Accurate and adaptive vehicle dynamics modeling is critical for the safety of autonomous driving systems, particularly under extreme maneuvers and time-varying parameters. Although the deep Koopman model provides a global linear representation, its online adaptation in high-dimensional lifted spaces can be sensitive to rank-deficient excitation and parameter drift. To address these issues, this paper proposes a tire-force-driven adaptive deep Koopman framework with offline physics-informed training and online normalized-projection adaptation. In the offline stage, longitudinal-lateral planar force equilibrium residuals derived from a 7DOF vehicle model are embedded into the learning objective to improve the physical consistency of the learned dynamics. In the online stage, the Koopman operator is updated via a Physics-Informed Variable Step-Size Normalized Least Mean Squares (PI-VSS-NLMS) algorithm with offline-prior anchoring regularization. The update avoids explicit covariance-matrix inversion and admits an explicit bound on the adaptive operator deviation. Extensive Car-Sim simulations and dSPACE real-time tests validate the framework's robust prediction accuracy across various conditions. Furthermore, the algorithm achieves an average execution time of 0.421 ms, demonstrating strong potential for onboard deployment.
As large language model (LLM) services become widely adopted, the cost of GPU resources for serving these models in cloud environments has emerged as a critical concern. Spot instances offer up to 90% cost savings over on-demand instances, but their frequent interruptions and limited availability pose significant challenges for continuous LLM serving. GPU spot instances, in particular, exhibit lower and more volatile availability than CPU-based instances, making homogeneous clusters that depend on a single GPU type vulnerable to correlated failures. Heterogeneous clusters spanning multiple GPU types can address this by leveraging complementary availability across diverse spot pools, yet existing LLM serving systems are designed for homogeneous environments and suffer from load imbalance on heterogeneous GPUs. This paper presents ShuntServe, a cost-efficient LLM serving system for heterogeneous spot GPU clusters. ShuntServe employs a roofline model-based performance estimator and a dynamic programming-based placement optimizer that jointly determines node configuration, parallelization strategy, and layer assignment to maximize throughput. To enhance fault tolerance under spot interruptions, ShuntServe combines output-preserving request migration with concurrent initialization via a shared tensor store, minimizing downtime by overlapping replacement node preparation with ongoing serving. Evaluated on Llama-3.1-70B and Qwen3-32B with a heterogeneous AWS cluster of L4, A10G, and L40S GPUs, ShuntServe achieves 1.42 & times; and 1.35 & times; higher throughput than state-of-the-art baselines and 31.9% and 31.2% cost efficiency improvements over on-demand instances for offline and online serving, respectively.
The co-pyrolysis of hazardous antibiotic sludge (AS) and waste tires is a promising valorization pathway; however, its implementation is hindered by high activation energy barriers, the release of toxic nitrogenous gases (e.g., HCN), and the formation of low-quality bio-oil. To address these challenges, this study synthesized a bimetallic hierarchical zeolite (Ni-Co/AZ) to modulate the co-pyrolysis pathways of AS with polyurethane/ rubber tires. Hydrogen spillover, generated from tire-derived radicals on the Ni-Co alloy surface, suppressed HCN formation and promoted aromatization. Thermokinetic analysis revealed that Ni-Co/AZ shifted the reaction mechanism from random chain scission (F1) to a three-dimensional diffusion-controlled model (D3), reducing the apparent activation energy by more than 50% (to approximately 28 kJ/mol). In-situ gas monitoring and liquid-phase analysis confirmed that this hydrogen spillover redirected toxic HCN precursors into gaseous NH3 and stable nitrogenous compounds (i.e., fatty nitriles and aromatic amines), achieving gas-liquid biphasic detoxification. Concurrently, the catalyst promoted deoxygenation via decarboxylation and hydrodeoxygenation, reducing the liquid-phase oxygenate content to 5.8%. The size-exclusion effect of the hierarchical zeolite network coupled with bimetallic active sites enabled microporous shape-selective aromatization, increasing the yields of monoaromatic hydrocarbons and light fuel gases while suppressing heavy polycyclic tars. Building upon these mechanistic insights, machine-learning-driven multi-objective optimization was employed to predict temperature and feedstock combinations that maximized co-pyrolysis efficiency and volatile product selectivity. This study provides molecular-level insights into metal-acid bifunctional catalysis, establishing an efficient and sustainable route for valorizing multiple solid waste streams.
The co-gasification of waste polypropylene pyrolysis oil (WPPO) and cashew nut shell (CNS) was investigated in a two-stage downdraft gasifier to produce hydrogen-rich, low-tar gas. Experiments were conducted under air (ER = 0.3) and steam (S/C = 2.5) conditions, with selected runs incorporating activated carbon (AC) in the secondary reactor. Steam gasification achieved a maximum hydrogen concentration of 77.9 vol%. The WPPO/CNS mixing ratio affected performance, with the 1:1 blend increasing CGE from 70.3 to 89.7% to 108.9% and CCE from 68.2 to 71.5% to 83.9%, along with enhanced gas yield. The use of AC reduced gas-phase tar, and a downstream dried AC impinger further decreased tar concentration to 0.45 mg/Nm3. In contrast, the composition of condensed tar showed a shift toward heavier species during steam gasification. GC-MS analysis at a WPPO/CNS ratio of 1:1 indicated that steam gasification promoted the formation of heavier PAHs due to enhanced thermal cracking and subsequent polymerization of aromatic intermediates.
This study conducted an intra-disciplinary review exploring how multinational enterprises (MNEs) contribute to sustainability in local markets within the fields of international business (IB) and strategy. It comprehensively examines four key themes of corporate social responsibility, economic (regional) development, social performance, and environmental sustainability of MNEs. By doing so, it seeks to consolidate fragmented findings from existing independent studies while broadening understanding through problematizing assumptions made in research. In addition, it sheds light on the complexities that MNEs face to address sustainability challenges, and how to strategically integrate economic, social, and environmental goals while adapting to global and local pressures. In turn, we propose new research directions and draw implications for MNEs in addressing grand challenges. Overall, this study contributes to deepening sustainability research in IB and strategy fields, and provides valuable insights for MNEs’ sustainable business practice.
