Manufacturing semantic layer on Azure Databricks
Create one governed business language across production, inventory, quality, delivery and margin so Power BI, SQL and AI use the same grain, dimensions and KPI definitions.
Shared manufacturing context
One business language
Keep facts separate by grain, then standardize the business meaning shared across them.
OEE · yield
stock · shelf life
OTIF · fill rate
Different interfaces can present the answer differently. The governed business meaning should stay the same.
Same question
How much stock is available for production?
WMS report
Physical stock before exclusions
5,440 kg
Spreadsheet
Manual exclusions for blocked and quarantine stock
4,980 kg
Power BI
Report logic with its own filter interpretation
4,820 kg
Governed answer
4,820 kg available
Semantic controls
- grain: site × location × batch × SKU
- unit: base UOM
- owner: supply chain
Illustrative values. The goal is not one dashboard; it is one governed definition that every consumer can reuse.
Why semantics
The same business question can still produce different answers
Reports can use the same source data and still disagree when grain, exclusions, quality status, units or effective dates are defined independently in every consumer.
Grain determines what one row or measure actually represents.
Exclusions decide which states, lots or transactions count.
Units and time must be consistent before aggregation.
Ownership makes the definition reviewable and auditable.
Architecture boundary
A semantic layer defines business meaning. It does not fix the underlying data.
Identity, history, units and business process grain need to be reliable before data reaches the semantic layer. The semantic layer then gives teams a consistent way to interpret that data and calculate shared business metrics.
Governed data products
Reliable business data
Semantic layer
Shared business meaning
Separate facts by business process; reuse dimensions across them.
Manufacturing context
Define the grain before you define the metric
Manufacturing systems describe different business processes at different grains. The semantic model should make those grains explicit before facts are connected through shared dimensions.
Production run is not the same grain as a production order or machine event.
Inventory batch is not the same grain as an inventory movement.
Quality sample is not the same grain as a batch release decision.
Shipment is not the same grain as a customer order line.
Business model
Reuse dimensions without mixing business grains
A good manufacturing semantic model does not become one giant factory table. Keep facts separate by process and connect them through governed shared dimensions.
Shared dimensions
site · product · time · customer
Separate facts
production · inventory · delivery
Explicit paths
only valid business relationships
A Product dimension can be reused across production, inventory, delivery and margin.
Production and inventory facts keep their own natural grain.
Cross-process analysis uses governed dimensions instead of fact-to-fact joins.
One set of dimensions, multiple process-specific data products.
Shared dimensions
Production performance
run × line
Yield & material loss
run × material
Inventory & shelf life
batch × location
Delivery & OTIF
order line × shipment
Margin
order line × product/customer
Do not connect facts directly just because they share a date or product code. Reuse governed dimensions and explicit relationships.
Governed source
production_run_fact
Slice governed measures
The measure stays the same while the business view changes.
Fields
Good output
Yield
Downtime
OEE
Measure logic lives in one place
Reports and analytics can group the same governed measure by site, line, product, shift or date without rebuilding the KPI logic.
Unity Catalog Metric Views
Define measures once, then slice them by governed fields
Metric Views centralize reusable measures while keeping fields such as site, line, product, recipe, shift and date available for grouping and filtering.
Reusable measures
Standardize KPI logic once instead of recreating it in every report.
Governed fields
Expose the dimensions people need to analyse the metric consistently.
Explicit relationships
Model relationships according to the actual business cardinality.
Business definitions
A KPI formula is not the complete business definition
Reusable semantics need scope, grain, exclusions, units, ownership and effective dates in addition to the calculation itself.
Business concept
Available Stock
Metrics and terminology are related, but different
Metric Views
Govern reusable measures, fields and semantic relationships.
Business definitions
Add the context people need to interpret a concept correctly.
Certification
Signals which shared assets the organization has approved for reuse.
Operating model
Govern semantic definitions like data products
A shared definition needs ownership, reconciliation and change control. Otherwise the semantic layer becomes another uncontrolled metadata catalog.
Business owner approves meaning, scope and exclusions.
Known operational cases validate the definition before reuse.
Effective dates and versioning preserve historical meaning.
Treat shared meaning as something that is reviewed and maintained.
Define
Agree formula, grain, scope, exclusions, unit and owner.
Validate
Reconcile against known operational cases.
Certify
Mark the approved shared definition as trusted for reuse.
Publish
Expose the data product, metric or governed business concept.
Change
Review version and effective-date impact before changing logic.
Deprecate
Retire outdated definitions instead of leaving competing versions.
Common mistakes
Avoid semantic models that hide business ambiguity
Most semantic-layer failures come from mixing grains, duplicating KPI logic or trying to solve structural data-quality problems too late.
Avoid
KPI logic inside every Power BI model
Better default
Define reusable business measures upstream
Avoid
One giant factory semantic model
Better default
Separate business processes by grain
Avoid
Joining facts directly to other facts
Better default
Use shared dimensions and explicit relationships
Avoid
Using semantics to repair dirty data
Better default
Fix identity and conformance in Silver or Gold
Avoid
A KPI without a business owner
Better default
Assign an owner and technical steward
Avoid
Changing formulas without effective dates
Better default
Version definitions and manage history
FAQ
Frequently asked questions
Practical answers about manufacturing semantic layers on Azure Databricks.
What is a manufacturing semantic layer?
A manufacturing semantic layer is the governed business-meaning layer above trusted data products. It standardizes business terms, relationships, fields and measures so concepts such as OEE, yield, available stock, OTIF and margin are interpreted consistently across reporting, analytics and AI.
Why is grain important in a manufacturing semantic model?
Manufacturing processes operate at different grains. Production runs, machine events, inventory movements, batches, sales order lines and shipments describe different events. The grain should be explicit before metrics and relationships are defined, otherwise joins and aggregations can produce plausible but incorrect results.
Is a semantic layer the same as a Power BI semantic model?
Not necessarily. Power BI semantic models remain useful for report-specific presentation and modeling, but a platform semantic layer can centralize shared business meaning upstream so Power BI, SQL and AI consumers start from the same governed data products and definitions.
Where do Databricks Metric Views fit?
Unity Catalog Metric Views provide a centralized way to define reusable measures, fields and semantic relationships on top of governed data products. They are an important implementation component of Databricks semantics, but they do not replace data quality, conformance or good fact-and-dimension design underneath.
Should all business logic move to Metric Views?
No. Identity resolution, row-level transformations, historical corrections, conformance and data-quality logic should remain in the appropriate Silver or Gold data products. Metric Views should standardize reusable business measures and fields rather than repair structural data problems.
Next step
Make manufacturing definitions reusable across BI and AI
Start with one business process, define its grain and critical measures, then publish a governed semantic contract that reporting, analytics and AI can reuse.
1. Pick the process
Production, inventory, quality, delivery or margin.
2. Define the grain
Agree what one row and one measure represent.
3. Govern the meaning
Definition, exclusions, units, owner and effective date.
4. Reuse it
Power BI, SQL, analytics and Ask Titan.