In this study, based on the synergistic strategy of sustained release reduction and in-situ surface nanostructuring, oxidation-resistant and rapid sintering of submicron copper (Cu) paste in an air atmosphere was achieved. First, the critical temperature points at which Cu undergoes severe oxidation were identified. Based on this, the reductant system, composed of ascorbic acid and polyethylene glycol (PEG30 0), was specially designed to match the oxidation threshold of Cu. Notably, due to the sustained release effect of the reductant system, reduction activity was sustained throughout the entire sintering process. During sintering, both CuO and Cu2O were ultimately reduced to ultrafine Cu nanoparticles. Simultaneously, in-situ surface nanostructuring occurred on submicron Cu particles, enabling oxidation-resistant and rapid sintering in air. At a sintering temperature of 250 degrees C for 5 min, the resulting joints exhibited a shear strength of 38.1 MPa, meeting the requirements for power device packaging. This work provides a novel strategy for oxidation-resistant Cu sintering and its industrial application. (c) 2026 Published by Elsevier Ltd on behalf of The editorial office of Journal of Materials Science & ( http://creativecommons.org/licenses/by-nc-nd/4.0/ )
Fuzzy modeling is effective for handling uncertain processing durations in the Flexible Job-shop Scheduling Problem (FJSP), but schedules optimized under static fuzzy assumptions often lose robustness when uncertainties are progressively revealed during execution. To address this issue, this paper proposes a cross-attention multi-scale graph reinforcement learning (CA-MSGRL) framework for online adaptive scheduling in stochastic-fuzzy FJSP (SF-FJSP). The problem is formulated as a Dual-Trigger Semi-Markov Decision Process (DT-SMDP), where decisions are triggered by both machine-release events and uncertainty-revelation events. A tunable stochastic-fuzzy mixture mechanism is designed to capture the transition from estimated fuzzy durations to realized stochastic processing times. The proposed cross-attention multi-scale graph network (CA-MSGN) adaptively integrates global workload distributions and local operation-machine decision frontiers, while a Fuzzy-aware Graph Transformer (FAGT) encodes fuzzy uncertainty as a reliability-sensitive signal to enhance risk-aware decision making. In addition, a reliability-aware randomization training strategy is developed to emphasize structural invariants over volatile temporal estimates. Experiments demonstrate that CA-MSGRL achieves sub-second inference while improving large-scale scheduling quality and execution stability over representative benchmarks. On 200 & times; 50 instances, CA-MSGRL reduces the coefficient of variation by 42.1% compared with the CP-SAT-based static baseline, demonstrating strong robustness against stochastic perturbations.
With the rapid growth of industrial big data, modern processes are increasingly characterized by strong dynamics, nonlinear behaviors, and non-Gaussian stochastic disturbances, which pose significant challenges for reliable fault detection. In this paper, a hybrid variational Bayesian state space model is developed for dynamic process monitoring. Since standard variational Bayesian inference suffers from biased posteriors under nonlinear and non-Gaussian conditions, a Metropolis-Hastings within Gibbs strategy is incorporated to improve posterior accuracy. Meanwhile, a hybrid parameter update mechanism is employed, where conjugate terms are updated via efficient closed-form Bayesian rules, and non-conjugate parameters are refined through sampling-based optimization. Therefore, the next state can be better predicted even in the presence of noisy disturbances, hidden correlations among multiple variables can be revealed, and residual-based control limits can be reliably obtained through variational lower-bound optimization. Finally, the effectiveness and superiority of the proposed method are verified through benchmark simulation model no.1, real wastewater treatment process and a multiphase flow process.
Diabetic nephropathy is a severe microvascular consequence of diabetes mellitus, marked by an insidious progression and considerable challenges in early detection, which highlights the urgent need for sensitive and noninvasive diagnostic approaches. In response to these challenges, this study presents a novel dual-modal molecular probe, designated HY, which is engineered to selectively target superoxide anion (O2 center dot-) and combines the strengths of fluorescence imaging and magnetic resonance imaging. Comprehensive characterization confirmed the probe's selectivity, structural robustness, and biocompatibility. The probe enabled in situ visualization and quantitative tracking of O2 center dot- fluctuations in live models of diabetic nephropathy. These results establish HY as a highly sensitive and innovative platform for the real-time monitoring of superoxide anion activity. Altogether, this work provides critical mechanistic insights into diabetes-driven organ injury and paves the way for the creation of advanced diagnostic and therapeutic strategies.
