Prognostics of the Motor Coupling Based on the LS-SVM Regression using Features in Time Domain

A coupling is a device used to connect two shafts together at their ends for the purpose of transmitting power. The primary purpose of couplings is to join two pieces of rotating equipment while permitting some degree of misalignment or end movement or both. Coupling failure can stop the system operation that leads to loss of production, so that, maintenance strategy that including coupling condition monitoring is needed to determine the state of the coupling. When the coupling experience failure, it is needed to know in its initial stage and predict its future development so that it is enough time for the maintenance division to do action needed to avoid catastrophic failure. Prognostics of the coupling will do this task. This paper aims to develop prognostics of the motor coupling. Run to failure test is conducted on the coupling that mounted at the end of the motor shaft. The motor is run at the speed of 1500 rpm, and the coupling is loaded by 1.23 KN bending load. The vibration signal is acquired using two accelerometers that mounted at the motor with a sampling rate of 10 kHz. Ten features in the time domain are extracted from the vibration signal. They are: mean, rms, shape factor, skewness, kurtosis, crest factor, entropy estimation, entropy estimation error, histogram lower bound, and histogram upper bound. Then, the features are selected based on the monotonicity and trendability criteria that stated as the feature importance (FI). The selected feature will be used in the prognostics process. The rms feature, that has the highest FI is then used for prognostics of the motor coupling. After rms feature smoothing, the feature was divided into training data dan testing data. The training data are used to develop the Least Square - Support Vector (LS-SVM) regression model using RBF Gaussian Kernel that would be used to predict the future state of the coupling failure. The prediction performance was measured in the term of root mean square error (RMSE) and the mean absolute error (MAE). The result shows that the LS-SVM regression model developed can perform very well in predicting the motor coupling failure. The RMSE and MAE value are 0.0082 and 0.0061 respectively.