Large language models (LLMs) have become central to many applications, but their deployment often requires high-performance hardware, specialized libraries, and complex engineering, limiting accessibility for smaller organizations. Meanwhile, relational database systems (RDBMS) are widely used for portability, efficiency, and native support for managing large-scale data operations. This paper presents TranSQL1, a toolkit that enables transformerbased LLM inference within RDBMS. By translating neural operations into SQL queries and representing model weights as relational tables, TranSQL leverages database features like dynamic disk-to-memory data management and caching to reduce hardware and engineering demands for serving LLMs. Using the LLaMA3 8B model, we demonstrate TranSQL's ability to implement attention layers, KV-cache, and end-to-end text generation through SQL queries. TranSQL offers a cost-effective, portable, and scalable approach to making advanced AI technologies more accessible.

The quest for novel alternatives to hexavalent-chromium-based corrosion inhibitors is of utmost significance and urgency. Strict international health and safety regulations, due to growing concerns regarding the impact of hexavalent chromium on human health and the environment, have pushed the commercial introduction of many alternative inhibitor types, but the implementation of alternative active protective primers for structural parts in the aerospace industry is still pending. This endeavour has proven to be remarkably challenging, as the potential replacement coating types must meet numerous functional requirements encompassing cost-effectiveness and exceptional corrosion protection for intrinsically corrosion susceptible aerospace aluminium alloys. In recent years, considerable attention has been drawn to lithium salts as environmentally friendly corrosion inhibitors forming the basis for a novel active protective coating technology. The involvement of lithium ions has been shown to play a pivotal role in the conversion process of aluminium alloy surfaces by stabilizing the reaction products, thereby facilitating the gradual development of a protective layer with a multi-layered configuration, which exhibits considerable variability in morphology, depending on local chemical and electrochemical conditions. The versatility of the lithium-based corrosion protection extends to their application as corrosion inhibiting pigments in organic coatings or as a pre-treatment, directly forming conversion layers, thereby enhancing their practical implementation. However, previous chromate replacement reviews only introduced the promising outcomes provided by the lithium technology, omitting key details of its development and formation mechanism. This paper critically reviews and summarizes the studies conducted to date on lithium-based inhibitor technologies for the corrosion protection of aluminium alloys as well as topics to be investigated in the future.

In this work, the corrosion mechanism of AA2024-T3 covered by a lithium-based conversion layer is studied with high spatial and temporal resolution. Although the aluminium alloy surface is protected by a multi-layered conversion layer, areas around intermetallic phases (IMPs) represent weak spots due to an insufficient generation of a protective inner dense layer. For the freshly formed conversion layer, both the top and the inner layer undergo a gradual dissolution upon exposure to relatively dilute NaCl solution within 2 h due to their chemical instability. For the ambiently-aged conversion layer, most corrosion activity around IMPs is related to the S-phase and large constituent phases, due to their active nature and the lower local conversion layer quality, respectively. Moreover, S-phase-related corrosion activity lasts approximately 8 h due to fast dissolution whereas reactions induced by large constituent particles remain active over the entire re-immersion period of 12 h.

The advent of Large Language Models (LLMs) provides an opportunity to change the way queries are processed, moving beyond the constraints of conventional SQL-based database systems. However, using an LLM to answer a prediction query is still challenging, since an external ML model has to be employed and inference has to be performed in order to provide an answer. This paper introduces LLM-PQA, a novel tool that addresses prediction queries formulated in natural language. LLM-PQA is the first to combine the capabilities of LLMs and retrieval-augmented mechanism for the needs of prediction queries by integrating data lakes and model zoos. This integration provides users with access to a vast spectrum of heterogeneous data and diverse ML models, facilitating dynamic prediction query answering. In addition, LLM-PQA can dynamically train models on demand, based on specific query requirements, ensuring reliable and relevant results even when no pre-trained model in a model zoo, available for the task.

