Browsing by Author "Santos, Diogo Alexandre Vasconcelos"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Sistema Inteligente de Manutenção PreditivaPublication . Santos, Diogo Alexandre Vasconcelos; Marreiros, Maria Goreti CarvalhoMaintenance Tasks in a shopfloor are one of the most critical tasks regarding the direct effect on production costs and, consequently, profit. Up until now, maintenance tasks were based on both Run-To-Failure and Reactive paradigms, fixing a machine only when it breaks or at a regular time intervals, regardless of the assets needed the maintenance or not. However, with the Industry 4.0 Paradigm and the Smart Factories concept, machines are now equipped with sensors that monitor a large number of different and varied variables which are afterwards stored. This data can be used to predict machine failures, called Predictive Maintenance, with the aid of the manual registries of asset breakdowns. This project, carried out in the scope of the subject TMDEI of the Master in Informatics Engineering (MEI), aims to conceive and build a system capable of doing Predictive Maintenance, by combining sensors and manual inputted data on ERP systems. PrediMain employs different Machine Learning techniques, with a special emphasis on Ensemble Methods, making the generated machine learning models more robust and accurate, by not using a single algorithm for the predictions. For sensor predictions, before classifying them as failure or not, PrediMain uses the auto-ARIMA technique, being an autoparemetrized method generating more accurate predictions. In the end, the system correctly classifies a set of observations with an estimated 90% of accuracy. This system is also developed to be served as a Software-as-a-Service, allowing multiple Data Sources, and therefore shopfloors, to use the same software instance, consequently not compromising the performance of the existing systems.