Increasing customer demands in product quality and increasing complexity in production plants found the need of profound system knowledge to reduce downtimes, to extend the lifetime of components, and to reduce the deterioration of quality. In this contribution, the application of Wavelet Transform and Empirical Mode Decomposition to selected industrial data for fault identification is presented. The data originate from a seven-stand hot strip mill. The suitability of these methods to detect characteristic features to allocate the patterns to specific faults is shortly reviewd. Classification is done with a Support Vector Machine. The results of Empirical Mode Decomposition respectively the Intrinsic Mode Functions are classified by evaluating the results of a cross-correlation. In previous contributions, the authors analyzed the aptitude of selected data-driven methods to detect conspicuous behavior of production plant in the field of non-stationary signals [1]. Based on these previous results, the authors broaden in this contribution the statistical basis and thus improve the reliability of the interpretation by applying the method of four-fold cross-validation to the data. The results of these classifications are evaluated with the McNemar-Test. An approach for fault prediction is discussed.
doi: 10.12783/SHM2015/87