Project Portfolio: Predictive Maintenance for Reciprocating Compressors

Reciprocating Compressor>

Project Details

  • Category: Machine Learning and Predictive Maintenance
  • Client: Repsol
  • Project date: June 2023

Description

This project focused on implementing a machine learning-based predictive maintenance (PdM) strategy for reciprocating compressors, critical equipment in the gas industry. The main objectives were to avoid breakdowns, optimize maintenance, and improve performance.

Key Achievements

  • Data Collection and Preparation: Processed hourly operational data from multiple compressors over two years, including both normal and failure periods.
  • Model Development: Trained 25 unsupervised machine learning models for anomaly detection, including traditional methods (Isolation Forest, PCA, ICA) and deep learning methods (LSTM, Autoencoders, Transformers).
  • Severity Classification: Classified anomalies into four levels – Low, Medium, High, Severe.
  • Proof of Concept (POC): Developed a real-time streaming POC integrating sensor data ingestion to dashboard visualization, enabling proactive maintenance and operational improvement.

Results

Demonstrated the potential to significantly reduce downtime and improve compressor availability through advanced predictive maintenance techniques.

By leveraging machine learning for predictive maintenance, this project showcases the application of advanced data analytics and modeling to enhance the reliability and efficiency of critical industrial equipment.