Driven by the Industry 5.0 paradigm, tool wear monitoring has become essential for ensuring machining quality and production efficiency. However, the supervision asymmetry between readily available sensor signals and scarce labeled data hampers the industrial deployment of monitoring models. To address this issue, a physics-guided semi-supervised learning method with uncertainty quantification is proposed. First, multi-source heterogeneous features that characterize local cutting dynamics and historical degradation information are constructed via adaptive feature filtering. Second, the improved parallel symbolic regression learns trend-constrained symbolic degradation mappings and generates physics-guided weak labels that are consistent with the irreversible evolution of tool wear, thereby alleviating the supervision bottleneck caused by insufficient true labels. Subsequently, a staged conformal prediction strategy quantifies stage-related uncertainty of weak labels and constructs adaptive loss weights to reduce the interference of low-reliability weak labels during model training. Finally, a physics-guided cross-attention fusion module is designed to leverage degradation trend information as guidance for adaptive interactions among heterogeneous features, while a raw-feature bypass is introduced to preserve data-driven information and reduce the risk of negative transfer caused by local weak-label bias. Ablation studies confirm the advantages of the proposed method in preserving trend smoothness, suppressing heteroscedastic noise, and improving late-stage identification accuracy. Compared with the best-performing baseline, the proposed method reduces RMSE and MAE by 14.2 % and 10.9 %, respectively, while R2 increases by 1.2 %. Further validation shows that the proposed method can robustly adapt to different cutting conditions. Meanwhile, its lightweight online inference architecture enables deployment on resource-constrained industrial edge devices.

The ultrasonic phased array assembles multiple transducer elements into one system, greatly increasing the detection coverage and allowing for high-resolution imaging of reflectors. In recent years, full matrix imaging has gained its prominence in the development of ultrasound detection technology because of its comprehensive information, flexible imaging and efficient detection. In this review, the scope is focused on ultrasonic imaging methods targeting industrial products and manufacturing processes. The effectiveness of full-matrix imaging strongly depends on the imaging method, which is grounded in the wave propagation behavior of the measurement structure. The theoretical frame of the imaging methods is composed of wavefield reconstruction and imaging condition. The former reconstructs the wavefield in the measurement region through full wave equations, Green function, or one-way equation, and this step determines the imaging accuracy and efficiency, while the latter processes the wavefields to yield the imaging results. The existing wavefield reconstruction methods are compared through two simulation models, wherein the effects of spatial and temporal factors are investigated. Moreover, the existing full-matrix imaging methods are grouped into four categories according to the imaging theory, namely total focusing method, phase shift migration, time-reversal method and reverse time migration, wherein the characteristics and development directions are summarized. In addition, the relationship between full-matrix imaging methods and those for other ultrasonic acquisition modalities is also analyzed. Furthermore, the industrial application and technical gaps of full-matrix imaging methods are discussed and analysed. This review is expected to provide strong technical support in full-matrix imaging methods.

This study presents a comprehensive analysis of the coupled influence of acceleration and speed on displacement errors in ball screw feed systems. A micromechanical model based on ball elastic deformation is developed to investigate the respective contributions of centrifugal forces (from speed) and inertial forces (from acceleration) to displacement errors. Numerical simulations reveal that the displacement error increases approximately linearly with acceleration. Furthermore, the concept of an error jump value is introduced to illustrate the significant influence of preload on system stability during startup. Sensitivity analysis confirms that appropriately increasing the preload can reduce the error jump value, thereby improving dynamic stability. Based on these findings, a novel preload optimization strategy is proposed to achieve a balance between precision and stability. Compared with conventional empirical methods, the proposed strategy significantly reduces additional friction losses while maintaining high precision and operational stability.

High-end Computer Numerical Control (CNC) machine tools play an irreplaceable role in industrial production. CNC machine tools require high-speed spindle systems during the machining process, and the heat generated by the rotation of the motor and rolling bearings in the electric spindle, will cause significant thermal errors in the machine tool. To solve the temperature rise of the electric spindle, it is necessary to equip the motor and bearings with high-performance cooling water jackets. Nevertheless, traditional cooling water jackets have disadvantages in cooling performance and flow resistance. This paper presents a method of robust topology optimization (RTO) of the flow channel in cooling water jacket considering variable loading conditions. Establish a multi-physics field problem of convective heat transfer under flow-solid coupling issues. Use an interpolation method to map the design variables into the flow channel height information, which allows the three-dimensional design at the computational cost of two-dimensional topology optimization (TO), improving computational efficiency. Select a square heat exchanger as the two-dimensional numerical case, consider the variable loading conditions to optimize, and compare the differences in designs under constant and variable loading conditions. The results show that the RTO considering variable loading conditions can reduce the pressure drop of the flow channel while ensuring similar heat transfer performance. Finally, establish three-dimensional models of the traditional spiral jacket and RTO jacket to compare heat exchange performance, verifying the rationality and superiority of the RTO flow channel considering variable loading conditions.

