Overview

Developed an end-to-end IoT and machine learning solution for Webee.io to enable real-time equipment monitoring and failure prediction in industrial settings. This was part of my work with The Purdue Data Mine Learning Community.

Technical Details

Data Processing Pipeline

Engineered robust data cleaning pipeline for noisy sensor data using:
- Linear regression for trend analysis and outlier detection
- Gaussian Mixture Models (GMM) for multi-modal noise clustering
- Kalman filtering for sensor fusion and state estimation
Achieved 93% accuracy in noise reduction while preserving critical signal features

Machine Learning System

Developed hybrid anomaly detection system combining:
- One-Class SVM for novelty detection
- LSTM networks for sequence-based pattern recognition
- Achieved 83% accuracy in anomaly detection with low false positive rate
Implemented online learning to adapt to concept drift in sensor patterns
Built custom feature extraction for various sensor types (vibration, temperature, pressure)
Completely unsupervised learning process since the data was unlabeled.

System Architecture

Designed IoT product architecture using Docker and Kubernetes ensuring ease of deployment and scalability.
Built real-time alerting system with configurable thresholds and notification channels
No evaluation was done from Webee.io side but the system aligns

Poster/Report

For more details, you can view the full project report here: