This paper evaluates precision demotion in Marlim3, an open-source 1D multiphase flow simulator developed by Petrobras. We replace long double declarations, extended-precision literals, and long-double libm calls with double-precision equivalents. The original and modified versions are compared using nine simulation cases on x86-64 and ARM platforms, considering runtime and numerical differences in the outputs. On x86-64, the modified version improved all workloads, reaching up to 11.69× speedup. On Apple M4, most workloads remained close to the baseline.

The use of high-performance computing (HPC) solutions has expanded beyond the traditional context of scientific research and engineering, now supporting data-intensive applications across multiple domains. In this scenario, the storage subsystem becomes crucial for overall performance, scalability, and data availability. This paper discusses the role of three file system architectures in HPC: Network File System (NFS), Distributed File Systems (DFS), and Parallel File Systems (PFS).

This paper proposes the use of Large Language Models (LLMs) and Prompt Engineering to classify the runtime of reservoir simulations in HPC environments, aiming to optimize resource allocation. Three strategies (Naïve, Expert, and Few-Shot) were compared using 10,000 scripts. The Few-Shot approach, which incorporated historical statistics, achieved the best performance with an accuracy of 76.83%, outperforming traditional methods by interpreting code semantics without prior execution. Although promising, the results on imbalanced classes indicate that future improvements require fine-tuning or data augmentation strategies to enhance classification robustness in complex scenarios.

This paper presents a reflective systematization of design practice in innovation projects, supported by participant observation and interviews with experts at an innovation center operating in the Global South. The analysis of the empirical material followed the principles of Reflexive Thematic Analysis. The investigation stems from structural challenges and institutional frictions in R&D&I ecosystems to propose the Connect-Materialize-Sustain (CMS) model as an emerging operational synthesis for understanding the role of strategic design in innovation projects.

The application of service design methods in the construction of complex digital solutions proved fundamental in this empirical study for the development of a diesel hydrotreatment simulator at PETROBRAS. Faced with the limitations of legacy tools, the research investigated how design can align technical professionals in a participatory construction. The study began with a discovery phase, via interviews with basic engineering, researchers, and representatives from nine refineries, aimed at understanding the service and structuring a service blueprint, followed by iterative validations.

This paper investigates the role of generative AI as a technical-project-based mediation in the creative data visualization design process regarding climate disasters recorded in Brazil. It departs from a recurring problem in the field of visualization: the materialization of unconventional forms when a project requires programming, especially when conducted by designers without a computational background. Through Research through Design, the article analyzes three design cycles that involved manual experimentation, the use of visualization tools, and, subsequently, AI-assisted prototyping.

Requirements engineering is a critical phase of software development, demanding strategic tools for time optimization. This paper presents the development and evaluation of a parametrized AI agent for the automated mapping of requirements from meeting minutes and transcripts. This is an instrumental contribution aimed at automating information management and productivity during project handoffs. As a scope limitation, this preliminary study evaluated 42 requirements derived from a single digital product.

This paper examines the transformation of the user experience (UX) designer's role in building conversational interfaces powered by Large Language Models (LLMs). The probabilistic nature of these systems hinders the formation of accurate mental models and poses challenges such as a loss of control over the interaction and a mismatch between user expectations and the tool's actual capabilities. Based on a case study applied to a corporate assistant built on a RAG (Retrieval-Augmented Generation) architecture, this work proposes a resilience process structured around three pillars: Anticipation, Mitigation, and Recovery.

Job Safety Analysis (JSA) acts as a fundamental tool for occupational risk management in an oil and gas sector corporation. However, practical difficulties in its application motivated a study aimed at understanding its usability obstacles. To this end, the methodology was based on document evaluation, a literature review, and the analysis of data collected from interviews with the tool's actual users.

This paper presents a deaf-centered design approach applied to communicational inclusion in oil and gas industrial environments. The research examined the barriers encountered by deaf workers and co-developed accessible solutions through participatory methodologies grounded in inclusive design principles, which recognize diversity as a core component of the design process. The study included workplace immersion, interviews with both deaf and hearing employees, and collaborative workshops involving diverse members of the deaf community, Brazilian Sign Language (Libras) interpreters, and accessibility experts.

This paper introduces an AI-assisted qualitative analysis framework that integrates automated interview processing with human interpretation. The approach expands the capacity to identify patterns and opportunities across large datasets, while preserving critical expert validation and contextualization. The findings highlight the potential of human-AI collaboration to enhance research and support more evidence-based design decisions.

