https://ri.ufs.br/jspui/handle/riufs/2439 2026-07-13T05:40:44Z https://ri.ufs.br/jspui/handle/riufs/25476 Título: Uma arquitetura big data inteligente para auditoria na saúde Autor(es): Santos, Helder Prado Abstract: Context: The auditing of public health procurement, particularly for Orthotics, Prosthetics, and Special Materials (OPME), is a critical process hindered by data inconsistency in documents such as invoices. This lack of standardization leads to inefficiencies and creates vulnerabilities for irregularities. Although Artificial Intelligence (AI) solutions have already demonstrated effectiveness in small-scale item classification, their nationwide application is impeded by the absence of an infrastructure capable of processing a massive volume of data in a scalable, robust, and economically viable manner. Objective: This work aims to overcome this barrier on two fronts: first, to characterize the state of the art in Big Data architectures for healthcare, identifying the fundamental approaches, tools, and concepts for their construction; second, to present ALIAS (Architecture for Large-scale Intelligent Auditing of Healthcare Systems), a detailed, replicable, and open-source technical blueprint designed to democratize the development of data platforms and support the complete lifecycle of AI solutions (MLOps). Method: The methodology adopted a two-phase approach. First, a Systematic Literature Mapping consolidated the state of the art. These findings then guided the design, implementation, and evaluation of the ALIAS blueprint. Its effectiveness and applicability were investigated through a concrete case study: the large-scale classification of OPME items from invoices, scaling a pre-existing AI solution to a nationwide data volume and seeking evidence of its performance in a real-world auditing scenario. Results: The systematic mapping analyzed 219 articles and selected 16 primary studies, which guided the design of ALIAS. The practical application of the architecture overcame previous processing barriers, enabling large-scale analysis that had previously failed due to memory exhaustion. The adoption of the Parquet format reduced storage by approximately 80%, and data partitioning accelerated queries by orders of magnitude. Crucially, the architecture established an efficient MLOps workflow, which democratized access to distributed analysis and drastically reduced the cycle between experimentation and production. Conclusion: The research presents evidence that the ALIAS blueprint constitutes a robust and financially viable solution for public institutions to implement their own data analysis and AI platforms. By promoting technological sovereignty and offering a practical guide whose effectiveness has been demonstrated, this work empowers organizations to optimize complex processes such as OPME auditing, establishing the foundation for future innovations and ensuring greater transparency, efficiency, and quality in public health management. 2025-07-30T00:00:00Z https://ri.ufs.br/jspui/handle/riufs/25475 Título: An analysis of the literature on SysML characteristics Autor(es): Santos, Diego de Jesus Abstract: The growing complexity of modern systems, particularly those that integrate hardware and software across multiple domains, has intensified the need for effective modelling approaches in Systems Engineering. Since its introduction by the Object Management Group (OMG), the Systems Modelling Language (SysML) has been proposed as a standardised language to support the specification, analysis, design, and verification of complex systems. This study presents a comprehensive analysis of the characteristics of SysML, emphasising its practical application in industrial contexts and its role in addressing engineering challenges over the course of nearly two decades. This study is based on a Systematic Literature Review that evaluates the main characteristics of SysML as defined by the OMG. The review investigates how the language is applied at different stages of the system’s development life cycle and supports multiple architectural views. In addition to theoretical analysis, the study discusses a practical application of Model-Based Systems Engineering (MBSE) in the automotive domain, focusing on an embedded window control system. A case study demonstrates how SysML and Unified Modeling Language (UML) can be applied together to model system components and their interactions in a structured and consistent manner. The results highlight that the combined use of SysML and UML is widely adopted in system design, development, and analysis, especially in scenarios that require integrated modelling of software and hardware components. Key benefits include greater architectural consistency, improved traceability, enhanced communication between stakeholders, and support for the development of complex embedded systems. Based on the evidence collected, the study concludes by proposing possible improvements to SysML, aiming to increase the effectiveness and efficiency of model-based systems modelling, thus contributing to both academic research and professional practice in industry and encouraging the combined use of SysML and UML in Systems Engineering. 2026-05-26T00:00:00Z https://ri.ufs.br/jspui/handle/riufs/25469 Título: Estimativa de custos em ambiente de desenvolvimento de software ágil para empresas de pequeno e médio porte Autor(es): Santos, Marcos Venícius Abstract: Cost estimation in agile software development environments remains a recurring challenge for Small and Medium-sized Enterprises (SMEs), particularly due to constraints related to organizational maturity, methodological practices, and limited resources. In this context, this study examines how companies adopting agile methodologies perform cost estimation and explores more consistent and systematized alternatives to support this process. The main objective of this research is to develop a cost estimation model tailored to the specific characteristics of SMEs that employ agile methods in software projects. The methodological approach is grounded in the analysis of the scientific literature and in the identification of recurring challenges reported in previous studies, enabling the proposal of a conceptual model accompanied by guidelines to support its practical application in agile environments. The findings indicate that the proposed model demonstrates conceptual coherence and potential practical applicability, particularly in organizational contexts with limited resources, although the lack of empirical validation in a real-world setting prevents the acquisition of conclusive quantitative evidence. The contributions of this work include the provision of a theoretical artifact that systematizes cost estimation in agile environments for the academic community and the proposal of a model aligned with the realities of SMEs for industry practitioners. Furthermore, this study lays the groundwork for future research on agile cost estimation supported by Artificial Intelligence and hybrid techniques. 2026-03-23T00:00:00Z https://ri.ufs.br/jspui/handle/riufs/25466 Título: Generalização de modelos de super-resolução em ALPR: um estudo sobre o impacto da degradação sintética no desempenho em domínios realistas Autor(es): Oliveira, Cristiano Lima Abstract: Automatic License Plate Recognition (ALPR) systems face challenges when processing lowresolution images, in which the license plate area may represent less than 0.3% of the total capture, compromising optical character recognition (OCR). Super-Resolution (SR) emerges as a promising technique for reconstructing textual details prior to OCR; however, the literature lacks systematic studies on how synthetic degradation protocols — typically limited to bicubic interpolation with Gaussian blur — affect the generalization capacity of SR models to real-world capture conditions. This dissertation investigates two questions: (1) the impact of synthetic degradation protocol complexity on the cross-domain generalization of SR models; and (2) how the choice of perceptual loss function affects the semantic reconstruction of text on license plates. To this end, models based on three state-of-the-art architectures — RealESRGAN, ESRGAN, and LPR-RSR-EXT — were trained on five progressively complex degradation protocols, yielding 30 model configurations evaluated across 150 cross-domain scenarios, with performance measured by the character recognition rate using the YOLOv8 model. The results reveal a non-monotonic relationship between degradation complexity and generalization: models trained with two degradation stages recognized up to 18 times more characters in out-of-domain tests than models trained solely with Gaussian blur, whereas protocols with three or four stages reduced average accuracy by up to 59%, indicating destruction of semantic features due to excessive degradation. Regarding the perceptual loss function, architectural dependency was observed: ESRGAN with OCR Loss achieved 18.1% full plate match under the one-stage protocol, compared to 7.4% with VGG Loss, but collapsed completely (0.0% accuracy) with two or more degradation stages; RealESRGAN with OCR Loss reached an average of 1.22 recognized characters, against 3.56 with VGG Loss; and LPR-RSR-EXT, specialized in text, failed with VGG Loss (0.0%) but operated with OCR Loss (1.86 average characters). VGG Loss demonstrated consistent robustness across all evaluated conditions, establishing itself as a safe default choice regardless of the architecture or degradation protocol adopted. 2026-01-28T00:00:00Z