This study proposes a human-in-the-loop framework that integrates operator observations into a large language model (LLM) to control process parameters for defect handling in fused deposition modeling (FDM) 3D printing. Fully autonomous LLM-based control handles ambiguous sensor data poorly and cannot detect abnormal conditions that lie beyond the installed sensors. Operator observations may compensate for these limitations, but their actual impact on LLM decision-making has not been sufficiently validated. We therefore implemented the proposed framework and defined experimental scenarios involving erroneous parameter injection and environmental disturbances. The framework was evaluated in terms of LLM response quality and print quality. The LLM achieved over 80% response quality on the defined evaluation metrics and generated appropriate parameter adjustments, improving print quality by more than 55% on average. Comparative experiments further revealed that, without operator observations, the LLM sometimes failed to recognize defects. These findings demonstrate the effectiveness of human–LLM collaboration and provide a practical foundation for intelligent FDM process control.

This study investigates the ability of mechanical polishing to enhance the surface properties of sculpting clay ceramics without the use of glaze. To accomplish this, a ceramic tile machining system capable of efficient grinding and polishing was developed and utilized. Through experiments conducted under various conditions, key processing parameters— including tool rotational speed, applied pressure, and feed rate—were determined experimentally, resulting in significantly improved surface quality of the ceramic specimens. Analyses of glossiness, surface roughness, and contact angle confirmed notable enhancements in surface characteristics after polishing. These results demonstrate that a mechanically polished surface can achieve quality comparable to, or even surpassing, that obtained through traditional glazing processes, suggesting the potential to replace conventional glaze coatings with a mechanical polishing approach.

The damping ratio is a crucial modal property that influences the sound duration and acoustic quality of traditional Korean bells. This study presents a high-precision measurement methodology utilizing non-contact acoustic resonance excitation and lightweight piezoelectric patches, specifically designed to address the limitations of conventional contact-based methods. Traditional accelerometers introduce significant errors due to mass loading and cable stiffness; in contrast, the proposed method employs a lightweight piezoelectric patch combined with selective acoustic resonance to minimize parasitic damping. Specimens with varying tin and silicon content were fabricated and tested. Verification results indicate that the coefficient of variation for the measured damping ratio decreased from 24% (observed with the conventional method) to less than 10%, demonstrating improved repeatability and precision. Notably, the damping ratio reaches a minimum at approximately 13 wt% tin, providing scientific validation for the potential optimized chemical composition of the Divine Bell of King Seongdeok. Furthermore, the damping ratio was observed to decrease as the resonance frequency increased from 110 to 245 Hz. This research establishes a technical foundation for accurate acoustic simulation and the modern restoration of traditional metallic musical instruments.

Accurate prediction of cutting forces in milling is crucial for intelligent machining. However, traditional methods of collecting discrete data are often time-consuming and can lead to overfitting due to data sparsity. This study introduces a continuously variable data collection strategy designed to efficiently gather training data for machine learning models. By linearly varying the feed rate and radial depth of cut along a single tool path, we obtained continuous data that reflect the variations in machining conditions. Additionally, we applied trigonometric feature transformation to the tool rotation angle to maintain its periodic characteristics. The collected data were used to train Artificial Neural Network (ANN), Support Vector Regression (SVR), and Gaussian Process Regression (GPR) models. The results demonstrate that models trained with continuous data exhibit improved generalization performance under unseen conditions compared to those trained with discrete data. Furthermore, comparable predictive accuracy was achieved using a subset of the continuous dataset. These findings suggest that the proposed approach enhances data efficiency in model training and has the potential to reduce the experimental effort required in developing machining monitoring systems.

Here in, a high-quality automotive camera lens was developed based on an ultra-precision diamond turning core and cyclic olefin polymer (COP) injection molding process. To improve surface roughness and achieve the accuracy of plastic injection molding lens, systematic mold core machining process was developed and demonstrated using the diamond turning machine. The cutting tool path was generated by using NanoCAM 2D, and it was partly revised to prevent interference between the cutting tool and the workpiece. After the initial machining using the generated tool path, the compensation-cutting process was conducted based on the measured surface profile of an initially machined surface. After two times of compensation machining, the fabricated core mold showed a shape error of 100 nm between peak to valley (PV) and Arithmetic mean roughness (Ra) of 3.9 nm. The performance of the fabricated core was evaluated using an injection molding test. Injection molded aspheric plastic lens showed contrasts that were higher than 55% at 0.0 F, 30% at 0.3 F, and 20% at 0.7 F without any moiré phenomenon that meets the specification for automotive vision module with 1MP and 140° field of view.

Large optical mirrors for space applications require accurate surface evaluation between successive corrective machining steps. Without proper compensation for self-weight, gravitational deformation can affect the measured surface figure and thereby reduce the reliability of the machining process. In this study, an air-cylinder-based gravity compensation device was developed to maintain the target support loads for large optical mirrors. A proportional–integral–derivative (PID) controller with load-cell feedback was implemented, and the initial gains obtained using the Ziegler–Nichols method were retuned on the basis of the measured system response. Single-axis experiments showed that the retuned controller met the specified performance requirements for overshoot, settling time, and steady-state error, even when applied to air cylinders with different spring reaction forces. The validated control method was then applied, via a sequential multi-channel control scheme, to a five-axis gravity compensation system to assess its applicability to multi-point gravity compensation. The experimental results confirmed stable convergence to the target support loads and effective load regulation at steady state, demonstrating that the proposed system is suitable for the multi-point gravity compensation of large optical mirrors.

