Xu Li

Surface cracks on mandrel bars affect product quality and production stability in seamless steel pipe manufacturing. Existing vision-based methods mainly rely on bounding-box detection, which is insufficient for precise crack delineation and geometric characterization. This study proposes a lightweight segmentation framework for online mandrel bar crack inspection using grayscale industrial images. Based on YOLO11n-seg, the framework incorporates single-channel input adaptation, lightweight network reconfiguration, and crack-oriented feature enhancement to improve the extraction of weak, thin, and irregular cracks while reducing computational cost. A dedicated industrial dataset and a sample-balancing strategy are introduced to alleviate severe crack–background imbalance. Based on the predicted pixel-level masks, crack area, projected length, maximum width, and average width are calculated for online evaluation. Experimental results show that the proposed method achieves a mask mAP@0.5 of 88.5%, a false negative rate of 1.72%, and real-time inference at 204 FPS with 3.01 GFLOPs. Field deployment further demonstrates the effectiveness of the proposed framework for online crack inspection and geometric parameter calculation of mandrel bars.

Due to the asymmetry of pass profiles, F-section steel is prone to defects such as overfilling, underfilling, and twisting during production, which significantly deteriorates the dimensional accuracy, mechanical properties, and surface quality of products. To mitigate the occurrence of such defects, this study established a thermo-mechanical coupled three-dimensional finite element model for the finishing rolling process of F-section steel using ABAQUS 2022 incorporating the actual operating conditions of the steel plant’s production line. By analyzing the stress–strain fields of each pass, it was found that the maximum deformation of the rolled piece is concentrated at the junctions of the inner leg with the flange, the inner leg with the web, and the outer leg with the web. Additionally, underfilling was observed at the legs and flanges of the pass in each rolling sequence. Based on these findings, an in-depth analysis was conducted on the effects of friction coefficient, tension configuration, rolling temperature, and web reduction on pass filling degree. Conditions of low friction, small reduction, and high temperature facilitate the smooth filling of metal in the leg cavity; in contrast, conditions of high friction, large reduction, and low temperature promote the filling of surface metal and an increase in spread. Maintaining a low-tension state is a common favorable condition for improving the pass filling degree of both the legs and the surface. When the friction coefficient is 0.2, tension is 0, rolling temperature is 1040 °C, and web reduction is 4 mm, the pass filling degrees of the inner and outer legs reach their maximum values of 99.88% and 99.16%, respectively. When the friction coefficient is 0.4, tension is 0, rolling temperature is 1010 °C, and web reduction is 4 mm, the pass filling degrees of the upper and lower surfaces are maximized, reaching 98.95% and 98.22%, respectively. These findings provide data support and theoretical guidance for addressing defects encountered in F-section steel production.

High‐strength low‐alloy (HSLA) strip combines high strength with excellent formability, yet its constraints make cold rolling more challenging, leading to frequent shape defects during production. To investigate and control the causes of such defects, a multistand 3D continuous rolling simulation model was developed based on industrial data, integrating the widthwise property variations of HSLA thin strip with a data genetic mechanism. The relative errors of the calculated rolling force and center thickness for each rolling pass model are controlled within ±7% and ±0.15%, respectively. Based on this model, the differentiated regulation mechanisms of work roll bending (WRB) and intermediate roll bending (IRB) on the shape evolution of HSLA strip were elucidated. WRB regulation shows clear stage‐dependent behavior: weak in early stands with relative thickness and length differences within ±2.5, nonlinear in intermediate stands driven by genetic inheritance effects, and saturating in final stands. IRB improves flatness in intermediate stands by enhancing center reduction and reducing edge‐to‐center differences but has a weaker overall effect. WRB directly affects work roll deflection and exit strip crown, while IRB may cause overcompensation in the final stand. Overall, the low deviation flatness zones for all forest stands are concentrated within the range of ≈100–250 kN for WRB and 100–300 kN for IRB.

While data-driven fault diagnosis methods have been successfully developed in the past years, large amounts of high-quality condition monitoring data are generally required to ensure model performance. Due to the high economic and labor costs in data collection, it is difficult for a single user to build an effective database, and exploring data of multiple users for better training becomes a promising solution. However, data privacy is of great importance in the real industries due to conflicts of interest, and direct data aggregation from different users is hardly feasible. To address this issue, a flexible federated learning method is proposed in this paper. Different from most existing methods with identical models under the federation, different customized individual deep neural network models can be used at different clients. Public data are exploited for knowledge transfer. Only the scores on public data are communicated between clients and server, rather than the whole model parameters. That significantly reduces the communication and computational burden. Experiments are carried out on two real-world machinery fault diagnosis datasets, and the results show the proposed method is promising for data privacy-preserving federated learning with flexible models and light communications.