SCAE-Stacked Convolutional Autoencoder for Fault Diagnosis of a Hydraulic Piston Pump with Limited Data Samples

Area of Science:

• Engineering
• Artificial Intelligence
• Data Science

Background:

• Deep learning (DL) models excel in fault diagnosis but require substantial data.
• Limited data due to sensor issues poses a significant challenge for reliable DL model performance.
• Traditional machine learning (ML) methods have limitations in feature extraction and dimensionality reduction compared to DL.

Purpose of the Study:

• To develop a novel deep learning model, the stacked convolutional autoencoder (SCAE), to address the challenge of limited data in fault diagnosis.
• To enhance gradient information flow and extract richer hierarchical features for improved diagnostic accuracy.
• To evaluate the SCAE model's performance on fault diagnosis of a hydraulic piston pump using limited and noisy data.

Main Methods:

• Development of a novel stacked convolutional autoencoder (SCAE) model.
• Application of time-frequency visual pattern recognition for fault diagnosis.
• Evaluation of the SCAE model on limited data samples from a hydraulic piston pump.
• Comparative analysis against traditional DL models like DNN, SSAE, and CNN.

Main Results:

• The proposed SCAE model achieved excellent diagnostic performance with over 99.5% accuracy.
• The SCAE model demonstrated superior performance compared to DNN, SSAE, and CNN.
• The model exhibited robust diagnostic capabilities even under noisy data conditions.

Conclusions:

• The novel SCAE model effectively addresses the challenge of limited data in fault diagnosis systems.
• The SCAE model offers enhanced feature extraction and gradient flow for superior diagnostic accuracy.
• The proposed method provides a reliable and effective solution for hydraulic piston pump fault diagnosis, outperforming existing DL models.