Aplicação da regressão logística binária para manutenção preditiva em máquinas de ressonância magnética

Abstract

This study focuses on the application of logistic regression as a machine learning technique to improve the maintenance of General Electric GE 1.5 T Magnetic Resonance Imaging machines. Given the context of machine reliability and continuity associated with equipment failures, the need to optimize maintenance and minimize unexpected interruptions is crucial to ensure operational efficiency and customer satisfaction. The objectives of the study were to identify tools and theoretical references related to binary logistic regression and its methods, to collect data from temperature, pressure, and flow sensors through a datalogger installed on the machine, and to validate the capacity of the logistic regression model in the utilized database. To achieve these objectives, in addition to the data collection, Microsoft Excel was used for constructing the database, and the software Orange was utilized for the application of the regression model. For model validation, metric analysis such as the ROC curve and the confusion matrix were employed, also comparing with other prediction methods. The results demonstrated the efficacy of logistic regression in predicting potential failures, highlighting its potential as a tool for supporting predictive maintenance. The successful analysis of the test data confirmed the model’s ability to identify the operational status of the Magnetic Resonance Imaging machine, thus contributing to the optimization of maintenance management. The predictive capacity of this approach provides support for data-driven decision-making and contributes to the efficiency and reliability of Magnetic Resonance Imaging services.