MISFIRE DETECTION ON INTERNAL COMBUSTION ENGINE THRORUGH DIFFERENT TYPES OF MACHINE LEARNING MODELS

Resumo

Misfire is a phenomenon that can jeopardize the good yield of an engines function, this might result in
a lower efficiency than expected and an increase on pollution produced by elevated gas emission. However,
there are certain systems that are capable of detecting this type of flaw. This article presents a model based on
artificial intelligence that is capable to spot the misfire caused by a malfunction of the spark plug in a 2006
Zetec-Rocan ford motor that consists in a 4 stroke engine of internal combustion. As a result of its usage,
identifying the main source of the problem becomes an easier task and therefore reducing the time spent on its
maintenance. The model mentioned above uses vibration signals generated by an accelerometer. This signals
went through a pre-processing proceedure to extract features in which were used a multiscreen analysis and FFT
(Fast Fourier transform). After the extraction, those features were used as an entry on machine learning models
that allow them to be classified according to its signal so we are able to identify if theres a defect and where the
problem is located. The 5 machicne learning techniques used were Random Forest, Forest Tree, SVM(Support
Vector Machine), KNN(K-Nearest Neighbors) and Neural Network. The results showed that they were all
accurate both in train and in tests. External data validation also showed solid performance.