Design of MLP-NN Classifier Block with PCA-Type of Dimensionality Reduction Technique for Assessment of State of Degradation in Stator Insulation of Induction Motor
Amit J. Modak1, H. P. Inamdar2
1Prof. Amit J. Modak, Ph. D. Research Student, Department of Electrical Engineering, Walchand College of Engineering, Sangli, India.
2Dr. H. P. Inamdar, Ex-Professor & Head, Department of Electrical
Engineering, Walchand College of Engineering, Sangli, India.
Manuscript received on August 14, 2015. | Revised Manuscript received on August 26, 2015. | Manuscript published on September 05, 2015. | PP: 133-141 | Volume-5 Issue-4, September 2015 . | Retrieval Number: D2721095415 /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: In the present work, the design of discrete ‘ANN’ simulation model was done for the classification and qualitative assessment of the state of degradation of insulation present in the respective phases of three-phase ac induction motor. The ‘ANN’ simulation model consists of numbers of discrete neural network classifier blocks. The extraction of mathematical parameters of stator current data pattern, which are simulating the specific state of degradation of insulation based on Park’s current transformation model, were presented in the previous research papers. Further, the optimal design specifications of the various neural network classifier blocks, which were realized on the basis of ‘multilayer perceptron’ (MLP) and ‘radial basis function’ (RBF) types of neural network architectures were compared in the same papers. The striking generalizations, which were derived on the basis of the comparative performance analysis resemble that the general optimum design specifications, which are determined on the basis of ‘MLP’ network are preferred as an optimum choice over the ‘RBF’ network. The aim of the present research paper is to explore the possibility of any further reduction in the size of the ‘MLP’ network. The present investigation emphasis the use of ‘principal component analysis’ type of dimensionality reduction technique for the simplification and improvement in the design of discrete neural network classifier blocks, which were already designed on the basis of ‘multilayer perceptron’ (MLP) neural network architecture for the classification and qualitative assessment state of degradation of insulation in three-phase ac induction motor.
Keywords: Induction motor, stator insulation, dimensionality reduction technique, principal component analysis (PCA), sensitivity analysis (SA), artificial neural network (ANN).