Efficacy of Some Primary Discriminant Functions in Diagnosing Planetary Gearboxes
We consider the efficiency of some primary discriminant functions applied in planetary gearbox diagnostics. Real data for planetary gearboxes mounted in bucket wheel excavators working in field condition are elaborated. The aim is to perform condition monitoring (faulty or healthy) of such devices. The raw recorded data (vibration series emitted by the device) were first segmented and transformed to frequency domain using power spectra densities (PSD). Next, 15 variables denoting amplitudes of derived spectra were extracted. This yielded two data matrices A and B of size 1232×15, and 951×15, representing the faulty and the healthy device appropriately. The data are non-Gaussian and the covariances in both groups differ significantly.
Now, using Fisher’s discriminant criterion and the kernel methodology, we construct from a learning sample (counting only 600 items) a discriminant function able to provide a discriminant score for distinguishing between the healthy and faulty state of a gearbox. The function proved to be very effective: Both for the learning and the testing data samples (600 and 1483 data vectors respectively) we got 100% correct assignments to the ’faulty’ and ’healthy’ class, with a conspicuous margin between the two classes. The results are visualized in a 2D plane.
Keywordscondition monitoring discriminant analysis healthy state faulty state kernels nonlinearity gearbox diagnostics
- 2.Bartkowiak, A., Evelpidou, N.: Visualizing Some Multi-Class Erosion Data Using Kernel Methods. In: Rizzi, A., Vichi, M. (eds.) Proceedings in Computational Statistics, 17th Symposium Held in Rome, pp. 805–812. Physica-Verlag (2006)Google Scholar
- 3.Bartkowiak, A., Evelpidou, N.: Visualizing of some multi-class erosion data using GDA and supervised SOM. In: Saeed, K., et al. (eds.) Biometrics, Computer Security Systems and Artificial Intelligence Applications, pp. 13–22. Springer (2006)Google Scholar
- 4.Bartkowiak, A., Evelpidou, N., Vasilopoulos, A.: Visualization of Five Erosion Risk Classes using Kernel Discriminants. In: Pejaś, J., Saeed, K. (eds.) Advances in Information Processing and Protection, ch. 10, pp. 169–178. Springer (2007) ISBN: 978-0-387-73136-0 (Print) 978-0-387-73137-7 (Online)Google Scholar
- 5.Bartkowiak, A., Zimroz, R.: Outliers analysis and one class classification approach for planetary gearbox diagnosis. Journal of Physics: Conference Series 305(1), art. no. 012031 (2011)Google Scholar
- 7.Camastra, F.: Kernel Methods for Computer Vision, Theory and Applications, http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.23.9315
- 8.Camastra, F.: Kernel Methods for unsupervided Learning. PhD thesis. DISI-TH, Universita di Genova (2004)Google Scholar
- 9.Duda, R.O., Hart, P.E., Stork, D.G.: Pattern Classification, 2nd edn. Wiley (2001)Google Scholar
- 12.Liu, Z., Qu, J., Zuo, M.J., Xu, H.B.: Fault level diagnosis for planetary gearboxes using hybrid kernel feature selection and kernel Fisher discriminant analysis. Int. J. Adv. Manuf. Technol., doi:10.1007/s00170-012-4560-yGoogle Scholar
- 17.Zimroz, R., Bartkowiak, A.: Investigation on spectral structure of gearbox vibration signals by principal component analysis for condition monitoring purposes. Journal of Physics: Conference Series 305(1), art. no. 012075 (2011)Google Scholar
- 18.Zimroz, R., Bartkowiak, A.: Multidimensional data analysis for condition monitoring: features selection and data classification. In: CM 2012|MFPT 2012, London, June 11-14. Electronic Proceedings, art no. 402, pp. 1–12 (2012)Google Scholar