PumpGuard: AI-Powered Predictive Maintenance

PumpGuard is an intelligent industrial monitoring prototype designed to help identify developing pump faults before they escalate into costly equipment failures. Built for the AMD Developer Hackathon ACT III, it combines synthetic telemetry simulation, explainable rule-based diagnostics, machine learning, and an interactive 3D monitoring dashboard. The platform simulates evolving operating conditions across five scenarios: normal operation, overheating, bearing wear, cavitation, and imbalance. It analyzes telemetry such as temperature, vibration, RPM, pressure, and other operational indicators to assess equipment health, detect abnormal trends, identify potential faults, and recommend maintenance checks. PumpGuard's FastAPI backend provides validated telemetry, health assessments, historical readings, and diagnostic endpoints. A React, TypeScript, and Three.js frontend presents live simulated data, component-level alerts, telemetry trends, and visual representations of pump conditions. Its telemetry analysis service examines rolling statistics, rates of change, and threshold excursions to help explain why a condition may be deteriorating. A machine-learning pipeline and compact neural-network classifier complement the rule-based detection system. Optional AMD Token Factory integration supports AI-assisted diagnostic explanations, while deterministic rules and strict evidence-grounding checks help keep assessments tied to available telemetry. PumpGuard is designed as an offline-capable prototype with a modular architecture that can support future integration with real industrial sensors and maintenance workflows. Its current demonstrations use synthetic data, so real-world accuracy and deployment readiness require further validation with physical equipment and field data.