This work describes the evolution of the Internal Product Metrics Dashboard (PMProd) from a data visualization tool into a user-centric ecosystem, incorporating UX practices, periodic surveys, an AI agent, and support content. The initiative led to higher adoption rates, increased user engagement, and stronger community involvement. This experience underscores the critical role of user-centered design methodologies in ensuring the success of enterprise digital products.

AI-driven Semantic Segmentation has revolutionized Petrography by enabling quantitative analysis of thin sections. Nevertheless, the resulting thematic maps rely on color as the main means of distinguishing mineral classes. While color is a fundamental attribute in petrographic analysis, conventional palettes that are not optimized for visual deficiencies pose a significant accessibility barrier for color-blind professionals, thereby undermining inclusivity in scientific research. This paper presents a user-centered design process aimed at developing accessible visual representations for petrographic analysis tools.

The project develops a Digital Twin for integrity management and fatigue monitoring of rigid risers in operation. Through a scalable, event-driven software architecture, the platform integrates real operational data with physical models (finite element and VIV) to orchestrate and automate large-scale structural simulations. By accurately predicting accumulated damage, the solution feeds graphical dashboards that reduce analysis time and support agile decisions related to safety, sustainability, and cost efficiency.

This presentation introduces the architecture and principal technical challenges associated with developing a Digital Twin platform for fatigue and CO₂-induced stress corrosion cracking (SCC-CO₂) monitoring in flexible pipes deployed in subsea systems. The architectural decisions made to ensure the solution's modularity and scalability will be detailed, with a particular focus on integration with legacy operational systems and heterogeneous industrial databases. The platform adopts a microservices-based architecture and hybrid cloud connectivity. The sensor data ingestion, normalization, and validation pipeline, along with the automatic simulation reprocessing mechanisms, will be presented as a core element of the platform's robustness. Horizontal scalability strategies to efficiently and resiliently support multiple assets will also be discussed.

From an occupational safety and process perspective in industrial activities, recording undesirable occurrences with the potential to become safety events is paramount. Low digitization at this stage can compromise the quality of the collected data, making it essential to provide the observer with a quick way to report the event. This article presents the use of digital transformation techniques to map and review incident recording processes, focusing on user-centered design, making them more agile and accessible, and facilitating the integration of professionals.

This paper presents an empirical case study of a collaborative, interdisciplinary process engaging multiple departments and institutions to develop business operation strategies within a complex context. Rooted in Strategic Design and Innovation principles, the study draws on a four-month experimentation and Design project conducted by an Innovation Institute in partnership with an Oil and Gas Company. The project facilitated the design of operational strategies for a Decision Support System (DSS) aimed at managing Brazil's Water Resources and the associated risks posed by Climate Change.

This study explores advanced techniques leveraging Large Language Models (LLMs) and intelligent agents for the detection of domain-specific entities—including wells, fields, and oil platform names—within conversational search systems for the oil and gas (O&G) sector. Experiments conducted on domain-specific datasets, combining LLMs, intelligent agents, domain rules, data dictionaries, and data preprocessing pipelines, demonstrate effective and reliable extraction of key entities from both structured and unstructured data. In particular, for unstructured data, improved detection of domain-specific entities such as well names significantly enhances conversational search relevance, achieved through advanced prompt engineering, metadata enrichment, and performance optimization.

SAVIA, a solution developed at Petrobras, employs artificial intelligence to automate and support decision-making in well design. Among its components is the Complexity Indicator, which formerly required manual form completion and has now been automated through AI agents. These agents are able to interpret project data and respond to the indicator's queries, thereby reducing manual effort and subjectivity. This approach renders the process more efficient and consistent with Petrobras' innovation agenda.

This paper introduces an application of Physics-Informed Neural Networks (PINNs) to diesel hydrotreating, integrating experimental data with physicochemical knowledge to build models that are more interpretable and generalizable than purely data-driven alternatives. Preliminary findings demonstrate enhanced predictive accuracy for product quality variables. To the best of our knowledge, this represents a novel application of PINNs to diesel HDT, holding promise for the optimization and real-time monitoring of HDT units.

