Predictive Process Monitoring (PPM) forecasts the future trajectory of ongoing process instances, including next activities and remaining execution times. While deep learning has advanced the field, modeling irregular temporal dynamics and long-range dependencies remains a significant challenge, especially within complex object-centric environments.We propose a multi-task predictive framework based on the Temporal Graph Attention Network (TGAT). By transforming object-centric event logs into temporal graphs with time-aware edges, our approach explicitly encodes continuous time intervals directly into the attention mechanism. This allows the model to capture subtle temporal irregularities often missed by sequence-based methods. A shared temporal graph encoder is jointly optimized for three critical tasks: next activity prediction, next timestamp estimation, and remaining time prediction.We evaluate our framework using four real-world BPI Challenge datasets (2014, 2016, 2017, and 2019). The results demonstrate that our model consistently outperforms or matches state-of-the-art sequence-based, graph-based, and Transformer architectures. Furthermore, attention visualizations and ablation studies confirm that integrating temporal graph modeling with multi-task learning significantly enhances predictive accuracy in complex process environments.
Dimethyl sulfide (DMS), a biogenic trace gas emitted by marine ecosystems, plays a critical role in aerosol formation and climate regulation. However, its low atmospheric abundance and rapid reactivity hinder real-time detection, limiting our understanding of its climatic impact under warming-driven environmental changes. Here, we present a laser-based optical system using tunable infrared laser direct absorption spectroscopy (TILDAS) for the real-time DMS measurement at pptv levels. This system uses a 3.37 μm interband cascade laser and a 413.4 m multipass cell, targeting a strong DMS absorption band at 2969.7 cm⁻¹ . It achieves a detection limit of 46 pptv (3σ) with 2 min averaging of the 1 Hz data, improving to 17 pptv with 30 min averaging, verified by Allan variance analysis. Compared with existing optical DMS sensors, the system offers substantially improved sensitivity while retaining the high temporal resolution needed to resolve rapid atmospheric variability. Instrumental performance and stability were confirmed through year-round calibrations, which yielded a mean campaign-specific slope of 1.01 ± 0.02 between expected and measured DMS mixing ratios. Concurrent ambient-air measurements using TILDAS and conventional gas chromatography also showed overall agreement (slope = 1.01; r = 0.94), demonstrating reliable field performance. By combining pptv-level sensitivity, continuous 1 Hz output, minimal sample processing, and long-term operational stability, DMS-TILDAS provides a powerful tool for resolving short-timescale DMS variability, with potential future applications in marine sulfur cycling and associated aerosol and climate-related processes.
Balancing strength, ductility, and electrical conductivity remains challenging in Cu alloys because conventional strengthening approaches often increase electron scattering and compromise ductility. Here, a Cu-0.1Sn alloy was subjected to dynamic offsets and shear force adjustment cryorolling (DSCR) followed by annealing to regulate recrystallization and heterostructure formation. Compared with the symmetrically rolled and annealed sample, the DSCR-annealed (DSCRA) sample exhibits an earlier recrystallization onset and faster apparent recrystallization kinetics. This behavior is reflected in its lower softening temperature (594 versus 610 K) and apparent activation energy for recrystallization (113.1 versus 126.4 kJ/mol). After annealing at 598 K, DSCRA develops a partially recrystallized heterostructure comprising recrystallized soft domains and retained hard lamellar domains, some of which contain locally preserved deformation nanotwins. DSCRA also exhibits weaker rolling-texture inheritance and greater orientation dispersion, consistent with a recrystallization pathway involving competition among multiple orientations. DSCRA achieves an ultimate tensile strength of 401 MPa and a uniform elongation of 11.1% while maintaining an electrical conductivity of 87.6% IACS. The favorable property balance is associated with plastic accommodation by the soft domains and strength retention by the hard lamellar domains. Locally retained deformation nanotwins provide additional strengthening, while interactions between the soft and hard domains promote strain partitioning and hetero-deformation-induced hardening. Recovery and partial recrystallization concurrently help reduce defect-induced electron scattering. These findings indicate that regulating the initial deformation state and subsequent recrystallization pathway offers a route to tailoring the strength-ductility-conductivity balance of dilute Cu-Sn alloys. (c) 2026 Published by Elsevier Ltd on behalf of The editorial office of Journal of Materials Science & Technology.
Plant biomass-derived polymeric materials such as cellulose nanofibrils (CNF), cellulose nanocrystals (CNC), and lignin have been widely assessed by life-cycle assessment (LCA) as low-carbon alternatives to fossil-based polymers. Yet, despite numerous LCAs of these processed biomaterials, their parent structure—lignocellulosic pulp (LC)—has rarely been quantified as a standalone material, leaving its comparative performance unclear. Here, we address this gap with three aligned LCA analyses comprising a material-level LCA of LC produced via minimal pulping and grinding, as well as two product-level case studies on LC-based thermal-insulation foams normalized to equal thermal resistance (R = 1 m²·K·W⁻¹) and UV-blocking films matched at equal optical performance (98.25
In this study, the global mean characteristics and spatial distribution differences of the Lorenz energy cycle (LEC) were investigated using four atmospheric reanalysis datasets over 44 years from 1980 to 2023. The global mean values show that the energy terms exhibit relatively small differences among the datasets, with values ranging from − 7.4 to + 3.7
