End-to-end predictive maintenance pipeline for elevator operations

Abstract

This thesis develops a scalable predictive maintenance system for elevator operations by integrating gradient-boosted machine learning models, cloud-based automation, and interactive web visualization into a comprehensive end-to-end pipeline. A Gradient Boosted Trees (GBT) classifier was trained using PySpark and MLflow on the Databricks Lakehouse platform. The trained model was then deployed on a cloud-based virtual machine provisioned through the CSC Pouta Infrastructure-as-a-Service (IaaS) environment, enabling automated inference via scheduled execution. The system continuously monitors failure probabilities, triggers alerts based on defined thresholds, and delivers insights through automated email notifications and an interactive web dashboard built with Flask. The key contribution of this project lies in constructing a modular and reproducible predictive maintenance pipeline that demonstrates real-world applicability to industrial equipment monitoring tasks.