Abstract
Fault classification is an essential tool for Prognostics and Health Management (PHM), whose goal is the assessment of the health condition of a machine and its components. In this research, we propose using Poincaré Images (PI) as features for fault classification using Convolutional Neural Networks (CNN). The approach is validated for classifying 17 valve faults in a reciprocating compressor (RC). The proposed approach attained an averaged classification accuracy of 94.97% which represents an improvement of about 15% concerning using discrete features extracted from the Poincaré plot for classifying faults with Random Forest models. Additionally, results show that the classification performance with PI is comparable to using spectrograms as alternative input images to CNN-based fault classifiers.
Translated title of the contribution | Las imágenes de Poincaré extraídas de las señales de vibración son características útiles para la clasificación de fallas en un compresor alternativo |
---|---|
Original language | English (US) |
State | Published - 28 Dec 2022 |
Event | XIX The conference on Latin America Control Congress - CU Duration: 28 Nov 2022 → 2 Dec 2022 https://lacc-2022.github.io/ |
Conference
Conference | XIX The conference on Latin America Control Congress |
---|---|
Period | 28/11/22 → 2/12/22 |
Internet address |
Keywords
- Poincaré
- Vibration signals
- Reciprocating compressor
CACES Knowledge Areas
- 827A Industrial maintenance