Rafael Pieper

Estudo de técnicas de análise de sinais

Rafael Pieper — Wed, 09 Jul 2025 19:09:34 +0000

Estudo de Técnicas de Análise de Sinais em EEG de Pacientes com Crises Epilépticas

Author: Rafael B. Pieper

Programa de Pós-Graduação em Engenharia de Sistemas Eletrônicos (PPGESE) / UFSC

ABSTRACT – This paper presents a comparative analysis of signal processing techniques applied to EEG recordings containing epileptic seizures, using data from dataset “E” of the University of Bonn database. The study includes the Fourier Transform, windowed transforms, and the Wavelet Transform, focusing on the identification of frequency components and non-stationary variations over time. Additionally, Butterworth and Chebyshev filters were applied to isolate specific frequency bands associated with possible epileptic activity. The results highlight the differences among the approaches in terms of temporal and spectral resolution, emphasizing the potential of

RESUMO – Este artigo apresenta uma análise comparativa de técnicas de processamento de sinais aplicadas a registros de EEG com crises epilépticas, utilizando dados do conjunto “E” da base da Universidade de Bonn. O estudo contempla a Transformada de Fourier, Transformadas Janeladas e a Transformada Wavelet, com foco na identificação de componentes de frequência e variações não estacionárias no tempo. Além disso, foram aplicados filtros do tipo Butterworth e Chebyshev para o isolamento de bandas específicas associadas à possível atividade epiléptica. Os resultados evidenciam as diferenças entre as abordagens em termos de resolução temporal e espectral, destacando o potencial das técnicas para o estudo de sinais complexos como os de EEG.

Resenha Crítica – Análise de Sinais de Vibração em Propulsores de USVs

Rafael Pieper — Wed, 09 Jul 2025 18:56:44 +0000

Revisão crítica: Detecção de falha em propulsores de USV através de transformadas Wavelet em sinais de vibração

Author: Rafael B. Pieper

Programa de Pós-Graduação em Engenharia de Sistemas Eletrônicos / UFSC

Abstract– This critical review evaluates the study of Cho et al., which proposes a failure detection and classification system in UPS thrusters using vibration analysis via Continuous Wavelet Transform in a Visual Transformer (ViT) algorithm. The relevance of the article lies in the importance of monitoring and mitigation of failures and operational risks on the high seas. The technical analysis highlights the simulation of failure scenarios and the accuracy of the model, however, methodological gaps have been identified that deserve attention in future work. Despite these observations, the study reinforces the potential of integration between advanced signal analysis and machine learning for predictive maintenance monitoring. The case of an actual occurrence, presented with ropes in the thruster, illustrates the real-world applicability of the technology. It is concluded that, although promising, the work of Cho et al. highlights the need for greater methodological rigor to ensure the reproducibility and practical implementation of such innovations.

Resumo— Esta resenha crítica avalia o estudo de Cho et al., que propõe um sistema de detecção e classificação de falhas em propulsores de USVs utilizando análise de vibração via Transformada Wavelet Contínua em um algoritmo Visual Transformer (ViT). A relevância do artigo reside na importância do monitoramento e na mitigação de falhas e riscos operacionais em alto mar. A análise técnica destaca a simulação dos cenários de falha e a acurácia do modelo, contudo, foram identificadas lacunas metodológicas que merecem atenção em trabalhos futuros. Apesar dessas observações, o estudo reforça o potencial da integração entre a análise avançada de sinais e o aprendizado de máquina para o monitoramento preditivo de manutenção. O caso de uma ocorrência real apresentado, com cabos no propulsor, ilustra a aplicabilidade real da tecnologia. Conclui-se que, embora promissor, o trabalho de Cho et al. destaca a necessidade de maior rigor metodológico para reprodutibilidade e implementação prática de tais inovações.

Machine Learning and Naval Architecture

Rafael Pieper — Tue, 20 May 2025 07:30:00 +0000

Machine Learning (a.k.a AI) for Naval Architecture?

When we think of ship design, we often imagine massive containerships or tankers; But tugboats are a class of their own and as complex to design as another ship. Compact, powerful, and maneuverable, tugboats main mission is to push and pull and salvage large vessels. And as with any vessel, design starts with one thing: requirements.

