Architecture Diagram
Conveyor Maintenance Dashboard 1
Conveyor Maintenance Dashboard 2

Inspiration

We were inspired by the high costs and inefficiencies caused by unplanned conveyor belt failures in manufacturing. The idea of using AI and IoT to predict and prevent these failures—before they disrupt operations—motivated us to build a smarter, automated solution.

What We Learned

How to ingest realistic IoT sensor data for industrial equipment.
The importance of feature engineering (statistical and spectral) in predictive maintenance.
Orchestrating multi-layered cloud architectures for scalability and automation.
Integrating AI reasoning with real-world ticketing and scheduling systems.

How We Built the Project

We designed a five-layer pipeline:

1. Data Ingestion Layer

Used AWS Lambda to generate IoT data (vibration, temperature, current) every minute, publishing it via AWS IoT Core to Amazon S3.

2. Feature Extraction Layer

Lambda functions or SageMaker jobs computed features like RMS, mean, kurtosis, and FFT amplitudes for each data window.

3. Model Training & Inference Layer

Amazon SageMaker hosted ML models for anomaly detection and fault prediction, with retraining triggered by drift.

4. AI Reasoning & Decision Layer

AWS Bedrock AgentCore interpreted model results, applied logic (e.g., if (anomaly > 0.8) trigger alert), and generated insights or maintenance tickets.

5. Smart Scheduling & Visualization Layer

Integrated agents scheduled alerts/tickets, and React App dashboards visualized equipment health, anomalies, and maintenance timelines.

Challenges Faced

Ingesting data for real world anomaly detection and fault prediction
Ensuring feature extraction captured both statistical and spectral machine health indicators.
Ability to train a model for anomaly detection and fault prediction
Balancing automation with interpretability in AI-driven decisions.