Jasinski M., Radkowski S.

One of method of defects evolution early stages diagnostic and prediction, is the correlation of the objective technical condition with the diagnostic parameter received from the vibroacoustic signal. Principal Components Analysis (PCA) offers an approach for linear transformation of the problem variables so that the redundant information is reduced and the diagnostic model is more easily extracted. When empirically identified models are used for fault detection, differences in the input excitation between the identification data and the test data can cause difficulties. For processes with significant autocorrelation, the conventional PCA method may not be effective. As the statistical basis of the conventional PCA method is lost due to the violation of the time independence assumption, misleading results such as excessive false alarms may be generated, especially for small size disturbances. For use with dynamic systems, Dynamic Principle Component Analysis (DPCA) extends traditional PCA by creating a data vector which consists of values of the inputs and outputs over a window of time. This paper compare advantages and disadvantages using of the DPCA and traditional PCA to the early stages faults detection and to reduce differences in the input excitation between the identification data and the test data. The object chosen for the investigation presented in this paper is a gearbox. To conclude, the method presented here avoids the need for performing analytical model which are time consuming and costly. It is possible to estimate the type and stage of defect to any generated vibroacoustic signal from the objective parameters of the specially prepared signal without performing simulation models.