I use data to build, measure, and improve the things that matter :-)

Predictive Maintenance Health Indicator for Train Fleet Compressors

Built an LSTM-Autoencoder to detect compressor degradation in a train fleet using real-world operational data. Unified three data sources (9.5M+ sensor readings, 5M+ diagnostic records, and 69K+ maintenance work orders) into a modelling-ready dataset through end-to-end data wrangling and feature engineering. The model identified signs of degradation up to 30 days before recorded failures, with a median lead time of 13 days for detected events.

• Python
• PyTorch
• LSTM
• Autoencoder
• Predictive Maintenance
• Time Series