← Back to projects In progress

Predictive Maintenance MLOps Platform

Industrial equipment rarely fails without warning; the warning is just buried in sensor data. This project turns that data into a failure-prediction service with the full production loop around it: experiment tracking, a registry, drift monitoring and automated retraining.

View code on GitHub Live demo (coming soon)

Why this project

Before moving into data, I worked as a mechanical engineer, and I later spent nine years building failure-prediction models for engineering teams as a data analyst. Unplanned downtime is expensive and mostly preventable. Most ML portfolio projects stop at a notebook with a good F1 score; in industry the hard part starts after that, when the model has to run reliably as data shifts underneath it.

What it does

Trains gradient-boosted and deep-learning models (XGBoost, PyTorch) to predict remaining useful life and failure risk on the NASA C-MAPSS turbofan dataset
Serves predictions through a FastAPI service with input validation and streaming-friendly batch endpoints
Tracks every experiment in MLflow, with a model registry gating what reaches production
Monitors live inputs for data drift with Evidently and triggers automated retraining when drift crosses a threshold
Ships with tests, GitHub Actions CI/CD and a Docker Compose stack that starts the whole platform with one command

Architecture

IngestionSensor data loading and validation (C-MAPSS run-to-failure cycles)

↓

Feature pipelineRolling-window statistics, degradation indicators, train/test splits by engine unit

↓

Training pipelineXGBoost baseline and PyTorch model, every run logged to MLflow

↓

Model registryMLflow registry; only evaluated, approved models get promoted

↓

Inference APIFastAPI service, Dockerised, deployed with CI/CD

↓

MonitoringEvidently drift reports; drift beyond threshold triggers the retraining pipeline

Results

Headline numbers land here as the build progresses: model metrics against the published C-MAPSS benchmarks, P95 API latency, test coverage and a drift-detection walkthrough. Technical outcomes only, nothing that cannot be reproduced from the repo.

Skills demonstrated

Classical ML and deep learning on time-series sensor data
MLOps: experiment tracking, model registry, drift monitoring, automated retraining
API engineering with FastAPI and Pydantic
Docker, CI/CD with GitHub Actions, cloud deployment
Domain knowledge: a decade around industrial equipment and maintenance data