Indexable inserts feature a wide variety of types and undergo frequent updates, posing enormous challenges to automated visual inspection. Traditional deep learning methods require extensive data collection and retraining for each new type of insert, resulting in high training costs and overly stringent requirements for quality control personnel, which hinders their practical application. To address these issues, we propose a novel zero-shot anomaly detection method in this paper, specifically, a multi-scale multi-modal segmentation contrastive language-image pre-training (CLIP) model (termed M2S-CLIP). First, we designed a multi-scale segment anything model (SAM)-CLIP distillation learning adapter (MS-SAM adapter) that combines the semantic understanding capabilities of the CLIP with the fine-grained segmentation knowledge of the SAM, thereby enhancing the model's ability to detect fine details. Thereafter, we introduce a learnable textual prompt template based on prompt learning, which enhances the multi-modal large model's understanding of industrial scenarios. Subsequently, a multi-scale cross-modal fusion module (M2P-Fuse) is designed to extract visual features at multiple scales while dynamically guiding textual features with visual cues. Finally, a bottleneck-structured aligner is developed to achieve precise image-text alignment. Experimental results demonstrate that M2S-CLIP achieves an image-level AUROC of 95.2% and a pixel-level AUROC of 93.0% on our self-built dataset, significantly outperforming existing methods, while cross-domain tests on MVTec AD and VisA verify its strong generalization capability.

The remote center of motion (RCM) constraint plays a critical role in robot-assisted minimally invasive surgery, ensuring that surgical instruments move around a fixed point and thereby reducing tissue damage near the surgical incision. RCM constraint methods can be categorized into mechanical and algorithmic approaches. Mechanical methods provide high accuracy but lack flexibility, while algorithmic methods offer greater adaptability; however, existing algorithmic approaches typically maintain RCM constraints in real time at the velocity level, which can fail under task conflicts or in complex environments. To address this issue, this paper proposes a novel robot path planning method based on two-stage sampling, which enforces RCM constraints through geometric construction rather than real-time correction, explicitly considering the global feasibility of the path. First, the robot performs a collision-free first-stage sampling in the task space to generate an initial path of the working points. Then, based on geometric relationships, all initial path points are mapped through the RCM point to compute the corresponding robot joint-end poses. Finally, a second-stage sampling is performed on the obtained joint-end poses to generate the final path in the robot's joint or Cartesian space. To address sudden pose changes and increased RCM errors caused by initial path points being close to the RCM point, a distance-checking mechanism is introduced, and helical interpolation is applied to poses exceeding the threshold, ensuring path continuity and satisfaction of the RCM constraint. Experimental results demonstrate that the proposed method produces feasible paths that meet the precision requirements of medical robots under RCM constraints.

Reliable, real‐time monitoring of elastomer crosslinking is essential for ensuring service performance and manufacturing quality of elastomeric materials used in soft robots and adaptive systems. However, existing characterization techniques, such as nuclear magnetic resonance (NMR), mechanical testing, and swelling, are destructive, offline, costly, or lack temporal resolution, making them unsuitable for in situ monitoring. In this study, a magnetic levitation (Maglev)‐based platform is proposed for real‐time, nondestructive, and quantitative monitoring of elastomer crosslinking evolution under ambient conditions. A mathematical model is established to correlate levitation height with material density and crosslinking degree, enabling continuous monitoring throughout the curing process. Crosslinking degree is quantitatively determined using a pre‐calibrated density–crosslinking correlation, with the underlying mechanisms of network formation further analyzed. Taking liquid silicone rubber as a representative system, it demonstrates high sensitivity (241.61 mm cm 3 g −1 for density and 0.17 mm % −1 for crosslinking degree) and long‐term monitoring over 20 h, capturing a full curing profile from 15.25% to 31.49% crosslinking degree. This study establishes a high‐sensitivity and cost‐effective physical sensing platform for polymer curing analysis and introduces a new paradigm for in situ, process‐aware crosslinking quantification, offering strong potential for integration into real‐time quality control and adaptive materials development.

