Contact sensing along the soft catheter robots (SCRs) is crucial for enhancing surgical performance and intraoperative safety during minimally invasive procedures. Nevertheless, current approaches relying on medical imaging devices or optical fiber sensors are noticeably costly and pose challenges for clinical applications. This study develops a more implementable and cost-effective contact sensing method based on magnetic signals for SCR-based surgery. Specifically, several miniature Hall sensors are integrated into the SCR, and a heterogeneous magnetic field is applied. When the SCR comes into contact with the external environment, its deformation leads to changes in the magnetic signal detected by the Hall sensors. A contact sensing deep learning (ContactSenseDL) model is then developed to map the magnetic signal variation to contact position and force along the SCR. The proposed approach is validated on a catheter robot system prototype and achieves remarkable performance. In contact position prediction, the average error of axial arc length is as low as 1.97 mm (2.81% of the maximum insert length of the SCR), and the prediction of radial position exhibits high consistency with actual values. In contact force prediction, the average errors of friction and pressure are 2.43 (2.59% of the maximum friction) and 1.61 mN (2.45% of the maximum pressure), respectively. Additionally, contact sensing experiments are conducted on a knee model to demonstrate the potential application of this method. Overall, the proposed contact sensing strategy can effectively sense the contact position and force along SCRs in 3D space, holding promise for enhancing safety in SCR-based surgery.

Full-matrix capture (FMC) of matrix phased array has raised widespread attention in 3-D ultrasound imaging, while research on the imaging methods mainly concentrates on the total focusing method (TFM). However, the TFM simplifies wave beams into rays to estimate the propagation time, so its imaging accuracy is limited, and it has difficulties in adapting to the media with complex sound speed distribution. In this article, a 3-D ultrasound full-matrix imaging method is developed based on the frequency-domain reverse-time migration. Through the discretization of the 3-D wave equation in the frequency domain, the source and receiving wavefields are reconstructed simultaneously, and then, the imaging condition is implemented to recover the imaging result as the spatial reflectivity distribution. Moreover, online frequency-domain reverse-time migration is proposed to process the online FMC datasets efficiently. Herein, Green functions are initialized for the measurement region, through which the source and receiving wavefields are reconstructed efficiently with low memory consumption. Experiments show that the proposed method can image the defects in the components clearly, and the imaging accuracy and resolution of the proposed method are, respectively, 2 and 1.5 times higher than those of the TFM. In addition, the proposed method is applied to image the melt front during the injection molding process, where the merging process of the melt fronts from two sprues was first visualized in situ. Therefore, the proposed method shows great potential in the detection of defects in the components and online inspection of the manufacturing process.

The heating process contributes to over 60% of the energy consumption in material manufacturing, such as injection molding, metallurgy, die casting, etc. Temperature prediction is the crucial first step to realize energy management. However, the complex heating process poses challenges, such as strong multivariate coupling, large time delay, and frequent changes of heat transfer characteristics. To address these issues, this article introduces a temporal re-attention long short-term memory (TRA-LSTM) method for multivariate time series prediction. We adapt an encoder–decoder framework. A temporal attention module is utilized for extracting key historical features. A re-attention incorporated loss layer is proposed to rectify the temporal attention. TRA-LSTM's performance is compared with five advanced time series prediction methods across six predictions on four real-world datasets collected, including data from the injection molding barrel heating process and the battery thermal runaway process. Results show that TRA-LSTM consistently outperformed the other methods with high accuracy and efficiency. The alignment of re-attention weights with tested time delays in both heating and cooling processes further demonstrates TRA-LSTM's advantage. Ablation experiments underscore the significant contribution of the re-attention module, while hyperparameter sensitivity analysis offers guidance on selection and validates TRA-LSTM's robustness. Overall, the proposed method enables precise multivariate time series prediction and shows substantial potential for optimizing energy management in industrial production.

Abstract In the production of low‐voltage apparatus, assembly quality inspection is of great relevance for ensuring the final quality of the entire product. With the continuous improvement of production efficiency and people's requirements for production quality, traditional manual inspection methods can no longer meet the quality inspection requirements. In this paper, an Attention‐guided Feature‐fusion ConvNeXt Network (AFCN) for the automated visual inspection is proposed. By embedding the attention mechanism of the Coordinate Attention block into the residual channel of the ConvNeXt block, the position‐aware information and features of the low‐voltage apparatus images can be effectively captured to locate the quality problems. Then, an improved attention feature fusion module is adopted to merge the output features at different stages, which introduces a 3D non‐parameter attention SimAM block and adapts output accordingly. Therefore, this model can capture the key information of the feature map in a coordinated way in terms of channel and position, fully integrating multiscale features and obtaining contour texture information and semantic information of the low‐voltage apparatus. Experiments show the proposed approach can effectively classify defective and normal products.

Quality prediction plays an essential role in industrial process monitoring and optimization. However, the existed methods lack of the ability to extract features in multiphase and nonlinear time series data. To improve the quality prediction accuracy and simultaneously provide some suggestion of the process mechanism, this article proposes a phased dual-attention long short-term memory (PDA-LSTM) neural network, which consists of a phase segmentation module and a prediction module. The whole series data are first divided into phases in the phase segmentation module, and individual models are subsequently constructed for each phase in the prediction module to capture distinct patterns. The LSTM network with input and temporal attention mechanism is utilized to select the most relevant variables and time steps throughout the time sequence. The proposed method is verified with a numerical example and a real injection molding process. Compared with the existed methods, the PDA-LSTM method has the advantages of higher prediction accuracy and the ability to interpret the key input variable and time points of the process data.