A novel fault diagnosis method based on nonlinear-CWT and improved YOLOv8 for axial piston pump using output pressure signal

Axial piston pumps are key components in hydraulic systems. Their condition monitoring and fault diagnosis are essential in ensuring the safety and reliability of the entire hydraulic system. Output pressure signals are commonly used for fault diagnosis due to their sensitivity to pump health condit...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Advanced engineering informatics Jg. 64; S. 103041
Hauptverfasser: Xia, Shiqi, Huang, Weidi, Zhang, Jie
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.03.2025
Schlagworte:
Bayesian algorithm
Continuous wavelet transform
Deep learning
Fault diagnosis
Horizontal deformable convolutional networks
Fault diagnosis
Deep learning
Bayesian algorithm
Horizontal deformable convolutional networks
Continuous wavelet transform
ISSN:1474-0346
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Axial piston pumps are key components in hydraulic systems. Their condition monitoring and fault diagnosis are essential in ensuring the safety and reliability of the entire hydraulic system. Output pressure signals are commonly used for fault diagnosis due to their sensitivity to pump health conditions. However, these signals are non-stationary, containing transient and pulse components, making traditional time and frequency analysis methods insufficient for accurate fault diagnosis. To address this, a novel fault diagnosis method is proposed. First, a Nonlinear Continuous Wavelet Transform (Nonlinear-CWT) is developed to transforms one-dimensional output pressure signals into two-dimensional time–frequency images, amplifying key signal characteristics and reducing noise. Then, a Horizontal Deformable Convolutional Network (HDCN) is proposed to handle horizontal deformations in the images caused by varying sampling lengths, replacing standard convolution modules in the YOLO (You Only Look Once) model. Lastly, Bayesian Optimization (BO) is employed to automatically optimize the hyperparameters, thereby producing a BHDCN-YOLO model. The experimental data of six health conditions with pump output pressure of 21 MPa is collected. Model performances are analyzed through the ablation experiments, comparison of other five deep learning model, and dataset with signal-to-noise ratios (SNR) of 30–50 dB. The results show that the BHDCN-YOLO model achieves an average accuracy of 97.38 % and inference speed of 0.9 ms. BHDCN-YOLO model accuracy improved by 20.2 % compared to the YOLO model. The adaptability experiment verified that HDCN-YOLO also has good recognition accuracy on datasets with additional sampling lengths. This study provides a novel method for more accurately diagnosing faults in axial piston pumps.
ISSN:1474-0346
DOI:10.1016/j.aei.2024.103041