Fault Diagnosis Method for Reciprocating Compressors Based on Spatio-Temporal Feature Fusion.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Reciprocating compressors are vital in petrochemical and natural gas sectors but face component failures due to variable loads and impacts.
• Current fault diagnosis methods lack effective spatio-temporal feature extraction and correlation analysis.
• Failures in critical components like valves and piston rings necessitate improved diagnostic techniques.

Purpose of the Study:

• To develop an advanced fault diagnosis method for reciprocating compressors.
• To address limitations in existing methods regarding spatio-temporal feature extraction and correlation.
• To enhance the precision and reliability of fault identification in critical compressor components.

Main Methods:

• A spatio-temporal feature fusion model (STFFM) was developed.
• The model integrates a multi-layered stacked bidirectional gated recurrent unit (BiGRU) for temporal feature extraction and an enhanced graph isomorphism network (GIN) for spatial feature extraction.
• A bidirectional multi-head attention mechanism and cross-modal gated updates were employed for feature fusion and selection.

Main Results:

• The STFFM effectively integrates spatio-temporal fault features.
• The model achieved a high average accuracy of 99.14% on an experimental dataset.
• The method demonstrated superior performance in capturing complex fault characteristics.

Conclusions:

• The proposed spatio-temporal feature fusion method offers an effective technical pathway for precise fault diagnosis in reciprocating compressors.
• This approach enhances the reliability and safety of critical industrial equipment.
• The findings contribute to advancing diagnostic capabilities in the petrochemical and natural gas industries.