Oil leakage from cracks is a kind of common fault in mechatronics systems. Responses of artificial olfactory system to the volatile organic compounds (VOCs) emitted by leakage oil can be used for leak detection, which is a novel non-destructive method for fault diagnosis. An artificial olfactory system was applied to detect three different kinds of leakage oil such as diesel oil, machine oil and gear oil. The collected dataset was arranged in a-three dimensionality matrix (sample×time×sensor). Three methods as two-way principal component analysis (PCA), three-way unfolding PCA and parallel factor analysis (PARAFAC) were adopted to distinguish the VOCs. The results showed that PARAFAC was superior to the other two methods because PARAFAC took into account the true three-dimensionality structure of the dataset. The multivariate calibration method, principal component regression (PCR) was applied in the prediction of different concentrations of diesel oil. The results indicate that it is possible to use an artificial olfactory system to give an alarm by setting a threshold. 