In the case of tugboats, the most critical parameter that sets the foundation for the design spiral is the bollard pull—the ability to generate pulling force at zero speed, as I’ve written some time ago…
Once this value is determined, naval architects begin the intricate spiral of defining hull dimensions, propulsion power, stability, and arrangements.

As I previously wrote in 2021, the Ship Design Spiral is not a straight path; it’s an iterative journey. As each parameter is refined, other parameters must adjust. For tugboats, designers must balance power, stability, and compactness while ensuring the vessel meets class rules and high power delivery.

A Recent study by Karaçay et al. (2024) published in Applied Sciences proposes a machine learning model to predict the main particulars of diesel-powered Z-Drive harbor tugboats during the concept design stage.
The model uses Bayesian networks and non-linear regression to estimate key dimensions (length, beam, draft) and Power, based on inputs like bollard pull and service speed. The authors compiled a robust dataset of 476 tugboats, achieving interesting results (6.57% MAPE).

This study demonstrates that ML (AI for the buzzword lovers) models can support the early-stage concept design of specialized vessels like tugboats once we have a good historical dataset. The use of a data-driven framework can reduce the number of revisions down the spiral, which traditionally involved CFD simulations or towing tank tests.

What’s exciting is how machine learning can help accelerate early decisions, giving naval architects more room to focus on refinement and performance tuning.

I also developed a simple linear regression model, on june 2023, that predicts Bollard Pull based on Power and Propeller diameter, from Kongsberg and Caterpillar’s z-drive catalog data, check it here.

Full Paper: Karaçay, Ö.E.; Karatuğ, Ç.; Uyanık, T.; Arslanoğlu, Y.; Lashab, A. (2024). Prediction of Ship Main Particulars for Harbor Tugboats Using a Bayesian Network Model and Non-Linear Regression. Applied Sciences, 14(7), 2891.

Time Series in Maritime

Rafael Pieper — Fri, 28 Mar 2025 12:00:51 +0000

How Maritime sector could leverage time series analysis?

The maritime sector is characterized by complex operational dynamics that need rigorous analytical methodologies to optimize efficiency, mitigate risk, and enhance decision-making.

Time series analysis, a branch of predictive analytics, enables maritime stakeholders to understand temporal dependencies, extract structures in data, and deploy robust forecasting techniques to improve resource allocation and operational resilience.

At its core, time series analysis lands in decomposing and modeling sequential data to identify stochastic patterns, structural breaks, and seasonal variations. The ability to extrapolate these patterns facilitates the development of probabilistic models that generate predictive decision frameworks, thereby benefiting strategy in maritime operations.

Strategic Implications of Time Series Forecasting in Maritime Logistics

Maritime transport and fleet management are inherently dynamic and require data-driven methodologies to navigate fluctuating demand, maintenance contingencies, and external market influences. Two applications of time series forecasting highlight its strategic importance:

Freight Volume Forecasting

Problem Statement: Freight volume is subject to various macroeconomic variables, trade policies, and seasonal demand fluctuations, leading to non-stationary demand curves and suboptimal asset deployment.

Analytical Framework: Leveraging historical freight data through parametric and non-parametric modeling approaches, shipping companies can derive predictive insights into future demand distributions, enabling adaptive fleet utilization, optimal routing strategies, and efficient supply chain coordination.
Implementing hierarchical time series models to anticipate peak shipping periods enables dynamic fleet reallocation, reducing logistical bottlenecks and enhancing service reliability.

Advanced Models: Vector Autoregression (VAR), SARIMAX, Holt-Winters

Predictive Maintenance and Condition-Based Monitoring

Problem Statement: Ships operate in highly variable environmental conditions, necessitating continuous monitoring of engine health parameters to mitigate the risks of catastrophic failures and optimize lifecycle maintenance strategies.

Analytical Framework: Time-dependent sensor telemetry—such as vibration signatures, thermal anomalies, and oil viscosity degradation—can be modeled using probabilistic anomaly detection frameworks and multivariate predictive models to schedule maintenance interventions preemptively.

Advanced Models: Autoencoder-Based Anomaly Detection, Random Forests

A simple example is explained here

Those are just a few examples of applications within the maritime sector!

