Implementación de un modelo predictivo basado en redes neuronales recurrentes en una aplicación para gestionar mantenimiento predictivo de equipos de gases arteriales disponibles en unidades hospitalarias

Abstract

The medical device market has grown rapidly in recent years. As technological trends in the healthcare industry focus on improving human well-being, it is currently unknown exactly how many different types of medical devices are used and exist worldwide. However, in the Region of the Americas, emerging medical device markets are highly significant; with few exceptions, countries import more than 80% of their medical devices. Given this situation, it is essential for countries to prioritize patient safety and ensure access to high-quality, safe, and effective medical devices [1]. In this context, it becomes evident that reducing the high downtime of biomedical equipment in hospital units—caused by failures or maintenance—is crucial. These interruptions often lead medical personnel to make decisions based on patient symptoms and personal experience, rather than on reliable laboratory test results, increasing the likelihood of misdiagnosis.Therefore, this project proposes predicting potential failures in arterial blood gas biomedical equipment through the use of ICTs, specifically by implementing Recurrent Neural Networks (RNN) trained on data such as date, device serial number, type of service, city, failure, and solution—recorded by expert engineers through a standardized form. This document presents a methodological design for the proposal to be developed over a 12-month period, with an estimated cost of $9,662,000, under the monograph modality and aligned with research areas in industrial automation and ICT management. The project will be implemented at the Faculty of Technology of Universidad Distrital Francisco José de Caldas and directly supports the healthcare and ICT sectors.