Neuromorphic Digital Twin for Remaining Useful Life Prediction

This repository presents a Neuromorphic Digital Twin model for predicting the Remaining Useful Life of industrial machinery using Spiking Neural Networks. The project is developed as part of a Master's thesis in Data Science and Artificial Intelligence.

Overview

The goal of this project is to estimate the Remaining Useful Life of machinery using vibration signal data. The approach combines feature engineering, spike-based encoding, and neuromorphic learning to build an efficient predictive model.

Dataset

Dataset used: IMS Bearing Run-to-Failure Dataset Set 1

https://data.nasa.gov/dataset/ims-bearings

The dataset contains vibration signals collected from bearings until failure. These signals are processed and converted into structured features for model training.

Methodology

The workflow consists of three main stages:

Feature Engineering A total of 128 features are extracted from raw vibration signals, including both time-domain and frequency-domain features.

Spike Encoding Continuous features are transformed into spike trains using a deterministic encoding approach.

Model Development A Spiking Neural Network based on Leaky Integrate-and-Fire neurons is implemented and trained using surrogate gradients. A Multilayer Perceptron is used as a baseline model for comparison.

Training Configuration

Learning Rate: 0.001 Optimizer: Adam Loss Function: Mean Squared Error Batch Size: 32 Epochs: 100

SNN-specific parameters include time steps, membrane decay, and spike threshold.

Results

The Spiking Neural Network achieves high prediction accuracy with performance comparable to the baseline model, while demonstrating improved computational efficiency.

Repository Structure

src contains the main Python implementation

results contains generated outputs and visualizations

Thesis.docx contains the full thesis document

requirements.txt contains all required dependencies

How to Run

Step 1: Clone the repository git clone https://github.com/RVSVarma/Thesis---Neuromorphic-Digital-Twin.git

Step 2: Move into the project directory cd Thesis---Neuromorphic-Digital-Twin

Step 3: Install dependencies pip install -r requirements.txt

Step 4: Run the pipeline python src/ndt_full_pipeline.py --data_dir "path_to_dataset"

Make sure to replace the dataset path with the actual location of the IMS dataset on your system.

Applications

Predictive maintenance Industrial monitoring Edge AI systems

Author

Venkata Skandha Rajendra Varma

Notes

This project is intended for academic and research purposes.