control system for the drawing frame to enhance the stability of the sliver thickness detection and improve the sliver quality, based on the fundamental autoleveling control equation. The proposed ACS adopts a digital signal processor (DSP) as its core, integrating key modules including a touchscreen, speed sensor, servo drive, sliver thickness sensor, and analog-to-digital converter. The autoleveling, communication, and online monitoring algorithms for sliver quality have all been implemented successfully. The proposed ACS is subsequently evaluated on an experimental platform to assess its performance. The proposed ACS reduces the CV _5m of the sliver by 93.10% compared with the unleveled sliver, improving sliver quality significantly. Compared with the existing advanced ACS, the proposed ACS reduces the root mean square error and average CV of sliver weight by 11.75% and 5.59%, respectively. This work provides theoretical and technical support for advancing domestic high-speed drawing frame development.

Aiming at the existing fabric defect detection algorithms with their suboptimal detection accuracy, high model complexity, difficult deployment to the edge of the device, and insufficient ability to meet the problem of real-time detection of fabric defects, we propose a lightweight fabric defect detection model (LWFDD-YOLO) based on the improved YOLOv8n. First, a generalized efficient layer attention network with selective kernel attention (GELAN_SKA) is proposed to replace the C2f module, and selective kernel attention is added to the module to adjust the weights of the convolution kernel according to the different features of different scales, reduce the use of computational resources, thus improving the model’s detection performance and efficiency. Second, a cascaded group attention (CGA) mechanism is added to provide a different input segmentation for each head to enhance the feature diversity of the input attention heads and improve the model’s computational efficiency. An ultra-lightweight dynamic up-sampling operator (Dy_sample) is introduced, which employs a point-sampling-based approach to reduce the consumption of computational resources and improve model performance. Finally, the YOLOv8l framework is utilized to construct a complex model teacher network and the features learned from the teacher network are transferred to the lightweight network proposed in this paper, thus further improving the model performance of the algorithm. The experimental results show that on the self-built fabric defect dataset, the accuracy, recall, and mean average precision ( mAP ) of our algorithm reach 89.4%, 85.2%, and 87.9%, which are 8.9%, 7.2%, and 4.5% higher than the original model, respectively, the number of parameters of the model decreases by 23.4%, the amount of GFLOPs decreases by 25.6%, and the file size is only 9.2 MB. The detection speed can reach 163.4 FPS on GPU. On the AliCloud Tian Chi dataset, the accuracy, recall, and mAP were also improved by 7.4%, 2%, and 4.4%, respectively. The LWFDD-YOLO algorithm proposed in this study can realize real-time detection of fabric defects with relatively obvious improvement in accuracy, and the model requires less memory and is easier to deploy to edge devices, so can be used as a reference for real-time detection of fabric defects.

Aiming at the problems of low detection accuracy and high leakage rate in traditional detection algorithms, an improved YOLOv8 algorithm is proposed for automatic detection of fabric defects. A swin transformer block was added to the C2f module in the backbone network, which can transfer information between multiple attention layers in parallel to capture fabric defect information and improve the detection accuracy of small-sized defects. To enhance the model’s performance in detecting defects of various sizes, a bidirectional feature pyramid network (BiFPN) was incorporated into the neck. This allows for the assignment of different weights to defect features in different layers. A convolution block attention module (CBAM) was added to the feature fusion layer, enabling the model to automatically increase the weight of essential features and suppress nonessential features during training to solve the problem of leakage detection of small-sized defects due to occlusion and background confusion. The Wise-IoU (WIoU) loss function replaces the conventional loss function, addressing sample imbalance and directing the model to prioritize average-quality samples. This modification contributes to an overall improvement in the model’s performance. The results of the experiment proved that on the self-constructed fabric defect dataset, the algorithm in this paper achieved an accuracy of 97.7%, recall of 95.1%, and mAP of 96.8%, which are 4.4%, 9.4%, and 5.1% higher than those of the YOLOv8 algorithm, respectively. On the AliCloud Tianchi dataset, the algorithm achieves 52.3%, 49.2%, and 49.8% in terms of accuracy, recall, and mAP, respectively, which is an improvement of 4.4% in terms of accuracy, 2.8% in terms of recall, and 2.7% in terms of mAP compared with the baseline algorithm. The improved YOLOv8 algorithm has a high detection accuracy, low leakage rate, and a detection speed of 107.5 FPS, which aligns with the real-time defect detection speed in the industry.