Coleção de Artigos Acadêmicos
URI permanente para esta coleçãohttps://repositorio.insper.edu.br/handle/11224/3227
Navegar
2 resultados
Resultados da Pesquisa
Artigo Científico An Automated Electronic System in a Motorized Wheelchair for Telemonitoring: Mixed Methods Study Based on Internet of Things(2023) Gradim, Luma Carolina Câmara; Santana, André Luiz Maciel; José, Marcelo Archanjo; Zuffo, Marcelo Knörich; Lopes, Roseli de DeusBackground: Wheelchair positioning systems can prevent postural deficits and pressure injuries. However, a more effective professional follow-up is needed to assess and monitor positioning according to the specificities and clinical conditions of each user. Objective: This study aims to present the concept of an electronic system embedded in a motorized wheelchair, based on the Internet of Things (IoT), for automated positioning as part of a study on wheelchairs and telemonitoring. Methods: We conducted a mixed methods study with a user-centered design approach, interviews with 16 wheelchair users and 66 professionals for the development of system functions, and a formative assessment of 5 participants with descriptive analysis to design system concepts. Results: We presented a new wheelchair system with hardware and software components developed based on coparticipation with singular components in an IoT architecture. In an IoT solution, the incorporation of sensors from the inertial measurement unit was crucial. These sensors were vital for offering alternative methods to monitor and control the tilt and recline functions of a wheelchair. This monitoring and control could be achieved autonomously through a smartphone app. In addition, this capability addressed the requirements of real users. Conclusions: The technologies presented in this system can benefit telemonitoring and favor real feedback, allowing quality provision of health services to wheelchair users. User-centered development favored development with specific functions to meet the real demands of users. We emphasize the importance of future studies on the correlation between diagnoses and the use of the system in a real environment to help professionals in treatment.Artigo Científico Análise morfológica de nanofibras: uma abordagem por visão computacional e aprendizagem de máquina(2021) Barros, Guilherme Duarte de; Santana, André Luiz Maciel; Pereira, Vitor Matheus Ferraz; Pereira, Wesley da Silva; Santana, André Luiz MacielThe studies and applications of nanofibers have grown over the years. It was observed that the properties of the nanometer-scale yarns present advantages in applications in several areas, such as biomedical, energy storage and production, and applications involving water filtration. These materials are synthesized through a technical process and, for that reason, they are subject to the presentation of failures. The most common flaws are the formation of granules and pores. With the evolution of computing, applications that use machine learning resources can assist in detecting these failures. This work aims to evaluate and compare two different approaches to morphological analysis to see losses in nanofibers. Firstly, a data set was created using a Scanning Electron Microscope. After that, each image was analyzed by ImageJ software and by RNA solution. As a hypothesis, the article will assess whether the beads identification and the number of beads by the analog method are statistically similar (H0) or statistically different (H1) from the machine learning method. The preliminary results indicate that for the group that used 100 images and computer visualization, the analog method's accuracy was 7.23%. In order to accuracy increase, another test with 150 more distinct images was done, bringing a new result of 55.09%. The analysis time was considerably less when performed by the computational method. It was possible to conclude that the computational approach does not have the beads identification statistically similar to the analog way concerning the methodology used. Therefore, rejected H0. However, the directly proportional relationship of accuracy with the number of samples suggests that training with more various images can calibrate the algorithm.