卡特彼勒

Engine-out soot or particulate matter (PM) emissions are an unavoidable consequence for many direct-injection mixing-controlled compression-ignition engines. Understanding soot formation and oxidation processes is necessary for pollutant reduction and compliance with future Tier V emissions. Since nearly all exhaust PM measurements are averaged over many cycles, transient cycle variable behavior is generally unknown, and yet, awareness of these variations is essential for simulation and prediction. In this work, cycle soot variations are quantified and examined in a 2.53 L single-cylinder direct-injection compression-ignition engine using a novel laser-based extinction diagnostic in the exhaust runner. Measurement accuracy is better than 0.5 ppb exhaust soot volume fraction and temporal resolution is faster than 0.5 crank angle degrees (CAD). Exhaust soot volume fraction history is compared with cycle resolved apparent heat release rate to better understand in-cylinder processes that lead to PM formation. Engine cycle soot variations are high, that is, greater than 10% COV for a wide range of operating conditions. Minimum engine cycle PM variations are equal to injector shot PM variations for non-engine quiescent conditions without spray-wall interaction, suggesting that spray characteristics establish a governing baseline. For some engine operating conditions, exhaust averaged soot volume fraction can vary by as much as an order of magnitude from cycle to cycle. Skewed, non-normal cycle soot populations indicate instability and non-optimal operation points for the hardware employed, and thus, opportunities for improvement. Comparison of heat release rate (HRR) profiles for low- and high-soot cycles reveals statistically significant correlations between engine-out PM and specific combustion intervals. High-PM cycles generally exhibit common features, including: early ignition with advanced premixed burn, adverse spray-wall (piston and head) interactions with reduced heat release rates, and diminished late cycle burnout. This soot measurement and analysis approach represents a useful new tool for combustion system design, troubleshooting, and simulation validation.

Web application performance is a primary determinant of user experience, scalability, and operational cost, yet optimization remains challenging in modern distributed architectures due to dynamic traffic patterns and heterogeneous diagnostic data. Conventional techniques such as load balancing, caching, and database tuning are effective but typically require extensive manual analysis and expert intervention. This paper investigates the use of Google Gemini as a decision-support component for web application performance engineering, operating offline in an advisory capacity to analyze workload summaries, slow-query traces, and system logs, and to generate actionable optimization recommendations. We evaluate four advisory dimensions: 1) query optimization support, 2) caching recommendations, 3) workload analysis for bottleneck identification, and 4) behavior-aware caching strategies based on aggregated user interaction patterns. Experimental results show that Gemini-informed optimizations reduce average response latency by up to 2- 3 & times; , increase throughput by up to 2 & times; under high concurrency, and improve cache hit ratios to approximately 50% in read-dominant workloads, while maintaining comparable performance under low concurrency. These results demonstrate that large language models can provide practical, explainable, and scalable guidance for performance optimization when positioned as advisory systems within established engineering workflows. Cross-layer reasoning over SQL execution plans, application logs, and workload summaries reduces engineering analysis time by an estimated 85-90%, based on practitioner observation rather than controlled experimental measurement.

A hydrogen (H Subscript 2 2 $_2$ ) jet in crossflow (JICF) was studied using direct numerical simulations (DNS), large-eddy simulations (LES) and Reynolds-averaged Navier–Stokes (RANS) approaches, based on a geometric representation of port fuel injection (PFI) in an H Subscript 2 2 $_2$ -fuelled heavy-duty internal combustion engine. The focus of this work was on evaluating the turbulent species flux models used in LES and RANS approaches using DNS of an H Subscript 2 2 $_2$ JICF with realistic PFI flow conditions and mixture properties. Results showed that LES perform very well in predicting both the mean velocity and the Reynolds stresses. In contrast, RANS significantly under-predicts all Reynolds stress components, while predicting the mean flow field relatively well. Regarding the prediction in H Subscript 2 2 $_2$ mixing, LES show excellent agreement with DNS, while RANS significantly under-predicts the mixing process. The underlying reasons for the poor performance of RANS were systematically investigated by extracting relevant turbulent transport properties used in the RANS approach from the DNS data. It was shown that the turbulent diffusivity used in RANS is much smaller than that derived from DNS. This under-prediction was found to be attributed to the over-prediction on the turbulent Schmidt number ( italic Sc Subscript t Sc t $ \textit{Sc}_t$ ), as well as the under-prediction on the turbulent viscosity. By further analysing the anisotropy of italic Sc Subscript t Sc t $ \textit{Sc}_t$ , the commonly used assumption of isotropic turbulent diffusivity in RANS was demonstrated to be invalid for the present configuration. This study provided a unique DNS dataset for H Subscript 2 2 $_2$ JICF relevant to H Subscript 2 2 $_2$ PFI engines, and generated new insights on improved modelling of turbulent mixing.