The increasing use of technologies such as computers and the internet has revolutionized the way we interact, work, entertain ourselves, and manage our health. In this context, the Internet of Things (IoT) emerges as an innovation that connects a variety of devices and systems, aiming to enhance the daily experiences of users. Covering everything from lighting systems to appliances and entertainment solutions, IoT allows users to manage various aspects of their environment in an integrated and intelligent manner, using everything from smartphone apps to voice commands through virtual assistants.

This project focuses on developing an IoT system for monitoring and detecting faults in thermal systems, combining electronics, microcontrollers, and sensors with machine learning methods. The goal is to integrate these technologies to create a device that monitors thermal conditions in real time while identifying abnormal patterns that may indicate potential failures. This predictive approach to the maintenance of heating and cooling systems aims to prevent disruptions before they occur, enhancing system efficiency and reducing operational and maintenance costs.

At the conclusion of this project, experimental results will be compiled and presented, demonstrating the effectiveness of the proposed system. It is expected that the developed device will provide detailed information on the performance and operational conditions of the thermal systems. Furthermore, the implementation of this IoT monitoring system promises to enhance security and energy efficiency, also providing users with a robust tool for remote control and management of their environments, aligning technology and practicality in everyday life.

Systems that originate from second-degree differential equations, or simply second-order systems, are highly relevant in the study of control systems, due to their broad power to model practical situations, such as mechanical vibrations, resonance, and even specific processes in areas such as fermentation in certain biological processes. Working with this type of model, instead of first-order state models, can bring some benefits, such as the possibility of developing more sophisticated controls and a clearer frequency analysis. In practice, these systems suffer from delays inherent in several parts, such as actuators and sensor measurements, which can generate harmful effects, such as reduced performance, instability, and oscillations. Considering that even small delays can generate significant disturbances in the system, it is valid to use an approach based on the system's receptances, which can accurately represent the delay through a study in the frequency domain. This dissertation aims to develop multivariable PID controllers for linear dynamic systems with multiple inputs and outputs (MIMO) that present delay in actuation and are described by a system of second-order differential equations. The proposed technique uses the frequency domain model known as the receptance matrix, which can be obtained accurately experimentally and which allows the delay effect to be treated without approximations. Closed-loop stability is ensured using the so-called generalized Nyquist criterion, verified by the so-called eigenloci diagrams. A given stability margin is obtained by imposing a minimum distance between the eigenloci and the critical point for stability. An optimization problem is formulated to calculate the PID controller gains that impose this margin and minimize a time response performance index, which is solved by Genetic Algorithm. Different delays for each actuator and different PID controller schemes are considered. Numerical examples illustrate the proposed approach.

This work presents the development and improvement of a prototype smart walker designed to assist people with reduced mobility during physical therapy rehabilitation. The device is based on an adaptation of a conventional walker, integrating geared motors, an Arduino Mega microcontroller, a BeagleBone Blue microcomputer, and sensors such as incremental encoders on the wheels, an inertial measurement unit (IMU) composed of an accelerometer, gyroscope, and magnetometer, and a Kinect RGB-D camera. The main focus of the research is to obtain reliable estimates of the walker’s position and orientation through odometry, its calibration, and sensor fusion. A gyroscope and a magnetometer are employed in this study, although the approach is applicable to other complementary sensors. Using motion and observation models, the Extended Information Filter (EIF) was implemented, allowing for efficient integration of sensor data and reducing the effects of noise and uncertainty, representing a promising framework for future comprehensive localization. The adopted approach demonstrated that, even without external sources of absolute location, it is possible to accurately estimate the walker’s trajectory and orientation, provided the sensors are calibrated and properly combined via EIF. This strategy, which combines odometry with sensor fusion, represents an important step in the development of a robust, reliable localization system applicable to real-world assistive scenarios.

This dissertation presents the development of an automated system for generating personalized wrist orthoses. The innovative approach utilizes a single photograph of a patient's upper limb to create a custom orthosis model. This model is then fabricated using a FDM 3D printer with thermoformable PLA, allowing for direct application to the patient without requiring an orthotist's intervention or the patient's physical travel.