Hydrotreating (HDT) is a critical process in fuel refining, ensuring compliance with current product specifications. Nevertheless, it is a high-cost operation, and its optimal performance relies on tools capable of providing accurate simulations of operational scenarios. In this context, artificial intelligence-based mathematical models were developed for low-sulfur diesel HDT to quantify the impact of main operational variables on process performance and to extend the models' predictive range. The applied methodology yielded reliable performance predictions, supporting the definition of HDT unit operating conditions, feedstock selection, and the reduction of off-specification diesel production.

This article highlights the partnership between CNOOC, CESAR, and UFAL to revolutionize offshore exploration in the Buzios pre-salt field. The project allocates R$18.5 million to CESAR for the development of advanced solutions using digital twins and AI, enabling the anticipation of operational failures and the creation of scalable technologies for the global oil and gas market

This work presents a modular and orchestrated pipeline that automates document ingestion, extraction, semantic enrichment, and indexing. The proposed framework integrates Artificial Intelligence (AI) and Natural Language Processing (NLP) models to extract and enrich semantic metadata from multimodal sources, including text, images, and tables.

The presented solution estimates fatigue damage based on commonly measured parameters, specifically BOP stack and riser angles. Combined with riser, BOP, and wellhead data, it obtains a time series of bend moment and shear force at the wellhead, and then a transfer-function approach based on FEA of just the wellhead system to obtain stress time history at critical locations along it. The whole process is automated through a web-based application that integrates internal databases to quickly retrieve data necessary for the assessment, perform all necessary analysis, and present results through user-friendly dashboards.

This paper reports a case in which a collaborative glossary was developed during user immersion and adopted as a guiding artifact for AI design and engineering through participatory sessions involving domain experts, designers, software engineers, and data scientists. The resulting living document defined technical and operational terms while translating everyday practices, metaphors, and user needs; it also mapped synonyms, terminological variations, and contexts of use to align ambiguous expressions with real intentions.

This study investigates the application of Information Design and Human Factors in redesigning Job Safety Analysis (JSA) in Petrobras refineries, aiming to improve risk communication in highly hazardous industrial environments.

This article presents the results of a design project aimed at including deaf employees in a Petrobras refinery, focusing on overcoming communication barriers in the industrial environment.

This article details the evolution of SAVIA, presenting results from the retraining process of machine learning models based on an updated database that included projects executed between 2023 and 2024. During the retraining process, in addition to the previously used models, the XGBoost model was introduced into the experimentation process.

This article presents an approach to the use of RISC-V in high-performance computing, outlining its main characteristics, applications, and current development.

This work evaluates machine learning models for predicting the execution time of oil reservoir simulation jobs in HPC clusters.

This study aims to understand the challenges of applying SSI to the constrained IoT through a Systematic Mapping Study (SMS). The results showed that although there is a limited number of identified articles, research has grown in the last six years, with more than half of the studies published in the last two years.

This article presents a digital application and a method developed to estimate and monitor the consumption of spare parts for industrial equipment, aiming to solve the lack of information for defining these items.

Sistemas complexos na indústria, como os ecossistemas de dados batimétricos, impõem alta carga cognitiva e desafios de design de interação. O objetivo do artigo é relatar a experiência de aplicação de métodos de co-design e validação tangível para reestruturar um sistema de análise de dados no setor de óleo e gás. Para isso, conduziu-se um workshop com 12 profissionais utilizando o protocolo Thinking Aloud e o Markup Físico (prototipagem tangível de baixa fidelidade) para mapear fluxos, externalizar a carga cognitiva e validar requisitos táticos.

Em organizações complexas, definir escopos de inovação (PD&I) envolve ambiguidades e pressões organizacionais associadas a estruturas rígidas de decisão. Este artigo apresenta a proposta do conceito Design Before the Deal, elevando o discovery design de etapa preparatória a infraestrutura

sociotécnica de decisão. A metodologia empregda foi uma pesquisa qualitativa baseada em revisão

teórica e análise temática de entrevistas com especialistas.

This article presents an application of the Automatic Query Expansion (AQE) technique for information retrieval in a collection of geoscientific documents in Portuguese. The proposed AQE uses a knowledge base composed of thesauri, glossaries, and lists of entities related to the Oil and Gas domain, selected with the help of experts in the geoscientific domain.

The article presents SAVIA (Selection of Alternatives via Artificial Intelligence), an AI-based solution developed under Petrobras' Internal Startups Program in partnership with CESAR, aimed at innovating the well design process known as SELEPOÇO. The tool assists well designers by suggesting alternative well configurations for investment projects, with a focus on improving technology integration and cost optimization.

