High Impact Factor : 4.396 icon | Submit Manuscript Online icon |

Encoder Decoder based Linear Discriminant Analysis Technique for the Condition Monitoring of Induction Motor using Stator Current Signal


Ambuj Kumar , National Institute of Technology Delhi; Aniket Pathak, National Institute of Technology Delhi; Maneesh Kumar Singh, Findish India Pvt. Ltd.


Autoencoder, linear discriminant analysis, LDA, condition monitoring, induction machine


Induction machines are used in every industry due to their robust design, easy construction and high reliability. In-spite of the advantages, these machines are also prone to various faults, which needs to be detected and rectified timely in order to safeguard the concerned industries from system failure. Fault detection plays a crucial part in order to find the most suitable diagnosis method for the machine. In this paper we demonstrate a novel fault classification system for the identification of various health conditions of Induction machine using Autoencoder based Linear Discriminant Analysis (LDA). A total of eleven statistical features are calculated for various health conditions (i.e., healthy, broken rotor bar, inner race bearing fault, outer race bearing fault) by varying the loading conditions. LDA classifier implemented on the reconstructed output feature of autoencoder which is almost similar to normally calculated features having less noise, more learnt information and non-linear transformation. Here, grid search algorithm is used to tune the hyper parameters of autoencoder. Autoencoder assists LDA to project the data in lower dimension with maximized separability that helps LDA to classify more accurately and avoid over fitting. With the advent of this technique fault identification and diagnosis can be done efficiently at real-time in the industries.

Other Details

Paper ID: IJSRDV7I110196
Published in: Volume : 7, Issue : 11
Publication Date: 01/02/2020
Page(s): 314-318

Article Preview

Download Article