The core motivation for this project stems from the significant challenges in rehabilitating patients with neuromuscular diseases, injuries, or degenerative conditions like Amyotrophic Lateral Sclerosis (ALS). These conditions often involve constant anatomical and functional changes, demanding orthoses that can be rapidly adjusted. Traditional manufacturing methods are manual, time-consuming, and expensive, frequently requiring the patient's physical presence at multiple stages. This poses a considerable barrier to access for bedridden patients or those in remote locations. Furthermore, standardized commercial orthoses rarely offer an ideal fit.

This project is directly inspired by the revELA Project, a collaborative effort between the Laboratory of Technological Innovation in Health (LAIS/UFRN) and the Ministry of Health. The revELA Project aims to develop technological solutions for patients with ALS and other neurodegenerative diseases, specifically by leveraging computer vision, 3D printing, and artificial intelligence to create rapid, personalized, and accessible orthoses.

The world is undergoing climate changes that were predicted decades ago, stemming from the uncontrolled exploitation of natural resources, with processes that degraded fauna and flora, often in areas of high biodiversity. The lack of environmental conservation and awareness of the impacts of these exploitations on life preservation on Earth has led to the environmental disasters and crises we witness today. The environment serves as a natural absorber of greenhouse gases and supports ecological balance. Without preserving these biodiversity-rich environments, which act as natural regulators of environmental balance, the conservation of life becomes unsustainable alongside high pollutant emissions and population growth. Natural reserves are being consumed unsustainably, leading humanity into a difficult situation regarding the maintenance of life itself.

In this work, a radio telemetry application was developed to reduce costs and assist researchers, biologists, and wildlife conservation foundations in tracking wild animals that are critical to environmental control and balance but are at risk of extinction. This solution integrates both hardware and software, using a geolocator (GPS), sensors (temperature, immersion), data storage (in both the tracker and receiver), and LoRa radio protocol communication to locate and download data collected and stored by the tracker. The application aims to provide a means for wildlife research and conservation, feeding Brazilian environmental foundations and research with a lower cost in the import of goods and services.

The coordination of the electricity sector's operating activities is carried out by a set of tools and technologies, the most notable of which is audio communication between real-time teams of agents and technicians in the field. These calls are recorded, stored, and consulted on demand for auditing and traceability purposes, as well as to support postoperative analyses. However, because of the difficulty in accessing their content, these recordings, which contain relevant information for both system operation and agent billing, are underutilized. Thus, by transcribing them, it is possible to trace and associate these audios with each other and with other records from different operating systems, allowing for the optimization of activities that require a significant amount of time from post-operation engineers. This dissertation describes a virtual assistant for energy generators, transmitters, and distributors' operation centers that transcribes all audio communications from these centers and uses natural language processing techniques to extract information from the audios, classify them into topics of interest to the operation, and correlate the recordings with each other and with operating systems. The system architecture is presented, as well as the audio-to-text transcription models used, the classification module, and the correlation between audios and with SCADA module. The project described here is intended to aid decision-making and contribute to greater efficiency, safety, and operational reliability in electrical system operation and post-operation.

Faced with the constant technological evolutions that modern society experiences, industries have evolved more and more in search of greater efficiency and reliability of their processes. The search for new methods to optimize the industrial activities has fueled the interest in the development of research focused on this area. Despite the various technological developments that industries constantly undergo, steam remains present in their processes since the first industrial revolution, being used to perform movements in machines, which revolutionized the production process of the time. Nowadays its presence remains vast in different types of industry, being widely used for heating and electricity generation. The equipment used for steam generation is the boiler. In the process of generating steam in a boiler, there are two variables of great importance for control, which are the water level and the steam pressure in the drum. The control of these variables guarantees the operational safety of the machine and compliance with the process parameters through the required steam pressure. Thus, the boiler control problem is of great interest in academia and industry. In order to control the variables reported, it is basically necessary to act on the water supply and saturated steam exit valves of the drum. Based on this, this work aims to control the water level and pressure in the drum of a water-tube boiler through artificial intelligence in order to avoid unwanted variations in level and pressure when going through load variations. For this, fuzzy controllers optimized by genetic algorithm are used. Tests were performed for the model and their results were satisfactory.

