A machine learning system for predictive maintenance, anomaly detection, and Remaining Useful Life (RUL) prediction for industrial machines using time-series sensor data.
This project uses LSTM Autoencoder networks to learn normal behavior patterns of machine sensors and detect anomalies that may indicate potential machine failures.
Modern industries rely on predictive maintenance to reduce downtime and avoid catastrophic machine failures. Traditional maintenance strategies like reactive or scheduled maintenance are inefficient.
This project implements a data-driven predictive maintenance system capable of:
• Detecting abnormal machine behavior
• Predicting machine failure patterns
• Estimating Remaining Useful Life (RUL)
The model is trained using sensor time-series data from machine bearings and identifies anomalies using reconstruction error from an LSTM Autoencoder.
- Time-series sensor data processing
- LSTM Autoencoder architecture
- Anomaly detection using reconstruction loss
- Remaining Useful Life (RUL) estimation
- Data preprocessing and feature scaling
- Sliding window sequence generation
- Visualization of anomalies and predictions
The project uses the IMS Bearing Dataset which contains vibration sensor data from industrial bearings.
Dataset includes:
- Vibration signals
- Sensor measurements
- Time series machine condition data
Dataset file used in this repository: merged_Dataset_BearingTest_2.csv
Reference:
IMS Bearing Dataset – University of Cincinnati
- Data Loading
- Data Preprocessing
- Sliding Window Sequence Creation
- Feature Normalization
- LSTM Autoencoder Model Training
- Reconstruction Error Calculation
- Anomaly Detection
- Remaining Useful Life Estimation
- Visualization of Results
The core model used in this project is an LSTM Autoencoder consisting of:
Encoder
- LSTM Layers
- Latent Representation
Decoder
- LSTM Layers
- Reconstruction of Input Sequence
Anomalies are detected when the reconstruction error exceeds a predefined threshold.
Programming Language
- Python
Libraries & Frameworks
- NumPy
- Pandas
- Scikit-Learn
- TensorFlow / Keras
- Matplotlib
- Seaborn
Development Environment
- Jupyter Notebook
Clone the repository
Install required libraries pip install numpy pandas scikit-learn tensorflow matplotlib seaborn Run the notebook
The model successfully:
- Detects abnormal machine behavior
- Identifies early signs of failure
- Estimates machine degradation
- Predicts remaining useful life
Visualizations include:
- Reconstruction error plots
- Anomaly detection graphs
- Sensor behavior trends
Predictive maintenance systems like this can be applied in:
- Manufacturing plants
- Power plants
- Aerospace systems
- Automotive industry
- Smart factories (Industry 4.0)
- Real-time anomaly detection
- Deployment using Flask / FastAPI
- Integration with IoT sensor streams
- Edge deployment
- Dashboard visualization
- Sujai Shukla
- Kumthe Kartik Sanjay
Machine Learning & AI Enthusiast
This project is for academic and research purposes.