Abstract
Rotating machines are very critical equipment in the manufacturing and industrial sectors. The unexpected failure of this equipment can result in huge maintenance costs. To avoid such implications, the degradation in bearings of the machinery must be analysed accurately. Bearing degradation is generally monitored by changes in time-domain features. The time -domain features such as root mean square (RMS) and kurtosis represent vibration signals collected from the machinery bearing periodically.
Furthermore, using a prognostic approach, the remaining useful life (RUL) of the machinery bearing can be predicted by developing a prognostic model based on the extracted degradation features. In order to reduce maintenance cost, expensive downtime and maximum productivity, prognostic approaches are more helpful than diagnostic approaches.
A lot of predictive models have been developed in recent years for solving different types of data driven problems. Artificial neural network (ANN) is the most common data-driven technique in predictive maintenance. It has been used in previous studies for different applications including prediction of the RUL of rotating machinery. ANN is well known for its ability to handle non-linear datasets. However, many studies also show that ANN has its shortfalls. These limitations are related to the standard optimization algorithms used to train the model.
To overcome these limitations and achieve higher prediction accuracy, this present study explores the use of a Hybrid PSO-ANN model for RUL prediction of rotating machinery bearings. Particle Swarm Optimization (PSO) is a strong and modern optimization method that has been proved to perform well on a variety of optimization tasks. The hybrid of PSO – ANN model has shown superiority in many existing studies.
In this study, available bearing vibration monitoring data collected experimentally was used to train and test the proposed hybrid model. This data was provided by the Centre for Intelligent Maintenance Systems (IMS), and according to the experiment an outer race failure occurred on bearing 1 at the end of the test. A fitted mathematical equation describing the bearing degradation has been developed and used to extract training and testing datasets. Furthermore, the model uses time and fitted measurements Weibull hazard rates of RMS and kurtosis from present and previous points as inputs, and the life percentage as the output. This
v
By Mazibuko T.W. Copyright © University of Johannesburg, South Africa
was done to reduce the influence of the noise factors on the prediction performance of the model.
To understand the impact of adjusting the model parameters on the prediction performance and identification of the best performing hybrid PSO-ANN network, 36 runs of parametric analysis were performed. The results of this research study points out that the accuracy of the PSO-ANN model architecture depends on parameters such as the number of neurons, population size and acceleration factors.
The RUL prediction of the model was tested by comparing the actual and predicted output of the test data and the results obtained show that the Hybrid PSO – ANN model can potentially and effectively predict accurately the RUL of rotating machinery bearings.