🤖 Predictive Maintenance System

A deep learning–based predictive maintenance system that estimates the Remaining Useful Life (RUL) of aircraft engines using LSTM networks on the NASA CMAPSS FD001 dataset.

This project demonstrates an end-to-end machine learning pipeline for Remaining Useful Life (RUL) prediction. It loads the NASA CMAPSS dataset, preprocesses multivariate sensor data, generates fixed-length time-series sequences, trains an LSTM network, evaluates prediction performance using RMSE, and outputs simple maintenance status recommendations.

---

✨ Features

• 🚀 Remaining Useful Life (RUL) prediction
• 🧠 LSTM-based sequence modeling
• 📊 Automatic time-series sequence generation
• ⚙️ Data preprocessing & Min-Max normalization
• ⏹️ Early stopping during training
• 📉 RMSE evaluation on test engines
• 🔧 Rule-based maintenance recommendations
• 📈 Automatic training loss visualization
• 💾 Automatic trained model saving

---

🛠 Tech Stack

• Python
• TensorFlow / Keras
• NumPy
• Pandas
• Scikit-learn
• Matplotlib

---

📂 Project Structure

Predictive-Maintenance-System/
│
├── data/
│   ├── train_FD001.txt
│   ├── test_FD001.txt
│   ├── RUL_FD001.txt
│   └── readme.txt
│
├── docs/
│   ├── research-paper.pdf
│   └── Damage Propagation Modeling.pdf
│
├── output/
│   ├── models/
│   │   └── final_model.keras
│   └── plots/
│       └── loss.png
│
├── src/
│   ├── data_loader.py
│   ├── preprocessing.py
│   ├── sequence.py
│   └── model.py
│
├── train.py
├── requirements.txt
├── README.md
└── .gitignore

__pycache__/ and .venv/ are local-generated files and are intentionally excluded.

---

📸 Training Result

Training vs Validation Loss

---

⚙️ Setup

Create a virtual environment (recommended).

Windows

python -m venv .venv
.venv\Scripts\activate

Linux / macOS

python3 -m venv .venv
source .venv/bin/activate

Install the required packages:

pip install -r requirements.txt

Recommended Python Version: Python 3.11 (TensorFlow currently has limited support for newer Python releases.)

---

▶️ How to Run

python train.py

The script will:

• Load the NASA CMAPSS FD001 dataset
• Preprocess sensor readings
• Generate time-series sequences
• Train the LSTM model
• Evaluate RMSE
• Save the trained model
• Save the training loss plot
• Print maintenance recommendations

---

📊 Dataset

The project uses the NASA CMAPSS FD001 turbofan engine degradation dataset.

FD001 contains:

• 100 training engine trajectories
• 100 test engine trajectories
• One operating condition
• One fault mode (HPC degradation)

Files used:

• train_FD001.txt
• test_FD001.txt
• RUL_FD001.txt

The loader reads the space-separated files, removes empty columns, and assigns the following schema:

id, cycle, op1, op2, op3, s1 ... s21

The original NASA reference paper is included in:

docs/Damage Propagation Modeling.pdf

---

⚙️ Preprocessing

The preprocessing pipeline performs:

• Removal of unnecessary columns
• Min-Max feature scaling
• Feature / label separation
• Engine-wise sequence generation
• Fixed sequence length of 30 cycles

Dropped columns:

• cycle
• op3
• s1
• s5
• s10
• s16
• s18
• s19

---

🔄 Pipeline Overview

The complete workflow of the project is illustrated below:

NASA CMAPSS Dataset
          │
          ▼
Data Loading
          │
          ▼
RUL Calculation
          │
          ▼
Data Preprocessing
(Min-Max Scaling)
          │
          ▼
Sequence Generation
(30 Time Steps)
          │
          ▼
LSTM Network
          │
          ▼
RUL Prediction
          │
          ▼
RMSE Evaluation
          │
          ▼
Maintenance Decision

---

🧠 Model Architecture

The LSTM network consists of:

• LSTM (50 units, return sequences)
• Dropout (0.2)
• LSTM (25 units)
• Dropout (0.2)
• Dense output layer (ReLU)

Training configuration:

• Optimizer: Adam
• Loss: Mean Squared Error (MSE)
• Batch Size: 32
• Epochs: 50
• Validation Split: 20%
• Early Stopping (Patience = 5)

---

📈 Outputs

During execution the project:

• Trains the LSTM model
• Evaluates RMSE
• Prints Remaining Useful Life predictions
• Generates maintenance recommendations

Maintenance status:

RUL < 25  → Critical

RUL < 50  → Warning

Otherwise → Healthy

Generated files:

output/models/final_model.keras

output/plots/loss.png

---

📄 Research Paper

A detailed project report is available at:

docs/research-paper.pdf

The original NASA reference paper is available at:

docs/Damage Propagation Modeling.pdf

---

📦 Dependencies

Install all dependencies with:

pip install -r requirements.txt

Main libraries:

• TensorFlow / Keras
• NumPy
• Pandas
• Scikit-learn
• Matplotlib

---

🚀 Future Improvements

• Support FD002, FD003 and FD004 datasets
• Save preprocessing scaler
• Hyperparameter tuning
• Compare LSTM with GRU and Transformer models
• Interactive prediction dashboard
• Docker support
• Model inference script
• Cloud deployment

---

👨‍💻 Author

Vihaan Jain

If you found this project useful, consider ⭐ starring the repository.