To address issues such as significant wall thickness deviation, diameter expansion, and cracking during the spinning of D406A ultra-high strength steel cylindrical parts due to improper process parameter combinations, a finite element model for power spinning was established using ABAQUS/Explicit software. The orthogonal experiment range analysis method was employed to investigate the influence of process parameters, including spinning roller feed per revolution (1-3 mm/r), thinning ratio (30-50 %), and spinning roller nose radius (4-8 mm), on wall thickness deviation. The results indicate that the order of influence of each parameter on wall thickness deviation is as follows: thinning ratio > spinning roller feed per revolution > spinning roller nose radius. Based on simulation results, spinning experiments using the optimal parameter combination: spinning roller feed per revolution of 2 mm/r, thinning ratio of 40 %, and spinning roller nose radius of 4 mm, successfully produced a cylindrical part with a wall thickness deviation of 0.053 mm, further verifying the accuracy of the finite element simulation analysis. (Received in October 2025, accepted in December 2025. This paper was with the authors 1 month for 1 revision.)
This study develops a numerical model using ABAQUS to investigate the stick-slip vibration mechanism of a near-horizontal drilling bit in underground mines, incorporating bit-rock cutting, drill string-hole wall interaction, and gravity. Key factors include feed speed, rotational speed, drill string length, and rock properties. Results show that stick-slip vibration consists of slow loading, sticking, slip, and high-speed rotation sections and is coupled with axial vibration. Higher feed speed increases torsional energy storage time, thereby prolonging the sticking section. Higher rotational speed raises the peak velocity of the bit during stick-slip, which can exceed twice the driving speed. Drill string flexibility is a root cause; shorter strings with higher stiffness reduce sticking tendency. Average drilling speed decreases with increasing string length, though the reduction rate diminishes. Increased rock hardness prolongs the sticking phase, but the increment gradually levels off. These findings provide a theoretical reference for safe and efficient near-horizontal drilling in underground mines. (Received in February 2026, accepted in April 2026. This paper was with the authors 2 weeks for 2 revisions.)
Equipment manufacturing is characterized by multi-variety, small-batch production and frequent dynamic disturbances. Conventional production simulation is mainly used for offline validation and is rarely integrated with production planning, which limits its applicability in dynamic environments. This study develops a simulation-driven closed-loop rolling optimization framework based on a multi-agent discrete event simulation model. The framework integrates rolling horizon control, real-time interaction between simulation and optimization, and disturbance-triggered rescheduling. A coupling mechanism between the simulation clock and optimization process is established to support continuous plan adjustment. Simulation experiments in a real manufacturing workshop show that the proposed approach reduces order tardiness and makespan while maintaining stable equipment utilization under dynamic disturbances, compared with offline simulation-based optimization and conventional scheduling methods. The results indicate that production simulation can be integrated into planning decisions and used for dynamic adjustment in complex manufacturing environments. (Received in January 2026, accepted in April 2026. This paper was with the authors 1 month for 2 revisions.)
In this article, we present a methodology that enables the automatic creation of metamodels based on artificial neural networks. We use a neural network architecture search technique, known as NAS. Work in this area so far has mostly relied on manual design of the network architecture, which is timeconsuming and the results depend on the researcher's experience. Our framework combines several interconnected steps: generating data using Latin Hypercube Sampling, automatically checking its quality, finding the best architecture using the Tree-structured Parzen Estimator (TPE) algorithm, and finally building a set of models that work together to make the final prediction. We have verified the entire procedure on a real production line controlled by the Constant Work-In-Progress (CONWIP) system. We achieved a prediction accuracy of R2 = 0.9955 and an average percentage error of 1.34 %. Compared to direct simulation, our metamodel is 312 times faster, enabling optimization and scenario analysis directly during decision-making.