Tugboat allocation

Rafael Pieper — Tue, 21 Jan 2025 17:23:53 +0000

Presentation of an optimization problem for the course at the master’s studies.

Details also on GitHub: https://github.com/RafaPieper/Tugboat-Allocation-Optmization

Fault Detection and Diagnosis – what’s up?

Rafael Pieper — Fri, 15 Nov 2024 19:20:14 +0000

Review on Valesco-Gallegos (2023)

The relationship between ship maintenance and data technology remains foggy, but we hope this will change. While they can work together, it is challenging because new technologies are being adopted slowly—perhaps too slowly—by maritime stakeholders.

I’ve been studying the current state-of-the-art methods of data-driven maintenance applied to ships, and it is interesting how the research is growing in this field. Even class societies are looking at new applications for Fault Detection and Diagnosis Systems (FDD).

Today, I read another paper by Valesco-Gallego et al. (2023). In this paper, the authors conducted a literature review on recent advancements in data-driven methods for FDD, focusing on data preprocessing, fault diagnosis, and prognosis in marine systems from 2016 to 2022.

The idea of applying data-driven methods to maintenance arises because the effectiveness of a Machine Learning algorithm is way higher than that of a human when evaluating a high volume of data and its attributes. The possibility of identifying the initial phase of a failure enables the ship operator to take action, avoiding unexpected breakdowns.

What is being applied for FDD?
Machine Learning methods such as Deep Learning, Artificial Neural Networks (ANNs), and Support Vector Machines (SVMs) are among the most commonly applied techniques in many of the papers reviewed by the authors.

For many researchers, however, the lack of labeled data (fault/no fault or fault classification) remains a significant challenge. As a result, many are turning to anomaly detection and unsupervised learning methods to achieve the primary goal: fault detection.

New equipments are being delivered with high-tech sensors, and with possibilities of data collection. With good data analytics and data science, it give the operator good insights into the state of the equipment and opens an opportunity to apply condition-based maintenance and predictive maintenance, which can improve current maintenance strategies.

The image attached is an overall mind map I made connecting the information from Valesco-Gallego’s paper.

Link to the paper: https://www.sciencedirect.com/science/article/pii/S002980182301661X?via%3Dihub

Research Paper on Brazilian roads accidents

Rafael Pieper — Thu, 03 Oct 2024 22:58:24 +0000

Propostas para tomada de decisão de ações do Plano Nacional de Redução de Mortes e Lesões no Trânsito (PNTRANS) baseado na análise de dados de trânsito dos anos 2020 e 2021 para a região sul do Brasil

Abstract. This article proposes practical solutions for the southern region of Brazil related to 3 actions (A5004, A5006, A6020) of the National Plan for Reducing Traffic Deaths and Injuries (PNATRANS) based on traffic data from the years 2020 and 2021. Using data science techniques, it evaluates locations and segments of federal highways with high accident rates in different years. It was observed locations that stand out in all analyzes, and areas that have higher rates of accident occurrences due to a lack of respect for traffic laws, making it possible to propose resolutions and places of action for PNATRANS actions.

Resumo. Este artigo propõe resoluções práticas para a região sul do Brasil relacionadas a 3 ações (A5004, A5006, A6020) do Plano Nacional de Redução de mortes e lesões no trânsito (PNATRANS) baseado nos dados de trânsito dos anos de 2020 e 2021. Utilizando técnicas de ciência de dados, avaliou-se localidades e trechos de rodovias federais que possuem altos índices de acidentes em anos distintos. Foi observado trechos que se destacam em todas as análises e localidades que tem maiores taxas de ocorrências de acidentes por falta de respeito às leis de trânsito possibilitando propor resoluções e locais de atuação para as ações do PNATRANS.

Machine Learning Basics – Cheat sheet (Portuguese)

Rafael Pieper — Sat, 20 Jan 2024 22:34:04 +0000

Strong Foundation for AI Implementation

Rafael Pieper — Sat, 16 Dec 2023 23:08:00 +0000

The Importance of Data Quality and Business Intelligence

In the current business environment, Artificial Intelligence (AI) is one of the main technological trends. Companies have shown a strong interest in developing internal models and connections with AI technologies, enabling growth and innovation in organizations.

Machine learning (ML) is not new, it has existed since 1950~. However, before moving towards implementing AI, companies must understand the need to have a solid data foundation and have a properly structured Business Intelligence (BI) and data analysis team.

