Kotra/turbofan-engine-predictor

✈️ Turbofan Engine Predictive Maintenance Dashboard

(https://static.streamlit.io/badges/streamlit_badge_black_white.svg)

This project is an end-to-end data science application that predicts the Remaining Useful Life (RUL) of turbofan engines using the NASA C-MAPSS dataset. It features a live, interactive dashboard for fleet monitoring, maintenance recommendations, and model-driven insights.

🚀 Key Features

• —Fleet-Wide Monitoring: The sidebar features a "Red Alert" list that automatically identifies engines requiring immediate maintenance, allowing managers to prioritize action.
• —Real-Time RUL Prediction: Select any engine to see its live predicted Remaining Useful Life (RUL) in cycles, calculated by our V2 tuned Random Forest model.
• —Actionable Business Logic: The dashboard provides clear, three-tiered maintenance recommendations (🟢 Green, 🟡 Yellow, 🔴 Red) based on a cost-benefit analysis that balances failure risk (~$2M) against maintenance cost (~$250k).
• —Model Explainability (XAI): A "Model's Top 5 Predictive Features" chart explains what the model found most important for making predictions across the fleet.
• —Sensor Degradation Analysis: Visualizes the degradation trends of the top 4 most critical sensors for any selected engine.

🛠️ Tech Stack

• —Data Analysis: Python, Pandas, NumPy, SQL (SQLite)
• —Machine Learning: Scikit-Learn (RandomForestRegressor, RandomizedSearchCV)
• —Dashboard & Deployment: Streamlit, Streamlit Community Cloud
• —Data Versioning: Git & Git LFS (for handling the >100MB model file)
• —Data: NASA C-MAPSS (FD001 dataset)

📈 Data & Modeling Pipeline

This project was built in a clear, phased approach:

1. 1.Phase 1: Data Preparation:
2. 2.Loaded the raw FD001 text files into Pandas.
3. 3.Cleaned the data by dropping constant-value sensors.
4. 4.Engineered the `RUL` (Remaining Useful Life) target variable by calculating the time-to-failure for each engine in the training set.
5. 5.Stored the final, cleaned dataset in a persistent SQLite database (turbofan.db).

1. 1.Phase 2: Exploratory Data Analysis (EDA):
2. 2.Used SQL queries to analyze engine lifespan distributions.
3. 3.Built correlation heatmaps to identify the most predictive sensors (e.g., sensor_11, sensor_4, sensor_7, sensor_12).

1. 1.Phase 3: V1 Model (Baseline):
2. 2.Engineered time-series features (e.g., sensor_4_avg, sensor_11_std) using a 5-cycle rolling window.
3. 3.Trained a baseline RandomForestRegressor, achieving **

1. 1.Phase 4: V2 Model (Tuning):
2. 2.Used RandomizedSearchCV to tune the model's hyperparameters.
3. 3.This improved the final model performance, creating our V2 production model.

📊 V2 Model Performance

• —R-squared (R²): 0.62
• —Root Mean Squared Error (RMSE): 47 cycles

This improved RMSE allows for a more precise maintenance threshold. The app's business logic (get_recommendation) was updated to use this new, more accurate rmse=47 value for its "Red Alert" buffer.

🔧 How to Run This Project Locally

1. 1.Clone the repository:

    git clone [https://github.com/KotraHaridutt/Engine_Predictive_System.git](https://github.com/KotraHaridutt/Engine_Predictive_System.git)
    cd Engine_Predictive_System

1. 1.Install Git LFS (required to download the model file):

    git lfs install
    git lfs pull

1. 1.Install dependencies:

    pip install -r requirements.txt

1. 1.Run the Streamlit app:

    streamlit run app.py

💡 Future Improvements

• —V3 Model (LSTM): Experiment with a Long Short-Term Memory (LSTM) neural network, which is specifically designed for time-series data, to potentially achieve a significant boost in R² score.
• —Fleet-Wide Visuals: Add a histogram to the dashboard showing the RUL distribution of the entire 100-engine fleet, color-coded by alert status.
• —Cost Calculator: Implement the "Cost Avoidance Calculator" to show the total dollar amount saved by servicing the red-alert engines.