Metro-Tunnel-Digital-Twin-Pipeline

Description

This dataset contains the Python source code and simulated sensor data used in the study "Improved Predictive Maintenance Management of Metro Tunnels by Integrating Digital Twin, Process Mining, and Explainable AI" (ITcon, 2026). Contents: Sensor_Data.py — Generates synthetic sensor data (temperature, accelerometer, extensometer) for a 150 m tunnel section with 36 sensors over 30 days. Risk_Score.py — Computes an initial composite risk score for each sensor based on signal statistics (max, std, P95, exceedance rate). integrated_pipeline.py — Main pipeline that combines Digital Twin, Process Mining (PM4Py), and Explainable AI (XGBoost + SHAP/LIME). Output files include event logs, process models (DFG, Heuristics Net, Inductive Tree), XGBoost risk predictions, SHAP/LIME explanations, confusion matrix, BIM-ready CSVs, and Excel dashboards. Data generation: Synthetic sensor signals were calibrated against real metro bogie vibration measurements from an open GitHub dataset (Cui, 2020). The dataset covers 30 days of operation (06:00–23:00) at a train speed of 10 m/s (36 km/h). How to use: Run the scripts in the following order: python Sensor_Data.py python Risk_Score.py python integrated_pipeline.py Interpretation: The outputs include event logs (936,292 events across 36 sensor cases), process models, risk predictions, and color-coded BIM-ready sensor health data. The four KPIs (Unplanned Failure Reduction, Maintenance Cost Reduction, MTBF Increase, Decision Time Reduction) are computed automatically. Citation: Mirzaei, A., Pourrostam, T., Jafari Nadooshan, M., Asghari, P., & Majrouhi Sardroud, J. (2026). Improved predictive maintenance management of metro tunnels by integrating digital twin, process mining, and explainable AI. Journal of Information Technology in Construction (ITcon), 31, 1-X. License: The source code is licensed under CC0-1.0. The associated paper is published under CC BY 4.0.

Steps to reproduce

1. Install Python 3.9+ and the required dependencies: pip install -r requirements.txt 2. Generate synthetic sensor data for the 150 m tunnel section with 36 sensors: python Sensor_Data.py 3. Calculate the composite risk score for each sensor: python Risk_Score.py 4. Run the integrated pipeline (Digital Twin + Process Mining + Explainable AI): python integrated_pipeline.py 5. The pipeline generates the following outputs in the `outputs/` folder: - Event logs (936,292 events, 10 activity types) - Process models (DFG, Heuristics Net, Inductive Tree) - XGBoost risk predictions with SHAP/LIME explanations - Confusion matrix and performance metrics - BIM-ready CSV files with color-coded sensor health - Excel dashboards (EN and FA versions) - KPI comparison charts 6. The four KPIs are computed automatically: - Unplanned Failure Reduction: 57.6% - Maintenance Cost Reduction: 19.2% - MTBF Increase: 135.7% - Decision Time Reduction: 66.7%

Institutions

• Islamic Azad University, Tehran

  Tehran, Tehran

Categories

Licence