Mechanistic models for milling carbon-fiber-reinforced polymer (CFRP) are often limited by static cutting coefficients calibrated under specific machining conditions. As a result, their predictive accuracy often deteriorates when process parameters or tool geometries change. To address this problem, this study develops a physics-based cutting force prediction framework for CFRP milling. The framework uses the instantaneous fiber cutting angle and instantaneous uncut chip volume as geometry-resolved inputs, and extends the baseline prediction across cutting conditions through size effect modeling. The model is validated using both a straight-flute tool and a fine-pitch nicked router. The predicted force waveforms agree well with the experimental results in phase, magnitude, and waveform characteristics. For the nicked-router case, the model also captures the runout-induced force asymmetry under low-chip-load conditions, with peak-force deviations controlled within 9.35%. The model also provides high-resolution transient force descriptors for characterizing rapid local force variations associated with the staggered engagement of nicked cutting edges.

Quadrics are the most common types of surfaces used in weldments. Extracting multiple welds formed by quadric surfaces from a single 3D point cloud is an essential and challenging step in robotic welding for complex weldments. Relevant studies mostly focus on weld extraction from weldments with a single type of quadric. A weld extraction method for weldments with general quadratic surfaces is required. (1) This paper proposes a quadric fitting method for all kinds of quadrics, efficiently solving the quadric models with linear equations. It can be observed from the test results that the fitting error of the method proposed in this paper grows at a rate of about 1/5 of that of the SVD method in the literature as the point cloud noise grows; and the method proposed in this paper improves the operating efficiency by about 40 %. (2) For noisy point clouds with multiple intersecting quadrics, a quadric segmentation method based on region growing is proposed. The proposed segmentation method reduces 30 % similar to 50 % of segmentation errors during the tests compared to the ICP registration approach in the literature. (3) A region growing method based on ETVPS (End Tangent Vector Projection Sorting) for weld extraction with the unorganized raw intersection points from the segmented quadrics is proposed. All the mentioned methods are verified with solid experiments with physical weldments. The proposed weld extraction method proves to be robust to noisy and defective point clouds.

Flexible film sensors are pivotal components for human-machine interaction, and their performance critically depends on the fabrication of key functional materials. Ultra-soft silicone films are widely used in multilayer flexible sensors as dielectric, encapsulation, and mechanical buffer layers. However, their extremely low viscosity and flow-dominated rheology hinder integration with conductive inks within a unified manufacturing process. Inspired by traditional Chinese art-sugar-painting-we propose a wetting-driven thermal-freezing printing (WDTFP) strategy that regulates spreading dynamics and crosslinking behavior to enable controlled fabrication of ultra-soft silicone films. The resulting films exhibit tunable thicknesses from 80 to 1000 mu m and nanoscale smoothness (Ra approximate to 5 nm). More importantly, this approach overcomes intrinsic limitations of conventional thin-film manufacturing in forming conformal films on curved surfaces, perforated films, and graded-thickness films, while enabling co-printing of dielectric and conductive layers, thereby establishing WDTFP as a scalable platform for the integrated fabrication of structurally complex ultra-soft silicone films.

CNC machine tools are a fundamental pillar of modern manufacturing, essential for ensuring machining accuracy and optimizing production efficiency, which are crucial for maintaining high processing quality and consistent product standards. Despite their widespread use, traditional CNC machines face persistent challenges in handling complex machining processes and dynamic operational environments, especially with limited precision control and path planning optimization. In recent years, advancements in artificial intelligence (AI) have led to significant breakthroughs in CNC machining, offering innovative solutions to address these challenges. Notable improvements include real-time machining process monitoring, error compensation techniques, intelligent path planning, and digital twin modeling, all of which contribute to enhanced performance and capabilities. In this paper, an in-depth review of the key AI-driven technologies and their applications in CNC machining over the past five years is presented to examine their impact on process optimization and machine tool performance. Furthermore, the review systematically outlines the critical challenges facing AI integration, including system architecture, data management, performance limitations, and application scalability. Finally, emerging trends in next-generation CNC machine tools are discussed with a focus on self-optimization, autonomous decision-making, and adaptive control, offering a forward-looking perspective on the evolution of intelligent machining.

