Aircraft Predictive Maintenance

This project uses Recurrent Neural Networks (RNNs) to predict the Remaining Useful Life (RUL) of aircraft engines based on historical sensor readings. The aim is to anticipate potential failures and schedule maintenance efficiently, enhancing aircraft safety and reducing downtime.

---

📁 Project Structure

Aircraft-Predictive-Maintenance-main/
│
├── Aircraft_Data.ipynb      # Main Jupyter notebook with data processing, modeling, and evaluation
├── PM_train.txt             # Training dataset (sensor readings until failure)
├── PM_test.txt              # Test dataset (sensor readings, truncated before failure)
├── PM_truth.txt             # Ground truth for RUL of test dataset
└── README.md                # Project documentation (you're reading it!)

---

🧠 Problem Statement

Aircraft engines are equipped with sensors that generate time-series data over operational cycles. Each engine eventually fails, and we aim to predict the number of cycles remaining until failure (RUL) using this sensor data.

---

📊 Dataset Description

The dataset is sourced from NASA's C-MAPSS dataset. It consists of three parts:

• PM_train.txt: Sensor data for multiple engines up to the point of failure.
• PM_test.txt: Sensor data ending before failure.
• PM_truth.txt: Actual RUL values for each engine in the test set.

Each row in the train/test files includes:

• Engine ID
• Operational cycle
• 3 operational settings
• 21 sensor measurements (s1 to s21)

---

🔍 Workflow Overview

1. Data Preprocessing

• Load datasets and clean missing values.
• Rename columns for clarity.
• Sort records by engine ID and cycle.
• Calculate RUL for training data: RUL = max_cycle - current_cycle.
• Normalize both train and test datasets using MinMaxScaler.

2. Exploratory Data Analysis (EDA)

• Visualize trends in sensor values over time.
• Example: Sensor 1 tends to increase, while Sensor 6 tends to decrease as engines approach failure.
• Observation: Certain sensor patterns correlate with approaching failure.

3. Label Generation

• Define binary classification target based on a threshold RUL (e.g., 30 cycles).
• Useful for modeling binary classification and regression tasks.

4. Sequence Generation

• Convert sensor time-series into input sequences for RNN models.
• Sliding windows are used to create overlapping sequences of cycles.

---

🔁 Model Architecture

Several RNN-based deep learning models are trained:

🔹 Simple RNN (1 Feature)

• Uses a single sensor (e.g., s1) as input.
• Basic sequential model with SimpleRNN, Dense, and Dropout layers.

🔹 Simple RNN (25 Features)

• Uses all 25 features (3 settings + 21 sensors).
• Deeper model with increased input dimensionality.

🔹 Bidirectional RNN

• Uses bidirectional RNN layers for improved context understanding.
• Helps capture both forward and backward dependencies in the sequence.

---

📈 Evaluation Metrics

Model performance is evaluated using:

• Accuracy
• Precision
• Recall
• Confusion Matrix

Training and validation losses/accuracies are plotted to monitor learning progress.

---

🔬 Results & Observations

• Sensor patterns like rising s1 and falling s6 are strong indicators of imminent failure.
• Models using all 25 features outperform those with a single feature.
• RNNs effectively capture temporal dependencies in sensor data.
• Proper normalization and sequence generation are key to model performance.

---

🚀 How to Run

1. Clone the repository:

   git clone https://github.com/Pouriadia/Aircraft-Predictive-Maintenance.git
   cd Aircraft-Predictive-Maintenance

2. Install dependencies:

   pip install -r requirements.txt

3. Open the Jupyter notebook:

   jupyter notebook Aircraft_Data.ipynb

---

📌 Future Work

• Integrate more complex architectures like LSTM, GRU, or Transformer-based models.
• Add model explainability (e.g., SHAP or LIME).
• Explore ensemble modeling for improved RUL prediction.

---

📚 References

• NASA C-MAPSS Dataset: https://www.nasa.gov/cmapps
• TensorFlow Documentation: https://www.tensorflow.org/
• Original paper: Saxena, A., & Goebel, K. (2008). "Turbofan Engine Degradation Simulation Data Set".