Test for Efficient Increase/ Decrease Factors for Resilient Backpropagation using Combustion Engine Vibration Signals
N. D. Liyanagedera1, A. Ratnaweera2, D. I. B. Randeniya3
1N. D. Liyanagedera Department of Computing & Information Systems, Wayamba University of Sri Lanka Kuliyapitiya, Sri Lanka.
2A. Ratnaweera Department of Mechanical Engineering, University of Peradeniya, Peradeniya, Sri Lanka.
3D. I. B. Randeniya, Department of Mechanical Engineering,  University of Peradeniya, Peradeniya, Sri Lanka.
Manuscript received on October 15, 2015. | Revised Manuscript received on October 29, 2015. | Manuscript published on November 05, 2015. | PP: 45-51 | Volume-5 Issue-5, November 2015 . | Retrieval Number: E2751115515/2015©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: Resilient backpropagation is a recently emerging neural network with a high potential. This neural network is capable of handling a network structure with a large number of input nodes which many other networks fail. This is achieved based on the way the weights are updated in the algorithm. Resilient backpropagation uses two constant values, decrease factor [ ] and increase factor [ ] to update the weights and to get the optimal solution. This experiment checks on to find how the neural network performs when different values were used for these constants. The training, testing and validation of the neural networks were done using vibration signal data collected from a combustion engine corresponding to 16 different fault combinations available in the combustion engine. The performances of the networks were compared using mean square error, time and epoch. The final results indicate that, when the decrease factor is in the range of 0.5 to 0.6 and when the increase factor is in the range of 1.2 to 1.3 the resilient backpropagation algorithm has the best performance.
Keywords: Resilient Backpropagation Algorithm, Increase/Decrease Factor, Combustion Engine, Vibration Signals.