How aditya12333/predictive-maintenance-platform-ml works
This is a predictive-maintenance platform that turns historical engine data into governed RUL models, then scores live telemetry and exposes fleet health and alerts. The principal workflow is data preparation and model training, candidate packaging and approval, streaming ingestion and quality-aware inference, persistence, and dashboard/API consumption. The graph also includes batch orchestration and monitoring/retraining. Relationships are grounded in the README and shown source excerpts or SOURCE INDEX references; unsampled implementations are represented only at supported subsystem paths.
Main entry points: app.py Telemetry API · consumer.py Stream consumer · fleet-dashboard.tsx Fleet dashboard · train.py Model training
Components 25
Data foundation
- Dataset preparation
src/predictive_maintenance/data/prepare.py - Data validation
src/predictive_maintenance/data/validate.py - Data lineage
src/predictive_maintenance/data/manifest.py - Feature generation
src/predictive_maintenance/features/builder.py
Model lifecycle
- Model training
src/predictive_maintenance/training/train.py - Model evaluation
src/predictive_maintenance/training/evaluate.py - Candidate packaging
src/predictive_maintenance/training/package.py - Candidate registry
src/predictive_maintenance/training/registry.py - Human approval
src/predictive_maintenance/training/approval.py - Model promotion
src/predictive_maintenance/training/promotion.py
Telemetry serving
- Approved model loading
src/predictive_maintenance/inference/registry.py - Telemetry API
src/predictive_maintenance/api/app.py - Event streaming
src/predictive_maintenance/streaming/producer.py - Stream consumer
src/predictive_maintenance/streaming/consumer.py - Inference decisions
src/predictive_maintenance/inference/worker.py
Operations
- Operational persistence
src/predictive_maintenance/storage/operational.py - Alert lifecycle
src/predictive_maintenance/storage/alerts.py - Monitoring reports
src/predictive_maintenance/monitoring/report.py - Retraining policy
src/predictive_maintenance/monitoring/retraining.py - Candidate pipeline
airflow/dags/rul_training_pipeline.py - Monitoring schedule
airflow/dags/monitoring_pipeline.py
Fleet experience
- Fleet dashboard
frontend/components/fleet-dashboard.tsx - Dashboard API client
frontend/lib/api.ts
Other components
- Maintenance operator
- Telemetry producer
Connections 26
- Maintenance operator → Fleet dashboard: reviews fleet
- Telemetry producer → Telemetry API: submits telemetry
- Dataset preparation → Data validation: validates data
- Dataset preparation → Data lineage: publishes lineage
- Model training → Feature generation: uses features
- Model training → Model evaluation: compares models
- Candidate packaging → Candidate registry: registers candidate
- Candidate registry → Human approval: submits evidence
src/predictive_maintenance/training/approval.py - Human approval → Model promotion: authorizes promotion
src/predictive_maintenance/training/promotion.py - Model promotion → Approved model loading: updates champion
- Approved model loading → Candidate registry: verifies registry
src/predictive_maintenance/inference/registry.py - Telemetry API → Event streaming: publishes events
src/predictive_maintenance/api/app.py - Stream consumer → Inference decisions: processes events
src/predictive_maintenance/streaming/consumer.py - Stream consumer → Event streaming: publishes dead letters
src/predictive_maintenance/streaming/consumer.py - Inference decisions → Feature generation: builds features
src/predictive_maintenance/inference/worker.py - Stream consumer → Operational persistence: persists outcomes
- Telemetry API → Operational persistence: serves fleet data
src/predictive_maintenance/api/app.py - Telemetry API → Alert lifecycle: handles alerts
src/predictive_maintenance/api/app.py - Monitoring reports → Operational persistence: evaluates persisted data
- Monitoring schedule → Monitoring reports: runs reports
airflow/dags/monitoring_pipeline.py - Monitoring schedule → Retraining policy: checks failures
airflow/dags/monitoring_pipeline.py - Retraining policy → Candidate pipeline: requests candidate
- Candidate pipeline → Model training: orchestrates training
- Dashboard API client → Telemetry API: calls operations API
- Fleet dashboard → Dashboard API client: loads fleet data
- Fleet dashboard → Alert lifecycle: manages alerts