With rapid development of smartphones, performance requirements for injection molded ultraprecision optical lenses have become increasingly stringent. However, current research remains inadequate, particularly in elucidating the correlation between process parameters and lenses geometrical deviation. This study investigates the molding process of an ultraprecision aspheric optical lens, which feature a thickness of only 0.815 mm. Effects of key process parameters (injection velocity, packing pressure, back pressure and melt temperature) on geometrical deviation PV of the lenses were systematically studied. Notably, due to relatively small volume of lenses, packing pressure cannot be effectively transmitted to the cavity through the runner system, thus exhibiting limited influence compared to other parameters, which differs significantly from that in conventional molding. The geometrical deviation is unevenly distributed across the lenses and is strongly correlated with the part mass. A 2.6% increase in mass was found to reduce convex surface deviation by 74.8%, but simultaneously caused a 211.1% increase in concave surface deviation. These findings validate the feasibility of controlling geometrical deviation through part mass adjustment, providing important guidance for optimizing process parameters in manufacturing and improving product quality of ultraprecision optical lenses injection molding.

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. Note to Practitioners-This research is motivated by the growing demand for more feasible and cost-effective contact sensing methods in SCR-based minimally invasive surgery to ensure surgical efficacy and safety. In this work, a magnetic signal-based contact sensing method is developed. The proposed method dispersedly integrates several Hall sensors into the SCR and utilizes a deep learning model to map the measured magnetic signal to contact information. After the training of the deep learning model, the presented contact sensing method only requires Hall sensors and electromagnets in hardware to achieve effective contact position and force estimation along the SCRs in 3D space. Therefore, this method features low operational and maintenance costs, offering potential for widespread adoption in resource-constrained healthcare settings.

Precise control and real-time monitoring of key parameters such as concentration and density in paramagnetic solutions are essential for ensuring accuracy in detection science, biomedicine, and precision manufacturing. However, these parameters inevitably fluctuate over time due to water evaporation, repeated use, or prolonged storage, potentially compromising measurement reliability. This study proposes a sensitive, accurate, universal, scalable, and cost-effective magnetic levitation method for real-time, quantitative monitoring of concentration, density, magnetic susceptibility, and water evaporation in paramagnetic solutions. By simply tracking the levitation height of a standard-density sphere acting as a probe, this technique enables rapid, sensitive, and high-precision parameter determination. The method's performance was comprehensively assessed through sensitivity and accuracy evaluation, real-time detection, long-term dynamic monitoring, application demonstrations, and four validation frameworks, covering effectiveness, applicability, universality, scalability, and reliability. As a representative case, the MnCl2 solution concentration increased from 1.0001 M to 1.0481 M over 870 min. The method achieved concentration accuracy of 0.0045 M (corresponding to 0.00047 g cm-3 in density) and sensitivity up to 45.00 mm M-1 (471.70 mm cm3 g-1). This method offers an efficient, accurate, reliable, and scalable approach for real-time detection, long-term monitoring, and simultaneous sample measurement and solution calibration in Maglev systems. Its broad adaptability and practical operational advantages make it a valuable tool for precision manufacturing, biomedical analysis, and analytical detection applications, while addressing the challenge of coupling continuous solution monitoring with non-destructive sample testing in Maglev-based systems.

Carbon-fiber-reinforced polymer (CFRP) milling exhibits highly anisotropic fracture behavior and spatially non-uniform cutter engagement, making local cutting-state characterization particularly challenging, especially for complex tools such as fine-pitch nicked routers. To address this issue, this study develops a mechanism-informed geometric framework that characterizes the local cutting state through two complementary descriptors: the instantaneous fiber cutting angle, which describes how the material is removed, and the instantaneous uncut chip volume, which describes how much material is removed. A unified laminate-tool representation is established to enable high-fidelity calculation of these descriptors for arbitrary effective cutting-edges, axial positions, and time instants. Experimental observations, validation results, and case analyses show that the proposed descriptors are consistent with measured force responses, observed damage features, and the spatiotemporal heterogeneity of nicked routers. This work therefore establishes a mechanism-informed geometric representation of local cutting state in CFRP milling.