In this work, four different techniques were concurrently applied to study the interplay between local electroactivity and electrode surface characteristics of free-standing, polycrystalline boron-doped diamond (BDD). Scanning electron microscopy, electron back-scatter diffraction, Raman mapping and scanning electrochemical microscopy were used to probe the electrode morphology, grain orientation and boundaries, composition, and local electrochemical activity, respectively. Both nucleation and growth BDD surfaces together with the cross-section area were carefully investigated for the first time in a single study using the combination of all four techniques. This enabled us to obtain significant insights into the highly heterogeneous nature of the polycrystalline BDD material. Notably, boron dopants were confirmed to be non-uniformly distributed over the BDD material, which is characterized by a distinct columnar structure and composition of grains of various orientations. Particularly, the highest electrochemical activity was recorded on the highest doped (111) crystal orientation. In contrast, the averagely boron-doped (100)-oriented facet showed non-conductive nature. This highlights that the local electrochemical activity of the BDD surface is strongly grain-dependent and the most significant factors governing the obtained responses are crystallographic orientation and boron doping. Moreover, increased boron and sp² carbon content in the boundary regions was recognized by Raman mapping. However, such localized enrichment in impurities did not translate into enhanced electrochemical activity, which implies that boron atoms at the inter-grain areas are predominantly inactive. Finally, it is crucial to consider all characteristics of the polycrystalline BDD including crystal orientation, which is particularly relevant if micro- and nanoscale probing is intended.

The present article investigates the influence of chemical composition and phase fractions on the corrosion behaviour of industrially produced quenching and partitioning (Q&P) martensitic stainless steels. Localised corrosion was analysed by scanning Kelvin probe force microscopy (SKPFM) and scanning electrochemical microscopy (SECM) in 3.5 wt.% NaCl solution. SKPFM revealed a Volta-potential difference of around 40 mV between inclusions and the matrix, which is larger than the Volta potential variations within the matrix. This difference in surface potential is a driving force for selective dissolution (corrosion initiation) at inclusions and inclusion/matrix interfaces. SECM detected early pitting initiation, particularly in alloys containing MnS and TiN inclusions. Results suggest that pitting initiation and propagation occur at those specific regions. This study emphasised that irrespective of chemical composition and phase fraction, localised corrosion initiation in Q&P-processed martensitic stainless steels is predominantly governed by the presence of inclusions.

This study investigates the microbiologically influenced corrosion (MIC) of X80 steel accelerated by the phototrophic bacterium Rhodopseudomonas palustris TIE-1. The photorespiration plays a key role in promoting extracellular electron transfer (EET)-induced MIC. In the early corrosion stage, unstable localized corrosion dominated in the dark, while intense diffusion-controlled corrosion occurs in light. Compared to the sterile anaerobic medium, R. palustris TIE-1 accelerated corrosion of X80 steel, with a significantly higher corrosion rate under light conditions, approximately three times that of dark conditions. Inhibition of photosynthetic electron transfer or cessation of photostimulation resulted in pronounced reduction in the corrosion rate.

In this work, three distinct heteroepitaxial single-crystal boron-doped diamond (SC-BDD) electrodes were fabricated and subjected to detailed surface analysis and electrochemical characterization. Specifically, the heteroepitaxy approach allowed to synthesize large-area (1 cm²) and heavily-doped (100)-oriented SC-BDD electrodes. Their single-crystal nature and crystal orientation were confirmed by X-ray diffraction, while scanning electron and atomic force microscopies revealed marked variations in surface morphology resulting from their growth on respective on-axis and off-axis substrates. Further, absence of sp² impurities along with heavy boron doping (>10²¹ cm⁻³) was demonstrated by Raman spectroscopy and Mott-Schottky analysis, respectively. Cyclic voltammetry (CV) in a 0.1 M KNO₃ solution revealed wide potential windows (∼3.3 V) and low double-layer capacitance (<4 μF cm⁻²) of the SC-BDD electrodes. Their highly conductive, ‘metal-like’ nature was confirmed by CV with [Ru(NH₃)₆]^3+/2+ probe manifesting near-reversible redox response with ΔE_p approaching 0.059 V. The same probe was used to record scanning electrochemical micrographs, which clearly demonstrated homogeneously distributed electrochemical activity of the heteroepitaxial SC-BDD electrodes. Minor differences in their electrochemical performance, presumably resulting from the somewhat different morphological features, were only unveiled during CV with surface sensitive compounds [Fe(CN)₆]^3−/4− and dopamine. The latter was also used to show the possibility of applying herein developed heteroepitaxial SC-BDD electrodes for electrochemical sensing, whereas experiments with anthraquinone-2,6-disulfonate revealed their enhanced resistance to fouling. All in all, heteroepitaxial SC-BDD represents a highly attractive electrode material which can, owing to the fabrication strategy, easily overcome size limitation, currently preventing broader use of single crystal diamond electrodes in electrochemical applications.