Phenomena such as mechanical vibrations, resonance and oscillations can be mathematically described by second-order differential equation systems, commonly referred to as second-order systems. Working with this type of model, instead of first-order state models, brings numerical benefits, but there are inherent difficulties in determining their physical parameters. The challenges become even more significant when considering the existence of delays between state measurements and actuation signals, leading some approaches to require post-analysis for determining the stability of calculated solutions. An alternative to overcoming the difficulties of parameter measurement is the frequency response approach which utilizes models based on receptance.  In view of this issue, this work focuses on the design of PID controllers (Proportional-Integral-Derivative) for linear dynamic systems with delay, modeled by second-order matrix differential equations. It adopts the receptance approach because it is based on the frequency response of the system, enabling precise treatment of closed-loop stability, without the need for approximations or back-testing of delay terms. To ensure robustness and performance, as well as minimize the Absolute Error Integral relative to the tracking of a constant reference, it formules an optimization problem for the determination of controller gains. Also, it takes into account a pre-established stability margin. To solve the optimization problem, it implements a Genetic Algorithm. Unlike related works in the literature, the proposed method can be equally applied to systems with open-loop poles in the right half-plane.

This work proposes a method for designing controllers by state-derivative feedback of linear systems represented by second-order matrix differential equations with delay in the control signal. From the frequency domain representation, known as the receptance model, an optimization problem is formulated for the computation of controller gains that limit the frequency peak of a sensitivity function, thus ensuring robust stability in closed-loop even in face of delay. Genetic Algorithm is used to solve the optimization problem. To deal with systems with a singular mass matrix, a structure with a filtered Smith predictor is proposed, which guarantees the regularization of the system and the elimination of impulsive dynamics for consistent initial conditions. Numerical experiments illustrate the effectiveness of the proposed method, including the elimination of impulsive dynamics, and bring comparisons with state feedback controllers.

Type 1 Diabetes Mellitus is a disease that affects millions of people around the world. Recently, the incredible progress of embedded devices has given rise to proposals of devices that pump insulin subcutaneously, with the purpose of automatically regulating blood glucose level in diabetic patients. This way, the Artificial Pancreas could provide a better quality of life with more autonomy and comfort to the patients. The goal of this work is to design a nonlinear intelligent controller with a radial basis function (RBF) artificial neural network as an uncertainty estimator for an artificial pancreas using the IVP (Identifiable Virtual Patient) model for blood glucose regulation to simulate the dynamics of the virtual patient. The virtual patients and meals are randomly generated following normal distributions and the parameters of the patients vary in a sinusoidal way over the course of the simulation. The proposed control approach neither has knowledge of the system dynamics nor is alerted when a patient has a meal. The first controller analyzed was based on the feedback linearization (FBL) technique with a RBF estimator and a projection algorithm, and the second one was based on sliding modes control with a RBF estimator. On the first part of the tests, 200 virtual patients underwent a 7-day, 3-meal per day simulation. The controllers had equivalent performances with worst case scenario resulting in 115,97 mg/dL mean blood glucose and 97,14% of the time in normoglycemic regime. On the second part, 1 patient underwent a 63-day, 3-meal per day simulation with the goal of analyzing the long-term behavior of the controllers. In the worst case scenario, the simulations resulted in 119,20 mg/dL mean blood glucose and 93,67% of the time in normoglycemia. On this part, FBL technique showed better perfomance, suggesting that, in the long run, the projection algorithm provides greater stability in the update of the neural network weight vector. The results indicate that, due to its continuous learning and adaptation abilities, the proposed intelligent controller has proven to be fit for the problem of efficient blood glucose regulation in patients with type 1 diabetes mellitus.

