Improved Fault Diagnosis in Hydraulic Systems with Gated Convolutional Autoencoder and Partially Simulated Data

Area of Science:

• Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Hydraulic systems are critical in many industries but prone to failures.
• Effective fault detection is essential for operational reliability and safety.
• Traditional methods can be time-consuming and data-intensive.

Purpose of the Study:

• To develop and evaluate a novel neural network architecture for hydraulic system fault detection.
• To reduce the reliance on extensive real-world data for model training.
• To achieve high accuracy in identifying hydraulic system states.

Main Methods:

• A gated convolutional autoencoder neural network architecture was proposed.
• The model was trained on a simulated dataset augmented with a small percentage (0.2%) of real test bench data.
• Fault detection performance was validated on a physical test bench using a 10-second sampling window.

Main Results:

• The proposed model achieved over 99% accuracy in recognizing hydraulic system states.
• The training process was significantly accelerated by leveraging simulated data.
• The model demonstrated the ability to analyze decision boundaries in a 2D embedding space.

Conclusions:

• The gated convolutional autoencoder is effective for high-accuracy hydraulic system fault detection.
• Minimal real-world data augmentation can create robust fault detection models.
• The developed model offers an efficient and accurate solution for industrial applications.