Tool wear monitoring is crucial for optimizing CNC machining processes in next-generation intelligent manufacturing systems. However, existing methods struggle to capture the dynamic relationship between highfrequency features and wear evolution. Small-sample training and the uneven distribution of labels across the domain exacerbate bias in feature migration, limiting model generalizability and adaptability. To address this, a frequency domain-aware and bionic-aligned collaborative modeling approach for domain shift mitigation is proposed. Firstly, a smoothed wavelet convolution feature extraction method is introduced, enhancing the capture of sensitive frequency bands and stabilizing gradient propagation through a Softplus smoothing mechanism. The method's ability to suppress domain offset during the initial feature extraction stage is validated by comparing feature activation distributions across two domains. Inspired by bat echolocation, an attention mechanism is proposed that integrates energy guidance, echo alignment, and time-frequency focusing modules to enhance high-frequency signal mapping and mitigate domain shift. The method's effectiveness in highfrequency feature response is validated through enhancement metrics and variance distribution within the attention focus region. Additionally, interpretability of dual-domain feature alignment is improved by calculating working condition similarity, integrating a priori knowledge, and optimizing the MMD loss function. Systematic ablation experiments demonstrate that the proposed method achieves average RMSE, MAE, and R2 values of 0.078, 0.063, and 0.817, respectively. It outperforms all ablation models, yielding average reductions of 31.6 % and 32.5 % in RMSE and MAE, and an average improvement of 42.7 % in R2. Furthermore, the proposed method outperforms the best-performing method among the four mainstream methods, reducing RMSE and MAE by 13.3 % and 2.5 %, and improving R2 by 5.1 %. This method effectively suppresses domain bias in feature extraction, mapping, and training under small sample conditions, providing critical technical support for intelligent manufacturing in complex, variable working environments.

With the increasing demands of high-end manufacturing industries for processing quality and efficiency, real-time and accurate monitoring of tool wear during milling has become a critical issue in the field of mechanical processing. To address this, this paper proposes an online monitoring technology for tool wear based on the Integrated Multi-Scale Temporal Convolutional-Attention (IMSTCA) model. By synchronizing data from multi-source sensors, such as cutting force, vibration, and acoustic emission, the model integrates the advantages of multi-scale feature fusion and deep temporal feature extraction to monitor and warn of tool wear in real time. IMSTCA consists of two key modules: the Multi-Scale Selective Kernel Convolutional Network (MS-SKCNN) and the Residual Temporal Convolutional Network with Integrated Attention Module (RTCN-IAM). The former processes high-dimensional multi-channel signals through multi-scale convolutions, using a selective kernel strategy to adaptively choose the optimal branch for various wear characteristics. The latter, using deep stacked temporal convolution and an integrated attention mechanism, automatically focuses on key patterns from both the channel and time dimensions, capturing long-term and short-term dependencies in the wear evolution. Through ablation experiments on the PHM dataset, the IMSTCA model demonstrates good monitoring accuracy and generalization ability, outperforming other models in terms of RMSE, MAE, MAPE and R2.

The machining of Carbon Fiber Reinforced Polymers (CFRP) presents substantial challenges in preserving surface integrity due to the material's pronounced anisotropy, high stiffness, and low thermal conductivity. These characteristics lead to significant tool wear, considerable thermal damage, and pronounced surface defects during the machining process. To assess the machining performance of Polycrystalline Diamond (PCD) tools under different wear conditions and fiber orientations, this article introduces an experimental study on the milling of T700 / epoxy CFRP components using pre-wear PCD tools. The CFRP workpieces were specially designed in a gear shape to minimize tool wear progression and facilitate subsequent measurements. The results, observed using a Scanning Electron Microscope (SEM), indicated variations in surface integrity related to tool wear progression. At zero Fiber Cutting Angles (FCAs), worn tool causes severer matrix peel up from the fiber. At acute FCAs, slightly worn tools produced a better surface finish than fresh tools due to matrix smearing, whereas severely worn tools resulted in chip adhesion and degraded surface quality. Down milling achieved greater surface integrity than up milling. Additionally, surface integrity was found more critical at low feed rates with severely worn tool. At obtuse FCAs, surface integrity was highly dependent on the feeding orientation, and was sensitive to tool wear. Up milling produced saw-tooth surface cavities at 150°FCA, while down milling showed significant cracks at 120° and 150°FCAs. The cutting force was more sensitive to tool wear progression at 30°, 60° and 90°FCAs, and less sensitive at 120° and 150°FCAs. Additionally, cutting forces were higher in up milling than in down milling, especially as tool wear progressed. These findings provide valuable insights for further research on optimizing CFRP machining processes to reduce tool wear and enhance component quality.

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.