Estudio de factibilidad para la implementación de un sistema de bombeo para la comunidad Airón Cruz ubicada en el cantón Guamote de la parroquia Cebadas provincia de Chimborazo

Abstract

The objective of this work was to compare machine learning models for the detection of failures in double flow turbojets. The applied solution in the prediction was performed by analyzing the data, obtained in the measurements of the equipment whose machine learning is a technology that predicts the results based on a model by training it with data and analyzing its output behavior. In order to predict machine failures before they occur; then a comparative study of the existing set of machine learning algorithms to announce failures and the remaining useful life of the engine was performed. The models were built based on turbofan engine data sets extracted from the NASA Prediction Center of Excellence Forecasting Repository due to its three operational configurations and 21 sensor columns. In a training set a model was built and verified with a test data set, for this purpose 5 algorithms were selected: Random Forest, XGBoost, Gradient Boosting, SVM and a hybrid model created as a result of the union of the XGBoost and Gradient Boosting models. The results obtained were compared with real data to verify the accuracy of each algorithm, resulting in the Random Forest algorithm as the best model executed with normal and optimized parameters with an f1-score of 99.949% and 99.99% respectively. Finally, it was found that in the database it is not possible to perform feature extraction using machine learning due to its peculiarity in the operating conditions. It is also important to mention that the SVM model was not run with hyperparameters. It is advisable to use deep learning comparison methods because of their accuracy when classifying the data, drastically reducing the computational load when running the model.