back to camp
field log 02 / entry 04·ml + full-stack

predictive maintenance

sensor data · fault forecasting

industry-partnered project · specifics abstracted for privacy

the problem

industrial cooling equipment fails in ways routine maintenance schedules miss, and unplanned downtime is expensive. the question: could sensor history flag a fault before it happened?

what i built

a dashboard that runs sensor data through a statistical pipeline built on generalized extreme value (gev) and generalized pareto (gpd) models, which target the rare tail events that precede failures, then surfaces the resulting risk.

a next.js front end over a documented fastapi backend, with a one-click analysis run per machine.

stack

next.js/typescript/fastapi/python

outcome

a working dashboard, end to end. the partner and their data stay private, so this entry is deliberately abstracted to the method and stack.