In recent years, many studies have been carried out on the topic of exoskeletons related to locomotion assistance. The main goal of these devices is to assist the elderly and physically challenged persons in daily activities, replacing or increasing the movement of body articulations. Although significant progress has been achieved, many challenges still remain. A desirable feature for exoskeletons is the planning of autonomous movements, so that the movements are automatically adapted according to the environment that the user is facing. Therefore, it is indispensable the use of a computer vision system to provide information about the environment where the user is located, classifying the structures of the scene as walkable or not. In this sense, we propose a new strategy for ramp and step detection which are in accordance with technical standards and can be climbed for exoskeleton users. Initially, we use a RGB-D sensor to acquire depth information of the environment. Next, we perform plane segmentation using a hypothesis and verification methodology. Through the plane normal analysis, it is possible to establish ramps and steps candidates. Finally, plane dimensions are verified in order to decide whether the ramps and steps are qualified to be climbed. Results were obtained applying the method considering different environments, where it was possible to detect and model ramps and steps satisfactorily for the presented scenarios.

We propose the design and development of an autonomous sailboat, the N-Boat II, which is being built by the associated laboratories Natalnet (UFRN) and LAICA (IFRN). We introduce the electromechanical design, from design and construction to the control phase. Its great differential is distributed hardware by affinity along isolated watertight compartments. With the premise of being a prototype capable of replication and have complete open source, this robotic model will be used, in principle, to monitor water quality. However, this platform can be used in numerous tasks of various natures. Thus, it is expected at the end of this work, provide the community a complete infrastructure, with potential to be used scientifically, commercially or militarily, in proposals that require a robotic water surface vehicle with sufficient autonomy for medium and long missions.

In this work, it is proposed the creation of a computer vision system for an active
orthosis with the objetive of guaranteeing greater autonomy in its movements. This equipments
are generally present means of understanding the work environment, which makes
difficult the Orthosis to organize such environments and predict actions of their users,
for example. In this context, a strategy is presented which, through an RGB-D camera,
obtain a point cloud scenery to detect planar regions of the environment. Starting from
the principle that most walking sites are flat regions, we can classify regions transposable
to the base with no set of points that can be classified as ground, for example. To verify
the quantify and refinery are an classification for an update, is necessary that has been
the testing and filtering with specific date of data to processing. The preliminary results
were obtained in simulation environments and in real environment, to which the indicators
were satisfactory with the considerable reduction of the set of points, but keeping the
shape of the plane close to the original.

In NGPU (Natural Gas Processing Units), one of the most profitable products is LPG (Liquefied Petroleum Gas) which is composed mostly of propane (C3) and butane (C4). In addition, pentane (C5) and ethane (C2) may also be present in the gas as contaminants. Measurement of the molar fraction of the components of the GLP has been done by gas chromatographs. However, chromatography is a slow process, making it impossible to monitor the quality of the product in real time. In this work, it is proposed a soft sensor, using artificial neural networks, with the objective of inferring the molar fractions of C3, C4, C5 and C2. In this way, it would be possible to estimate the molar fractions of these LPG components once per minute, considerably faster than chromatographers. The results obtained are promising, showing that the virtual sensor can infer the molar fractions of C3, C4, C5 and C2 and thus enable improvement in the monitoring of LPG quality and, consequently, profitability.

With advances in nanosatellite research, as an alternative to space exploration, the Instituto Nacional de Pesquisas Espaciais (INPE) is interested in using this new nanosatellite technology in its future missions. This makes it necessary to have a platform on which to study nanosatellite behavior in space. In this context, the following thesis implemented a simulator, composed with a navigation system and a proportional and derivative controller (PD). The sensors used to estimate the attitude were composed of a magnetometer and a solar sensor, simulated by using an ideal model with white noise. To represent the kinematics, the quaternion was adopted and the nanosatellite was a rigid body. In the navigation system, the norm-constrained Extended Kalman Filter had shown itself to be adequate to work with the sensors embedded in the nanosatellite. Thus allowing the estimate of the angular velocity and the quaternion that represents the attitude. For the controller, the values estimated by the Filter were directly used as a control signal, which has the advantage of dealing with the nonlinearities. Different tests were performed to assess the simulator and evaluate the stability of the navigation and control system. The results have demonstrated that the set up is able to stabilize and correct the attitude of the nanosatellite, even in the face of different perturbations, large angles and high angular velocities. Hence, the simulator generates adequate results, allowing it to be used in future nanosatellite studies.