Scanning electrochemical microscopy (SECM) is employed to characterize the evolution of local electrochemical surface activity during lithium-based conversion layer formation on legacy aerospace aluminium alloy AA2024-T3. Initially, three types of studied intermetallic particles - S-, θ- and constituent phases - act as active cathodic areas. Subsequently, θ- and constituent phases show passivation preceding that of S-phase particles during the later conversion layer formation stages. The entire surface, including the matrix region, shows a higher reactivity at the beginning and then gradually shows decreasing reactivity. Hydrogen evolution-generated bubbles attach to the alloy surface and locally hinder the conversion layer formation, weakening the corrosion protection the conversion layer provides at those locations.

The influence of ageing under ambient conditions on the corrosion protective behaviour of a lithium-based conversion layer on AA2024-T3 is studied in this work. Conversion layers aged at ambient conditions for relatively short times (0 h and 4 h), show an initial high degree of corrosion inhibition but a much lower protectiveness after the inhibition stage terminates. Conversion layers with relatively long ageing times (24 h and 72 h) show a rather stable corrosion resistance which is higher than that of short-time aged samples. It is hypothesized that the freshly-formed conversion layer has trapped a certain amount of lithium ions and water molecules, leading to ongoing and heterogeneous growth of the conversion layer with time under ambient indoor conditions. Moreover, conversion layers with short ageing times show early-stage active corrosion protection by lithium-ion release.

The formation process of a lithium-based conversion layer on AA2024-T3 and its corrosion protective behavior are studied using electrochemical noise (EN). Wavelet transform, as well as noise resistance analysis, have been employed to interpret the EN data. The EN data confirmed five different stages during the conversion layer growth, accompanied by anodic dissolution, increasing corrosion protection of the conversion layer, and adsorption, growth and desorption of hydrogen bubbles simultaneously. The detachment of hydrogen bubbles, localized and uniform corrosion generate different features in the EN signals with energy maxima in high, intermediate and low frequency bands, respectively. In addition, EN results show that the lithium-based conversion layer still provides efficient protection after re-immersion in a corrosive environment, even though localized damage occurs. Moreover, the EN data corresponds well with the morphological layer formation and breakdown observed with microscopy techniques. The results demonstrate that EN is a powerful tool to provide continuous time- and frequency-resolved information about inhibition efficiency.

The microbiologically influenced corrosion of pure iron was investigated in the presence of Shewanella oneidensis MR-1 with various levels of exogenous riboflavin (RF) serving as electron shuttles for extracellular electron transfer (EET). With more RF available, a larger and denser phosphate layer was formed on the surface of pure iron by the bacteria. The results of electrochemical impedance spectroscopy, linear polarization resistance and potentiodynamic polarization tests showed that the product layer provided good corrosion protection to the pure iron. Using electrochemical noise, we observed that the addition of RF accelerated the corrosion at the initial stage of immersion, thereby accelerating the deposition of products to form a protective layer subsequently.

The influence of outward extracellular electron transfer (EET) of Pseudomonas aeruginosa in accelerating corrosion of 304 stainless steel was investigated. With less NO₃⁻ available as electron acceptor, P. aeruginosa biofilm accelerated the pitting corrosion. The ICP-MS and XPS results indicated that P. aeruginosa promoted the bioreductive dissolution of iron oxides in the passive film of stainless steel. Using in situ scanning electrochemical microscopy, we established a relationship between this accelerated deterioration of the passive film and the EET process mediated by the conversion of the redox states of pyocyanin secreted by P. aeruginosa.

Microbiologically influenced corrosion of metals is prevalent in both natural and industrial environments, causing enormous structural damage and economic loss. Exactly how microbes influence corrosion remains controversial. Here, we show that the pitting corrosion of stainless steel is accelerated in the presence of Shewanella oneidensis MR-1 biofilm by extracellular electron transfer between the bacterial cells and the steel electrode, mediated by a riboflavin electron shuttle. From pitting measurements, X-ray photoelectron spectroscopy and Mott-Schottky analyses, the addition of an increased amount of riboflavin is found to induce a more defective passive film on the stainless steel. Electrochemical impedance spectroscopy reveals that enhanced bioanodic and biocathodic process can both promote the corrosion of the stainless steel. Using in situ scanning electrochemical microscopy, we observe that extracellular electron transfer between the bacterium and the stainless steel is bidirectional in nature and switchable depending on the passive or active state of the steel surface.