Artificial Neural Network for the Clustering of Vibration Signals for Condition Monitoring of Rotating Machines
Vibration analysis is commonly used to provide valuable insights into the condition of a machine, which is crucial for ensuring reliability and reducing maintenance costs. However, analysis of vibration signals using artificial neural network (ANN) is mostly via development of classification models...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
College of Engineering of Afe Babalola University, Ado-Ekiti (ABUAD), Ekiti State, Nigeria
2025-01-01
|
| Series: | ABUAD Journal of Engineering Research and Development |
| Subjects: | |
| Online Access: | https://journals.abuad.edu.ng/index.php/ajerd/article/view/949 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Vibration analysis is commonly used to provide valuable insights into the condition of a machine, which is crucial for ensuring reliability and reducing maintenance costs. However, analysis of vibration signals using artificial neural network (ANN) is mostly via development of classification models, which cannot be suitably applied to several varied machine types and specifications. This study investigates the use of ANN in the clustering of vibration signals for machine condition monitoring of several rotating machines. Data obtained from different rotating machines for 4 years was utilized for the study. The data contained values of vibration signals taken at 12 different pickup points, power ratings, year and equipment location. The obtained data was preprocessed and analyzed statistically. Then, silhouette scores and within-cluster sum of squares (WCSS) were used to obtain the optimum number of clusters for the analysis. Afterwards, different clusters were created using ANN, which were then explored to gain insights for potential applicability of the technique for assessment of the conditions of rotating machines. The result of ANOVA showed that there were significant variations between readings obtained from different pickup points and readings obtained from the different machines, with p-values far less than 0.05 for both cases. It was found via silhouette and WCSS that 9 was an optimum number of clusters for the analysis. Calculated mean of standardized values informs that 6 clusters contained machines with different forms of faults, having positive mean values far greater than 0. Also, there were 2 clusters with machines having good working conditions with negative mean values, while one cluster had machines that were moderately okay with mean values close to 0. The study has shown that ANN can effectively cluster a set of machines based on their conditions using vibration signals taken at different pick-up points. The developed framework is a suitable alternative to ANN-based classification methods which have limited applicability.
|
|---|---|
| ISSN: | 2756-6811 2645-2685 |