Torque2Pixels: Time Series Anomaly Detection & Clustering for Wire Cutting

This project focuses on analyzing and classifying anomalies in a wire-cutting process for electric motor manufacturing at Bosch. It utilizes advanced signal processing and machine learning techniques to transform torque-time data into visual representations for deep learning-based classification.

Key Features:

• Time series data analysis using Continuous Wavelet Transform (CWT) and Short-Time Fourier Transform (STFT)
• Feature extraction via wavelet scalograms and spectrograms
• Deep learning classification using ResNet50 architecture
• Unsupervised clustering with K-Means and Mean Shift algorithms
• Anomaly detection in industrial cutting processes

Technologies Used:

• Python
• Scipy
• Scikit-learn
• PyTorch
• Librosa (for audio processing techniques applied to time series data)

This repository contains the code, data preprocessing pipelines, model architectures, and analysis scripts used in the project. It serves as a comprehensive resource for time series analysis and anomaly detection in industrial processes.

Future Work:

• Implement ensemble methods for multi-phase anomaly detection
• Explore advanced neural network architectures (e.g., Vision Transformers)
• Optimize feature extraction for different phases of the cutting process

For more details, please refer to the full project report in the repository.