How industrial-edge/reference-architecture-for-industrial-ai-on-azure works
This repository provides an Azure-oriented Industrial AI MLOps reference implementation, with image-classification and state-identifier workflows. The main path is pipeline preparation and submission, data preparation, model training and scoring, then model registration; separate packaging and delivery pipelines support edge deployment. The graph includes the documented Azure/Industrial AI context and sampled runtime modules. Some component wiring and deployment details are not shown in the excerpts, so those relationships are omitted rather than inferred.
Main entry points: pipeline/mlops_pipeline.py Image pipeline · pipeline/mlops_pipeline.py State pipeline · prepare_execute.py Prepare and execute · packaging_pipeline.py Packaging pipeline
Components 29
Pipeline orchestration
- Prepare and execute
mlops/common/pipeline/prepare_execute.py - Compute selection
mlops/common/pipeline/get_compute.py - Environment selection
mlops/common/pipeline/get_environment.py - Job execution
mlops/common/pipeline/pipeline_execution_utils.py
Image classification
- Image pipeline
mlops/image_classification/pipeline/mlops_pipeline.py - Image data asset
mlops/image_classification/src/data_asset_creation/data_asset_creation.py - Image preparation
mlops/image_classification/src/prep/prep.py - Image training
mlops/image_classification/src/train/train.py - Image scoring
mlops/image_classification/src/score/score.py - Image registration
mlops/image_classification/src/register/register.py - Image packaging
mlops/image_classification/src/package/packaging.py - Vision payloads
mlops/image_classification/src/package/payload.py
State identification
- State pipeline
mlops/state_identifier/pipeline/mlops_pipeline.py - State data asset
mlops/state_identifier/src/data_asset_creation/data_asset_creation.py - State preparation
mlops/state_identifier/src/prep/prep.py - State training
mlops/state_identifier/src/train/train.py - Feature pipeline
mlops/state_identifier/src/si/pipeline.py - Preprocessing transforms
mlops/state_identifier/src/si/preprocessing.py - State scoring
mlops/state_identifier/src/score/score.py - Scoring utilities
mlops/state_identifier/src/score/scoring_utils.py - State packaging
mlops/state_identifier/src/package/packaging.py
Packaging and delivery
- Packaging pipeline
mlops/common/pipeline/packaging_pipeline.py - Package validation
mlops/common/src/package_validation.py - Package registration
mlops/common/src/package_registration.py - Delivery pipeline
mlops/common/pipeline/delivery_pipeline.py - Model Manager delivery
mlops/common/src/package_delivery_mm.py
Other components
- MLOps operator
- Azure Machine Learning
- Industrial AI Model Manager
Connections 31
- MLOps operator → Prepare and execute: starts workflow
mlops/common/pipeline/prepare_execute.py - Prepare and execute → Compute selection: selects compute
mlops/common/pipeline/prepare_execute.py - Prepare and execute → Environment selection: selects environment
mlops/common/pipeline/prepare_execute.py - Prepare and execute → Job execution: submits job
mlops/common/pipeline/prepare_execute.py - Job execution → Azure Machine Learning: creates job
mlops/common/pipeline/pipeline_execution_utils.py - Image pipeline → Prepare and execute: uses orchestration
mlops/image_classification/pipeline/mlops_pipeline.py - State pipeline → Prepare and execute: uses orchestration
mlops/state_identifier/pipeline/mlops_pipeline.py - Image pipeline → Image data asset: prepares data
- Image pipeline → Image preparation: loads preparation
- Image pipeline → Image training: loads training
- Image pipeline → Image scoring: loads scoring
- Image pipeline → Image registration: loads registration
- Image data asset → Image preparation: calls preparation
mlops/image_classification/src/data_asset_creation/data_asset_creation.py - Image pipeline → Azure Machine Learning: constructs job
mlops/image_classification/pipeline/mlops_pipeline.py - Image packaging → Vision payloads: uses payload handling
- State pipeline → State preparation: loads preparation
- State pipeline → State training: loads training
- State pipeline → State scoring: loads scoring
- State pipeline → State data asset: prepares data
- State pipeline → Azure Machine Learning: constructs job
mlops/state_identifier/pipeline/mlops_pipeline.py - State training → Feature pipeline: uses pipeline
mlops/state_identifier/src/train/train.py - State training → Preprocessing transforms: uses transforms
mlops/state_identifier/src/train/train.py - State scoring → Scoring utilities: uses utilities
mlops/state_identifier/src/score/score.py - State scoring → Preprocessing transforms: uses preprocessing
mlops/state_identifier/src/score/score.py - State packaging → Scoring utilities: uses utilities
- State packaging → Preprocessing transforms: uses preprocessing
- Packaging pipeline → Azure Machine Learning: builds pipeline
- Packaging pipeline → Package validation: validates package
- Packaging pipeline → Package registration: registers package
mlops/common/pipeline/packaging_pipeline.py - Delivery pipeline → Model Manager delivery: dispatches delivery
- Model Manager delivery → Industrial AI Model Manager: sends delivery messages