1. Data Structuring and Quality: The First Step Towards AI

AI initiatives depend directly on the quality and availability of data. Before wanting to implement AI models, it is essential that companies carry out careful work of structuring and knowledge of this information.

This involves organizing, cleaning, and standardizing data in an appropriate format for analysis. By establishing a solid structure for their data, companies will be building a robust and reliable pillar to start thinking about the application of AI.

2. It’s Not Just About Tools, GPT Chat, and Colorful Dashboards

Data analysis is not just about using Power BI, Tableau, Excel, and Python, it is possible to achieve the same result in all of them. It is much more about human intelligence and knowledge applied to the interpretation and understanding of data. Although tools facilitate the process, it is the ability to extract relevant insights and make strategic decisions that make the difference.

This requires advanced analytical skills, the ability to identify patterns and trends, as well as knowledge of the business context. Not only knowing how to create a bar chart or avoid the pie chart. It is necessary to know how to formulate the right questions, select the appropriate metrics, and interpret the results to generate value.

3. BI and Analytics Team: Empowering Artificial Intelligence

A qualified and well-structured Business Intelligence team is a fundamental asset for the success of artificial intelligence initiatives. This team is responsible for analyzing data, validating the knowledge, and translating this information into strategic insights for the organization.

These professionals will work side by side with data scientists and engineers, ensuring that the analyses are accurate and relevant, and provide accurate data to start train and test of ML models.

With this well-defined structure, companies will be well-positioned to achieve the benefits of artificial intelligence:

Greater efficiency and accuracy in data analysis and utilization, ensuring the reliability and quality of information before applying AI;
Ability to extract valuable insights, identify emerging trends, and produce a solid and reliable foundation for training and testing AI models;
Simplification of the AI deployment process, reducing risks and optimizing resources, with the potential to develop more advanced and sophisticated models, leveraging the full potential of technology.

If you are considering implementing Artificial Intelligence in your organization, make sure to establish a solid foundation through proper data structuring and a qualified team.

These elements are essential to obtain reliable, relevant, and innovative results through AI. It may even be that you don’t need it, just a well-done analysis can be your goal.

Also on MEDIUM

TPC – Tonnes per Centimeter Immersion

Rafael Pieper — Sun, 01 Oct 2023 23:12:24 +0000

How much weight will cause a ship to sink by 1 centimeter?

The importance of knowing ship’s details and accurate port conditions

TPC stands for “Tons per Centimeter” of immersion, which means it is the amount of cargo in tons needed to sink the ship by 1 cm.

This value is important in the design of a cargo ship, because for all those involved in a maritime transport operation, the more cargo, the better the cost-benefit.

To calculate the TPC (tons per centimeter of immersion), we first need to know the waterline area (Aw) of the ship. Aw is calculated by multiplying the length, breadth, and waterline coefficient (Cw) of the ship. Cw is a coefficient calculated during the ship’s design, influenced by the shape of the hull and hydrodynamics.

With the waterline area in hand, the TPC is then calculated by multiplying Aw by the density of water in t/m³ and dividing by 100 to get the result in centimeters.

That’s it, we now have the TPC to load the ship.

It is essential to note that this value can change depending on the ship’s draft, because the geometry of the waterline plan will change according to the hull shape at each draft.

Each centimeter of draft of a ship is important for transporting more cargo. Similarly, it is indispensable to have precise control of the depths of ports in order to optimize operations.

So, if a ship has a TPC of 45t/cm, that means that it will sink by 1 centimeter if you add 45 tons of cargo to it.

A simple example

When a ship with a box-shaped hull sinks, the TPC (tonnes per centimeter) remains constant.

This is because the water plane area does not change as the ship sinks. Therefore, the weight of cargo required to sink the ship by one unit of draft remains constant for all draft (H) values.

A more complex example

In prismatic-shaped ships, as a ship sinks, its TPC increases due to the increased water plane area.

The TPC is dependent on the draft of the ship (designated as H), meaning that as the ship sinks further, more cargo is required to sink each unit of the draft.

Example of hydrostatic information of a ship and how they change in function of Draft.

Source: Tupper.E – Introduction to Naval Architecture 4th Ed