The growing complexity of steel manufacturing, loss of veteran operators, and demand for carbon-neutral operation make tacit operational knowledge increasingly difficult to sustain. Standalone large language models (LLMs) cannot adequately address terminology mismatch, schema fragmentation, undocumented tacit knowledge, and non-traceable hallucinations in on-site data. This study proposes a domain knowledge graph-based conversational operation assistant integrating a steel-process ontology with expert–LLM dialogue. An eight-class ontology is introduced, featuring the UnfilledSlot class to explicitly represent missing knowledge. A five-phase methodology constructs an initial graph, automatically fills available slots, prioritizes remaining deficiencies, and generates targeted multiple-choice questions for experts. The completed graph supports NL2SQL, analysis, explainable AI, and response generation, while operational logs reveal new knowledge gaps. Applied to a steel continuous casting process with 8,441 slots, automatic filling achieved 62% (5,233 slots), while 600 bundled expert questions resolved 2,702 additional slots, reducing the unfilled ratio from 38% to 6%. Comparisons with a standalone LLM, retrieval-augmented generation, and an ablation without UnfilledSlot, together with expert evaluations across three graph-completeness levels, confirm that explicit deficiency handling improves response reproducibility and traceability.

Omnidirectional gimbal systems use multiple actuators to achieve the desired line-of-sight (LOS) motion through thrust allocation. However, actuator degradation caused by thermal effects or aging alters the force constants, leading to model mismatch and diminished torque realization performance. Conventional pseudoinverse-based thrust allocation assumes constant actuator characteristics and, therefore, cannot effectively compensate for these variations. This paper proposes an adaptive thrust allocation algorithm based on Recursive Least Squares (RLS) for omnidirectional gimbal systems. The proposed method estimates the force constant of each actuator in real-time using measured torque and actuator current data. These estimated force constants are incorporated into the thrust allocation process to address actuator degradation. Additionally, a weighted thrust allocation strategy redistributes current demand to alleviate the burden on degraded actuators while maintaining the required torque. The proposed algorithm was validated through MATLAB/Simulink simulations under conditions of asymmetric actuator degradation. Compared to the conventional pseudoinverse-based method, the proposed algorithm reduced the torque realization RMSE by 99% and the LOS tracking RMSE by 24.5%. Furthermore, the weighted thrust allocation decreased the RMS current of the most degraded actuator by 35.8%, all while preserving torque realization and LOS tracking performance, demonstrating the effectiveness of the proposed method.

The knee adduction moment (KAM) is widely used as a surrogate measure of medial knee loading and is a primary target of gait modification. Toe-in and toe-out modifications of the foot progression angle are commonly assumed to reduce KAM peaks by displacing the center of pressure (COP), but this mechanism has not been firmly established. This study therefore aimed to clarify the mechanism underlying KAM reduction during these modifications. Healthy adults walked under normal, toe-out, and toe-in conditions. KAM, the moment arm of the ground reaction force (GRF), GRF magnitude, GRF angle, and COP were analyzed at the first and second KAM peaks. Toe-in and toe-out walking predominantly reduced the first and second KAM peaks, respectively. The moment arm was the primary determinant of both peaks (R² > 0.90). Contrary to conventional assumptions, the reduction in moment arm depended largely on changes in GRF angle, whereas the contribution of COP was minimal or even opposed the change in moment arm. These findings provide a mechanistic basis for foot progression modification strategies and should help improve their clinical effectiveness.

Cooperative 3D printing (C3DP) with multiple robotic manipulators can reduce build time through parallel deposition, but it requires layer partitioning that accounts for collision clearance, workload balance, G-code toolpath compatibility, and interlayer boundary alignment. This study presents a Voronoi- and graph-based layer partitioning framework for C3DP. STL geometry and G-code were integrated into layer-aligned data, and a 65 mm collision clearance was defined from the measured end-effector collision radius as the minimum separation preventing collisions between robots approaching non-adjacent Voronoi cells. Each layer was divided into Voronoi cells so that non-adjacent cells could be treated as collision-free regions. Cell adjacency and toolpath-based processing time were modeled as a weighted graph, and adjacent cells were clustered into workload-balanced task regions. Interlayer seed offsets staggered the partition boundaries, and graph coloring identified regions that could be printed simultaneously.The framework was evaluated by workload-balance simulations and printing experiments. Balance deteriorated when clusters were excessive relative to graph nodes. In experiments, the end-effector separation always exceeded the 65 mm clearance. Partitioned printing reduced the layer printing time from 70.063 to 63.57 min, a 9.3% reduction, and the second layer covered the preceding partition boundary, confirming the staggered-boundary implementation.

