Apply Pruning Algorithm for Optimizing Feed Forward Neural Networks for Crack Identifications in Francis Turbine Runner
Raza Abdulla Saeed1, Loay Edwar George2
1Raza Abdulla Saeed, Faculty of Engineering, University of Sulaimani, Sulaimani, Kurdistan Region, Iraq.
2Loay Edwar George, College of Science, University of Baghdad, Baghdad, Iraq.
Manuscript received on September 01, 2012. | Revised Manuscript received on September 02, 2012. | Manuscript published on September 05, 2012. | PP: 166-175 | Volume-2 Issue-4, September 2012. | Retrieval Number: D0928082412/2012©BEIESP
Open Access | Ethics and Policies | Cite
© The Authors. Published By: Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: In this study the Feed Forward Artificial Neural Networks (FFANN) for crack identification and estimates the turbine operating conditions in Francis turbine type was investigated. The sets of vibration data were used as vibrational signatures for studied mechanical structure, and they fed to FFANN as input vector for identification purpose. Different arrangements of FFANN were taken into consideration to find out the best topology which can produce identification results with acceptable accuracy levels. In order to examine the performance of the FFANN and obtain the satisfactory arrangements, different numbers of input data sets are tested. The test results showed that the use of very large number of input data will cause a large increase in training time beside to it may lead to unstable FFANN with over-fitting. To avoid these deteriorated results, different data reduction techniques have been proposed for reducing dimensionality of the input data to achieve an acceptable data reduction level. The conducted results indicated that the FFANN models have been successfully employed for crack identification and estimates the turbine operating conditions using vibration data sets. Moreover the results revealed that the pruning mechanism which is based on the data reduction mechanism can led to satisfactory results.
Keywords: Crack Identifications, Feed Forward Artificial Neural Networks, Francis Turbines Runner, Pruning Algorithm