Advanced Data Augmentation Techniques for Enhanced Fault Diagnosis in Industrial Centrifugal Pumps DOI Creative Commons

Dong-Yun Kim,

Akeem Bayo Kareem, Daryl Domingo

et al.

Journal of Sensor and Actuator Networks, Journal Year: 2024, Volume and Issue: 13(5), P. 60 - 60

Published: Sept. 25, 2024

This study presents an advanced data augmentation framework to enhance fault diagnostics in industrial centrifugal pumps using vibration data. The proposed addresses the challenge of insufficient defect settings by integrating traditional techniques, such as Gaussian noise (GN) and signal stretching (SS), with models, including Long Short-Term Memory (LSTM) networks, Autoencoders (AE), Generative Adversarial Networks (GANs). Our approach significantly improves robustness accuracy machine learning (ML) models for detection classification. Key findings demonstrate a marked reduction false positives substantial increase rates, particularly complex operational scenarios where statistical methods may fall short. experimental results underscore effectiveness combining these achieving up 30% improvement 25% compared baseline models. These improvements highlight practical value ensuring reliable operation predictive maintenance diverse environments.

Language: Английский

Advanced Data Augmentation Techniques for Enhanced Fault Diagnosis in Industrial Centrifugal Pumps DOI Creative Commons

Dong-Yun Kim,

Akeem Bayo Kareem, Daryl Domingo

et al.

Journal of Sensor and Actuator Networks, Journal Year: 2024, Volume and Issue: 13(5), P. 60 - 60

Published: Sept. 25, 2024

This study presents an advanced data augmentation framework to enhance fault diagnostics in industrial centrifugal pumps using vibration data. The proposed addresses the challenge of insufficient defect settings by integrating traditional techniques, such as Gaussian noise (GN) and signal stretching (SS), with models, including Long Short-Term Memory (LSTM) networks, Autoencoders (AE), Generative Adversarial Networks (GANs). Our approach significantly improves robustness accuracy machine learning (ML) models for detection classification. Key findings demonstrate a marked reduction false positives substantial increase rates, particularly complex operational scenarios where statistical methods may fall short. experimental results underscore effectiveness combining these achieving up 30% improvement 25% compared baseline models. These improvements highlight practical value ensuring reliable operation predictive maintenance diverse environments.

Language: Английский

Citations

2