Demand for high-performance bonded magnets with complex geometries is growing in electric motors, sensors, and energy devices. In lithography-based composite manufacturing (LCM), the magnetic properties of anisotropic NdFeB composites depend strongly on particle alignment during curing. This study compares the magnetic characteristics of 70 wt% NdFeB composites fabricated under magnetic-field-assisted and non-aligned conditions. An in situ alignment module was integrated into the manufacturing platform to induce directional particle orientation. The magnetic flux distribution was analyzed in ANSYS Maxwell, and particle alignment behavior was simulated by discrete element method (DEM) modeling in ANSYS Rocky. Vibrating sample magnetometry confirmed that magnetic-field-assisted processing enhances the anisotropic magnetic response, in agreement with the simulation predictions.

Strain wave gears are widely used in applications that require high reduction ratios, compact designs, and high positioning accuracy. However, increasing torque capacity remains a key challenge, particularly due to the high stresses at the flex spline tooth root. This study presents the development and investigation of a new strain wave gear variant. The tooth profile is defined using a quadruple-arc geometry with tangential transitions. A two-dimensional tooth engagement analysis evaluates the meshing behavior under flex spline deflection and identifies potential collisions during the design phase. Consequently, an FE simulation was conducted to investigate stress distribution. In the final step, the developed strain wave gear was assessed for overall performance and load capacity. Performance evaluations, including hysteresis loss, showed that the developed strain wave gear performed equally well or better than a leading strain wave gear manufacturer across seven key performance metrics. In terms of load capacity, the maximum torque at 50% failure probability (T50%) reached 106.7 Nm, which is 30% higher than the development target of 82 Nm, thereby confirming the validity of the tooth-profile design.

This study proposes a hybrid wire arc directed energy deposition (WA-DED) process that uses 3D sand-printed supports to overcome the limitations of low-angle overhang fabrication. WA-DED is a metal additive manufacturing process offering high deposition rates, cost efficiency, and suitability for large-scale components. However, because of its high heat input and molten pool instability, low-angle overhang and hollow structures remain difficult to fabricate, as the molten metal tends to collapse under gravity and thereby degrade geometric accuracy and surface quality. To address this issue, sand-printed supports were introduced. The supports provide tailored mechanical constraint and guide the solidification of the molten pool during deposition. Experiments were conducted to evaluate the feasibility of the proposed process at various overhang angles. The results show that the hybrid process markedly improves deposition stability and enables the fabrication of low-angle and curved overhang structures that conventional WA-DED cannot produce. These findings confirm the effectiveness of sand-supported WA-DED and highlight its potential for industrial applications requiring complex geometries, such as aerospace, marine, and energy components.

Environmental issues have become a global concern recently. Countries worldwide are making efforts for carbon neutrality. In the automotive industry, focus has shifted from internal combustion engine vehicle to eco-friendly vehicles such as Electric Vehicles (EVs), Hybrid Electric Vehicles (HEVs), and Fuel Cell Electric Vehicles (FCEVs). For driving strategy, research on vehicle driving method that can reduce vehicle energy consumption, called eco-driving, has been actively conducted recently. Conventional cruise mode driving control is not considered an optimal driving strategy for various driving environments. To maximize energy efficiency, this paper conducted research on eco-driving strategy for EVs-based on reinforcement learning. A longitudinal dynamics-based electric vehicle simulator was constructed using MATLAB Simulink with a road slope. Reinforcement learning algorithms, specifically Deep Deterministic Policy Gradient (DDPG) and Deep QNetwork (DQN), were applied to minimize energy consumption of EVs with a road slope. The simulator was trained to maximize rewards and derive an optimal speed profile. In this study, we compared learning results of DDPG and DQN algorithms and confirmed tendencies by parameters in each algorithm. The simulation showed that energy efficiency of EVs was improved compared to that of cruise mode driving.

Hair-like surfaces in nature consist of high-aspect-ratio fibers with diameters below 100μm, falling to several tens of micrometers in softer hairs. These fine fiber arrays govern tactile softness, flexibility, surface texture, and mechanical response. Conventional fiber-spinning methods produce fine fibers effectively but offer limited control over the position, direction, and patterned arrangement of individual fibers. Here we propose a fused deposition modeling (FDM)-based strategy that combines melt extrusion with geometric drawing to fabricate PLA hair-like fibers. PLA melted fully at the processing temperature of 250oC, well below the thermal degradation onset near 330oC. DSC analysis showed that faster cooling suppressed thermodynamic crystallization, indicating that the final fiber structure is governed by drawing history and rapid cooling rather than by increased crystallinity. As the printing speed increased, the fiber diameter decreased nonlinearly, following D ≈ 106.3 v^-0.45, in excellent agreement with the D  v^-0.5 scaling predicted by the continuity equation. Tensile strength and modulus increased with printing speed, whereas elongation and toughness decreased, indicating drawing-induced molecular orientation. These results demonstrate that FDM can serve as a programmable platform for fabricating biomimetic hair-like fiber arrays with predictable diameter and mechanical properties.