Fault Diagnosis of Hydraulic Components Based on Multi-Sensor Information Fusion Using Improved TSO-CNN-BiLSTM

Area of Science:

• Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Hydraulic systems are critical in many industries but prone to faults.
• Accurate and reliable fault diagnosis is essential for system maintenance and safety.
• Existing diagnostic methods may lack efficiency or robustness in complex environments.

Purpose of the Study:

• To propose an advanced fault diagnosis model for hydraulic systems.
• To enhance diagnostic accuracy and robustness using sensor fusion and intelligent optimization.
• To leverage deep learning for automatic feature extraction and analysis.

Main Methods:

• Sensor selection using the random forest algorithm.
• Feature extraction and fusion via convolutional neural networks (CNNs).
• Sequential data analysis using bi-directional long short-term memory (BiLSTM) networks.
• Optimization of CNN-BiLSTM parameters using an improved tuna swarm optimization (ITSO) algorithm.

Main Results:

• High diagnostic accuracies achieved: plunger pump (99.07%), cooler (99.4%), throttle valve (98.81%), accumulator (98.51%).
• The model effectively extracts fused features and utilizes multi-sensor information.
• Demonstrated good robustness to noise interference across various signal-to-noise ratios (SNRs).

Conclusions:

• The proposed ITSO-CNN-BiLSTM model offers accurate and reliable fault diagnosis for hydraulic systems.
• The method shows significant potential for real-world applications requiring robust diagnostic capabilities.
• ITSO effectively optimizes deep learning models for complex signal processing tasks.