The ability to follow or move along with a specified person or object, capable of move, is a necessary skill in several autonomous agents to perform various tasks present in everyday life, and can be applied either in everyday tasks, as in supermarket carts or cleaning environments, as well in high-risk tasks such as in large industries or autonomous cars. The idea presented here is to develop a target tracking and following method applicable to mobile wheeled land robots that have restrictions on their movement, which mean that standard control techniques can not always be applied. The work developed here also takes into account the use of a target detection technique that can be adapted to practically any type of target stipulated by the designer according to the needs of its application. The development of the proposed methods was accomplished by adding standard recognition techniques, used in common RGB cameras, position estimation and orientation techniques, and intelligent control algorithms applicable to robots with movement restrictions, which have a low computational cost.

Several devices are used in artificial vision, like MS Kinect v1/v2, the stereo cameras PG Bumblebee XB3 and Stereolabs ZED, among others. Since they are all devices that estimate depth data, they may have errors. In this work, we present the design and implementation of a generic method for estimating RMS error in depth data provided by any device capable of generating data of type RGB-D, that is, an image and the depth map, to the same time. To verify the method, an embedded system was built based on the NVIDIA Jetson TK1 and three sensors, the two versions of the MS Kinect and the ZED stereo camera. At the moment of data collection, the mathematical models of the RMS error were established for each device and, at the end, an analysis was made of the accuracy of each one. It is intended that the proposed method be the basis for other future works, such as p. ex. achieve a 3D reconstruction using the mix of depth data captured by various 3D sensors, with the lowest possible RMS error. In addition, it is sought that the error information obtained through the method can be used in any application to obtain more accurate results.

Com o surgimento de técnicas de controle que utilizam estratégias não lineares combinadas com algoritmos da inteligência artificial, tem sido possível em diversas área da engenharia o controle eficaz de sistemas não lineares, mesmo na presença de elevado grau de incertezas. A lógica difusa (fuzzy) se destaca dentre as técnicas de inteligência artificial tanto pela facilidade de sua implementação quanto pela semelhança entre o seu processo de inferência e o raciocínio humano. Outra vantagem reside no fato da lógica difusa não necessitar de conhecimento prévio do modelo do sistema, quando aplicada ao controle de sistemas dinâmicos. No que tange às estratégias de controle não linear, a principal limitação da técnica de linearização por realimentação, por exemplo, está na necessidade do conhecimento do modelo do sistema. Os sistemas eletro-hidráulicos, por sua vez, possuem modelo não linear de difícil controle pelas técnicas tradicionais e são utilizados em diversas áreas da engenharia, como por exemplo nos setores industrial e aeroespacial. Desta forma, mostra-se extremamente importante que seu controle seja realizado de maneira eficiente, tanto por questões de economia quanto de segurança. Partindo das dificuldades apresentadas ao tentar controlar esse tipo de sistema, são apresentadas propostas de utilização da técnica de controle de linearização por realimentação em conjunto com a lógica difusa para compensar a não linearidade de zona morta e demais incertezas inerentes a este tipo de sistema. Para avaliar o desempenho das estratégias de controle propostas são realizadas simulações numéricas utilizando um modelo não linear simplificado desse sistema e também são desenvolvidos testes experimentais em um atuador eletro-hidráulico de bancada.

Propomos um sistema mecatrônico completo para monitoramento da qualidade da água em rios, lagoas (e represas) e no mar, capaz de realizar a coleta, processamento e apresentação dos dados na web, em tempo real, com o intuito de facilitar a sua análise, de forma rápida e precisa pelas comunidades interessadas. A título de aplicação, o sistema está sendo embarcado em um veleiro robótico autônomo, que será a plataforma responsável por coletar os dados em vários pontos pré-definidos em um sistema de planejamento de navegação. Este projeto une as vantagens da autonomia de um veleiro robótico com a necessidade de monitoramento rápido e preciso da qualidade da água. Introduzimos a arquitetura de hardware e software de um sistema de monitoramento genérico que dá suporte a diferente aplicações, focando sua utilização em um veleiro robótico autônomo não tripulado, facilitando o desenvolvimento de pesquisas nessa área.