arXiv subjects

Yumeng Zhao

Publications and source records attributed to Yumeng Zhao.

3 recordsLinked to original sources

arXiv2026

Motile Bacteria Modify Salt Precipitation Patterns in Dried Sessile Droplet

Motile Escherichia coli bacteria can alter salt crystallization patterns during the evaporation of sessile droplets. In dilute bacterial suspensions in deionized water, dried bacteria cells predominantly accumulate at the droplet periphery, consistent with the classic "coffee-ring" effect. At higher cell densities, however, the bacterial distribution becomes more uniform. In the absence of bacteria, pure Phosphate Buffered Saline also forms salt crystals in a coffee-ring pattern. When bacteria are present alongside the salt solute, additional isolated crystals appear near the droplet center, with their abundance increasing with bacterial concentration, while crystals at the periphery adopt dendritic morphologies that extend radially. To investigate these phenomena, we used a Stokes-based analytical model to estimate the evolution of internal flow fields and compare them with bacterial motility. Then a finite volume model is implemented for bacteria and salt transport and adsorption, and a stochastic model for salt nucleation was developed, which successfully explains the crystallization pattern seen in the experiments. Our results show that bacterial motility can overcome evaporation induced flow during early stage, enabling bacteria cells to serve as nucleation sites and thereby altering the final crystalline morphology. This work highlights the potential of motile microorganisms to actively control evaporative crystallization, with implications for porous media flow and microfluidic deposition processes.

cond-mat.soft↗

arXiv2026

Agentic AI for Particle-Based Simulation: Automating SPH Workflows for Debris Flow Modeling

Physics-based simulation underpins engineering analysis but remains difficult to deploy in practice due to complex setup, parameterization, and interpretation. While Large Language Model-based agentic systems have shown promise in automating engineering computing workflows, they have primarily targeted structured, mesh-based problems. We present the first agentic AI workflow for meshless simulation in computational mechanics, demonstrated on debris flow modeling using Smoothed Particle Hydrodynamics (SPH) with the software DualSPHysics. By integrating tool orchestration, multimodal inputs (text and sketches), and human-in-the-loop interaction, the framework enables end-to-end simulation workflows for a class of problems that are inherently less structured and more challenging to automate. Results show that multimodal inputs not only enhance user experience but also reduces failure modes over text-only descriptions. Human-in-the-loop is critical for resolving ambiguities and handling SPH-specific configurations. We further introduce a cognitive-task-based evaluation of post-processing, showing strong performance in visualization and data extraction, with remaining gaps in higher-level SPH-specific physical reasoning that are amenable to improvement through domain-aware modeling. These results establish the viability of agentic AI for particle-based simulation and underscore its potential to transform the accessibility and efficiency of computational mechanics workflows.

cs.CE↗

arXiv2026

Drift-Aware Online Dynamic Learning for Nonstationary Multivariate Time Series: Application to Sintering Quality Prediction

Accurate prediction of nonstationary multivariate time series remains a critical challenge in complex industrial systems such as iron ore sintering. In practice, pronounced concept drift compounded by significant label verification latency rapidly degrades the performance of offline-trained models. Existing methods based on static architectures or passive update strategies struggle to simultaneously extract multi-scale spatiotemporal features and overcome the stability-plasticity dilemma without immediate supervision. To address these limitations, a Drift-Aware Multi-Scale Dynamic Learning (DA-MSDL) framework is proposed to maintain robust multi-output predictive performance via online adaptive mechanisms on nonstationary data streams. The framework employs a multi-scale bi-branch convolutional network as its backbone to disentangle local fluctuations from long-term trends, thereby enhancing representational capacity for complex dynamic patterns. To circumvent the label latency bottleneck, DA-MSDL leverages Maximum Mean Discrepancy (MMD) for unsupervised drift detection. By quantifying online statistical deviations in feature distributions, DA-MSDL proactively triggers model adaptation prior to inference. Furthermore, a drift-severity-guided hierarchical fine-tuning strategy is developed. Supported by prioritized experience replay from a dynamic memory queue, this approach achieves rapid distribution alignment while effectively mitigating catastrophic forgetting. Long-horizon experiments on real-world industrial sintering data and a public benchmark dataset demonstrate that DA-MSDL consistently outperforms representative baselines under severe concept drift. Exhibiting strong cross-domain generalization and predictive stability, the proposed framework provides an effective online dynamic learning paradigm for quality monitoring in nonstationary environments.

cs.LG↗