Feature Selection for Diesel Engine Fault Classification

This paper investigates the supervised classification of an injection fault of an internal combustion Diesel engine using vibration measurement. The S-transform is used to produce a time-frequency representation of the vibration signal. The matrix representation of the time frequency image is then reduced to a lower size matrix using a two-dimensional non-negative matrix factorization. Four algorithms are tested for feature selection from this reduced size matrix and the features are sorted according to their ability in fault classification. A Neural Network classifier is then trained and applied to classify test data. The performances of the four considered selection methods are then evaluated by comparing their percentage of correct classification and the computer execution time. It has been found that the performance of the classifier is enhanced when the number of retained feature is increased for the four investigated selection methods.