Rotary machinery has to be maintained and repaired exactly once or several times before they totally failed to keep their good working status. In order to determine the maintenance interval of rotary machinery, machine condition should be predicted by some diagnostic methods and predictive methods. This study was conducted to establish the relationship between machine performance and machine vibration and to validate the prediction model of the degradation status with the experimental data. In the experimental work, the vibration of rotary machinery was measured in two directions - (i) feed and ii) radial directions. The prediction model was developed by using the artificial neural network. A new experimental formula of neuron number in hidden layer for Back Propagation (BP) neural network with three (3) layers was developed. An improved learning rate was used to find the appropriate values of the parameters for the BP training algorithm. The results showed that the Improved Variable Learning Rate (IVLR) model only required eight min, which is half of the Traditional Variable Learning Rate (TVLR) processing time to converge to the global minimum. With either the IVLR or TVLR models, the network was able to avoid settling at the local minima and reach around 12 m2/min2 as the best-record mean square error. However, their respective least mean square errors were 12.2 and 11 m2/min, respectively. In conclusion, the IVLR showed better nonlinear approximation ability, shorter convergent time and higher discrimination ratio than the TVLR.
Key words: Machine vibration, degradation status, rotary machinery, TVLR, IVLR.
Copyright © 2023 Author(s) retain the copyright of this article.
This article is published under the terms of the Creative Commons Attribution License 4.0