In this work we study the performance and accuracy of machine learning models trained for this task, and analyzed the impact of their deployment in a microcontroller-based System-on-Chip used by IoT devices, along with the impact of TinyML-based optimizations, such as vectorization and quantization.

This article discusses the development of an intelligent lighting control system based on the Internet of Things (IoT), focusing on Industry 4.0. The system aims to optimize energy consumption in industrial environments, promoting greater efficiency and sustainability.

This article presents a practical example of the use of digital twins in a TV production line, where different neural network models and computer vision were used for data analysis.

This article describes a monitoring and control system for industrial tanks, focusing on the precise measurement of tank level and the quality of the stored liquid. Ultrasonic, temperature, and gas sensors were used, allowing the system a comprehensive view of the tank conditions.

This work presents the development and implementation of a GPS tracker based on LoRaWAN technology as a solution to overcome these limitations. The proposal uses IoT devices equipped with LoRa modules and GPS receivers, integrated into a LoRaWAN communication infrastructure, to track inputs in real time, even in industrial environments with high levels of Wi-Fi network noise.

The purpose of this study is to develop a project for detecting toxic and flammable gases with applications in various environments. A prototype will be created using commercially available sensors, namely the MQ-2 sensor, capable of detecting concentrations of combustible gases and smoke in the air; the MQ-3 sensor, capable of detecting alcohol and ethanol vapors (a sensor commonly used in breathalyzer projects); and the MQ-135 sensor, capable of detecting concentrations of gases harmful to air quality.

This paper proposes automatic monitoring of the quality of multimedia content transmitted between a smartphone and a digital TV via a wireless transmission channel. The methodology combines two tools: the peak signal-to-noise ratio (PSNR) and the structural similarity index (SSIM).

The present study aims to explore the familiarity of managers, leaders, and developers with software visualization tools and how these tools are employed to facilitate the technical onboarding of new team members. Method. To address the research problem, we built upon the insights gained through the literature and embraced a sequential exploratory approach. This approach incorporated quantitative and qualitative analyses of data collected from practitioners using questionnaires and semi-structured interviews.

This paper investigates the integration of generative AI tools within software development, focusing on understanding their uses, benefits, and challenges to software professionals, in particular, looking at aspects of productivity. Through a pilot case study involving software practitioners working in different roles, we gathered valuable experiences on the integration of generative AI tools into their daily work routines.

This article discusses the adoption of technology in Brazil, where U˜4.0 (CESAR and partners) creates digital twins of ovens and heaters, reducing gas consumption by up to 15% using sensors and machine learning, generating savings and competitiveness.

This work presents a precise methodology to define the common requirements that should be present in any IoT platform. After defining these requirements, a detailed survey is carried out among 32 specialists to prioritize these requirements, identifying which of them are essential. As a result, among a set of 64 mapped requirements, 26 are discussed regarding their relevance.

This work presents the doctoral research for the design of a meta-architecture for Internet of Things platforms, that addresses the interoperability challenges and the complexity in building IoT solutions.

This chapter explores the applications of Digital Twin (DT) technology in the Brazilian energy sector and its impact on businesses and society. It highlights how DT applications have contributed to cost reduction, human error mitigation, operational optimization, and technical failure prediction. The chapter also discusses the implementation process and the requirements for developing these systems.

This chapter presents a summary of how the most diverse industry sectors such as smart cities, oil, gas sector, energy and other areas are applying digital twins as a tool to support the digitization of companies.

This work aims to propose a study to develop an electronic device using Arduino capable of monitoring the operation of a three-phase induction motor in industry.

This article proposes the development of a data manipulation and visualization layer for thermal sensors in BIM models, represented in a virtual reality environment. The methodology includes the creation of these models to allow real-time monitoring. The results indicate that the system assists in the analysis of thermal comfort conditions in indoor environments, contributing to the efficient management of facilities and the well-being of users, integrating objective data with virtual immersion.

This work proposes a framework based on the following strategies, widely used in the literature: Genetic Algorithms (GA), Particle Swarm Optimization (PSO) and Linear Programming, besides our Heuristic approach. Furthermore, we restricted the resource allocation optimization to a real-time scenario, facilitating its use in an industrial process.