Diagnóstico de falhas de ignição em motor diesel marítimo utilizando sinais de vibração e aprendizado de máquinas

Abstract

This work develops a method for misfire diagnosis in a Diesel engine using Machine Learning (ML) techniques and vibration signals. In addition, it performs the study of the minimum number of accelerometers necessary to allow the AM algorithms to maintain a good performance in misfire diagnosis. The database used in this study is composed of vibration signals obtained through 15 accelerometers installed in the engine. Three AM algorithms were used in this work: K-Nearest Neighbors (K-NN), Random Forest (RF) and Support Vector Machines (SVM). The results showed that these algorithms were able to obtain an F1-Score of 99.87%, 99.87% and 100%, respectively, for detection and identification of misfires. Regarding the study of the minimum number of sensors, two methods were used: Relief-F and general combinatorial analysis, with the second being able to show that, with two accelerometers, it was possible to achieve a misfire diagnosis performance with F1-Score above 95%, which is the minimum required, proving the efficiency of the methodology used in this study.