Internal combustion engine valve clearance fault classification using multivariate analysis of variance and discriminant analysis

This paper investigates the classification of a valve clearance fault in an internal combustion diesel engine using vibration time domain features extracted from signal segments measured at several points on the engine bloc. Signals containing a large number of engine cycles are used to obtain a number of observations of each feature. The set of features is thus considered a set of variables. A stepwise variable selection algorithm based on univariate and multivariate analysis of variance is then used to sort the variables according to their diagnostic ability. The algorithm is also used to construct sets of variables of increasing size used to improve fault classification. Four commonly used supervised classifiers are trained and then tested, giving roughly the same percentage of correct classification. The tested classifiers confirmed that the use of more variables selected by the stepwise variable selection algorithm increases the percentage of correct classification.