Released · Portfolio Demo

Mohammad Ahmadian

Production ML Platform for Predictive Maintenance

Production-oriented machine failure risk scoring — leakage-safe features, baseline vs LightGBM, chronological evaluation, FastAPI inference.

Open Performance Dashboard Architecture GitHub README

What this system does

Scores machine failure risk from industrial sensor features on the public AI4I 2020 benchmark, with train/serve separation and documented evaluation.

AI4I CSV→ Validate→ Features→ Train / Eval→ Artifacts→ FastAPI

Test-set performance

Chronological split by UDI · threshold tuned on validation F1 · model version 20260726T111950Z

Primary F1
0.800
LightGBM
Baseline F1
0.222
Logistic Regression
ROC-AUC
0.944
Primary · test
PR-AUC
0.781
Imbalanced failures

Architecture

Figure 1. System architecture (Released). Source: presentation/architecture.

Architecture diagram for Production ML Platform for Predictive Maintenance

Integrity controls

  • Failure-mode flags (TWF, HDF, PWF, OSF, RNF) excluded from features
  • Chronological split — no random shuffle across UDI order
  • Baseline always reported next to primary
  • Synthetic dataset — limitations documented

Predict contract

{
  "entity_id": "L55686",
  "features": { "air_temperature_k": 298.4, "..." : "..." }
}

Full sample: reports/sample_predict_payload.json

Run locally

pip install -r requirements.txt
set PYTHONPATH=src
uvicorn predmaint.api.main:app --host 0.0.0.0 --port 8000
# http://localhost:8000/docs