埃克森美孚

Real-time process monitoring requires methods that extract actionable information from high-dimensional multivariate time-series data. In this work, we present a new trajectory-based process monitoring framework that combines tools of topological data analysis (TDA) and neural ordinary differential equations (NODEs). In this approach, we partition the time-series data into overlapping windows and represent these windows as manifolds. We then use topological descriptors such as Euler characteristic curves to provide compact summaries of the structure (e.g., connectivity, holes, etc.) of such windowed data across filtration thresholds, and a NODE learns the temporal evolution of these descriptors. The norm of their predicted temporal derivatives is used as an event signature. The framework is evaluated using historian data from an industrial olefins plant. We compare the proposed TDA-NODE framework against reconstruction-based principal component analysis and autoencoders and against a trajectory-based Koopman autoencoder. The state-of-the-art approaches primarily capture the major events such as shutdowns and start-ups but exhibit limited sensitivity to the subtle minor events like coldbox temperature deviations, production rate cuts, and changes in ramp rates. In contrast, TDA-NODE produces distinct transient signatures aligned with both the major and minor events, even when their static topological representations overlap with normal operation. The pretrained NODE model also accurately identifies instability regions in an independent testing dataset from the same plant without retraining or recalibration.

As AI systems increasingly exhibit agentic behavior, discussions of autonomy often conflate what systems are technically capable of doing with what they should be permitted to do in practice. This paper introduces a governance framework that explicitly separates Allowed Autonomy Levels (AAL), which define the degree of autonomy an AI agent is authorized to exercise given risk, oversight, and accountability considerations, from Autonomous Capability Levels (ACL), which characterize an agent's inherent technical abilities. We present a structured set of autonomy levels spanning reactive execution, decision support, supervised action, goal-directed autonomy, and delegated operational authority, and describe how control, reversibility, and accountability change as autonomy increases. To operationalize this framework, we propose a risk-aware decision process for assigning allowed autonomy, analyze how risk and accountability evolve across autonomy levels, and demonstrate its application through a deployed enterprise data engineering agent, illustrating how a system assessed at a high capability level can be deliberately constrained to a lower allowed autonomy based on risk, reversibility, and organizational readiness. By distinguishing authorization from capability, this work provides practical guidance for the design, deployment, and governance of Agentic AI systems.

As packaging consumption continues to rise in China, improving End-of-Life (EoL) management is critical for resource recovery and environmental protection. This study employs material flow analysis to track and compare the EoL disposition of polyethylene (PE), paper, metal, and glass packaging across key applications. Results reveal that PE packaging exhibits the lowest recycling rate of 35% compared with paper (69%), metals (67%), and glass (66%). Packaging with simple compositions and mature recycling systems—such as aluminum cans (95%), beverage glass bottles (92.9%), corrugated boxes (80%), industrial-grade PE film (e.g., stretch film and heavy-duty sacks, >80%), and food-grade high-density polyethylene (HDPE) bottles (70%) — achieve higher recycling rates. In contrast, complex or contaminated types, including chemical barrels (15%), flexible household PE packaging (<10%), courier bags (16.4%), and kraft shopping bags (15%) are largely incinerated. These results underscore the need for material-specific data, coordinated actions across the packaging value chain, and targeted, data-driven policies to enhance China’s packaging waste management.

Ammonia is increasingly being considered as an alternative to fuel oils for powering vessels and as a carrier for hydrogen. A deeper understanding of the environmental and human health risks of unplanned ammonia releases will further inform this promising contribution to efforts to decarbonize shipping. Integral plume models, including Phast, have been previously applied to evaluate ammonia dispersion in air. However, most studies modeled terrestrial releases. In this study, Phast was used to model volatilization, transport, and atmospheric exposure of ammonia hypothetically released from a bunkering hose over water at a representative location along the Gulf Coast of the United States. The results indicated that fate and exposure estimated by Phast are dependent on how the release is configured in the model, with a resulting change in affected area of 0.1 km2 on average between model modes (∼16% difference of total affect area). Overall, Phast estimated that 71–90% of the ammonia mass would dissolve into the water rather than volatilize into the air. Phast results indicated there would be >1% probability of 0.016–0.083 km2 exceeding human health hazard thresholds from a hypothetical 5-min release over water, representative of typical unplanned releases. For a larger, catastrophic release, Phast estimated 0.6 to 7.8 km2 would be affected. There are insufficient water-side field studies with which to validate Phast model results, particularly the estimations of air-sea partitioning. Such atmospheric modeling is needed for human health risk assessment and to provide input to in-water dispersion models that estimate fate and effects of spilled ammonia.

Abstract During the Liza Phase 1 development of the Guyana deepwater field, conventional Reservoir Drill-In Fluids (RDIF) were employed. However, as well complexity increased, there arose a necessity for a more robust and versatile fluid capable of not only meeting or surpassing these challenges but also accommodating the diverse well types and completion methods slated for subsequent developments. This paper meshes global author experiences with Guyana-specific project experiences and includes some of the findings during laboratory testing conducted to validate the new RDIF used in Guyana and emphasizes the performance enhancements observed in the field. Notable improvements include a significant reduction in plastic viscosity and overall rheological profile, enabling efficient drilling while maintaining Equivalent Circulating Density (ECD) within the narrow Pore Pressure-Fracture Gradient (PPFG) window. Furthermore, the micronized ilmenite minimizes the risk of weighting agent sag issues and allows significant reduction in critical path RDIF filtering time prior to running completion screens compared to API barite systems. A breaker system was used on selected wells to partially dissolve filtercake to increase injectivity on gas-first injectors. Three primary RDIFs were trialed in Guyana during Liza Phase 1 and Phase 2: (1) Non-Aqueous Fluid (NAF) weighted with API Barite, (2) NAF weighted with Ultra Fine Grind (UFG) Barite, and (3) NAF weighted with 100% Calcium Carbonate (CaCO3). Each fluid exhibited varying degrees of success depending on the completion type, but limitations emerged impacting well productivity or injectivity. Key challenges included high ECD nearing the reservoir's Fracture Gradient, posing a heightened risk of lost circulation. Such risk could be mitigated with shorter reservoir intervals, controlled Rate of Penetration (ROP), and reduced tripping speeds. Other complexities called for increased vigilance and dilution to manage fluid properties, increased storage needs, and increased logistics for three distinct RDIF systems. In essence, conventional drill-in fluids had reached their technical limitation. For the second (Liza Phase 2) and third (Payara) developments, the Alliance Partners (Alliance) collaborated to design and implement a new high-performance, non-aqueous RDIF. This fluid, weighted with a specific Particle Size Distribution (PSD) of micronized ilmenite and sized CaCO3, aimed to deliver benefits equivalent to a 100% CaCO3 system with enhanced ECD control and adaptability to all well completion types. Increased well productivity and injectivity were observed with this micronized ilmenite fluid compared to previous systems. This paper presents results from actual wells using RDIF containing micronized ilmenite as the primary weighting agent showing increased productivity and injectivity compared to barite and 100% CaCO3 RDIF systems used in the first field development. Additionally, the innovative RDIF reduced ECD, critical path RDIF filtering time prior to running completion screens, dilution rates, Liquid Mud Plant (LMP) volumes, and proved to be effectively reusable across multiple wells without compromising performance.