Prognosis of Bearing Failure Based on Health State Estimation

Multi-step support vector regression and optimally parameterized wavelet packet transform for machine residual life prediction

“…Based on the position of the test query (represented as 'star') the training data are selected, i.e., 'res1 ¼ ', 'res2 ¼ #'. In the results section, it is validated that the selection of the training data points which are in the vicinity of the test point, enhances the prediction accuracy contrary to the techniques (Sotiris and Pecht, 2007;Kim et al, 2009) based on building the prediction model by using the training data points from all the severity levels. The prediction model is built using the two selected training data which have feature values closest to the test data point with unknown residual life.…”

“…In the first step, the two training datasets which are in the vicinity of the test query (test data-point with unknown residual life), are selected. In the second step, the SVR based prediction model is built with the selected training datasets, instead of building the prediction model with all the training data (Sotiris and Pecht, 2007;Kim et al, 2009). This enhances the overall prediction accuracy because defect propagation or RMS based feature values follow a nonlinear curve with respect to time as described by Paris's law (Pugno et al, 2006).…”

“…This subsection presents performance evaluation of the proposed technique in comparison with 1-step ahead prediction (Shao and Nezu, 2000;Berenji and Yan, 2006;Tran et al, 2008;Fagang Zhao et al, 2009;Samanta and Nataraj, 2009) and SVR (Sotiris and Pecht, 2007;Kim et al, 2009) techniques. Table 6a gives the performance evaluation for the actual vibration datasets and Table 6b presents results for the simulated vibration datasets.…”

“…Moreover, the cracks in the rotary machine components such as bearings propagate in a nonlinear way with respect to time, from inception to breakdown, as described in Paris's fatigue model (Pugno et al, 2006). The third contribution of RWMS is to develop a novel multi-step support vector regression based prediction model which incorporates the impact of nonlinearity in defect propagation and, thereby, enhances the residual life prediction accuracy of the equipment as compared to the techniques based on simple support vector regression (SVR) (Sotiris and Pecht, 2007;Kim et al, 2009). The paper is organized as follows: Section 2 presents the framework for the proposed technique; Section 3 illustrates the data acquisition and parameter optimization; Section 4 contains the results and validates the performance of the proposed predictive maintenance model in comparison with the existing scheme; and Section 5 presents the concluding remarks.…”

Βελτιώσεις Στην Προληπτική Συντήρηση Μηχανολογικού Εξοπλισμού

“…Παράλληλα με τις αριθμητικές μεθόδους γίνεται εφαρμογή και τεχνητής νοημοσύνης (artificial intelligence). Οι πιο διαδεδομένες μέθοδοι αφορούν στην χρήση: Τεχνητών νευρωνικών δικτύων (artificial neural networks -ANN) [109], [110] Support vector machines -SVM [111], [112].…”