Export OCPM semantic model into Fabric workspace

The new optimized DirectLake semantic model leads to faster and more memory-efficient analysis of processes and is now available also for object centric processes.

Semantic model description

When you publish a process to Fabric workspace, it creates a new DirectLake semantic model and an empty Power BI report connected to the semantic model. The semantic model is created on top of Fabric Lakehouse delta tables.

The following screenshot is an example of a semantic model structure published to Fabric.

Column naming

The naming of semantic model columns corresponds to the naming of the columns in your ingested dataset, including spaces and other special characters. The naming is subject to Power BI semantic model limitations. The limitations of naming in the Fabric Lakehouse delta tables, however, are larger (for example, spaces aren't allowed). Power Automate Process Mining performs validation and sanitization before the export to Fabric workspace is triggered.

Allowed characters include:

• a-z: lowercase English letters
• A-Z: uppercase English letters
• 0-9: digits
• _: underscore

All other characters are replaced with an underscore (_) as part of the sanitization process.

This process might result in rare situations where export isn't successful, as the ingested data source includes two columns that result in identical names after sanitization—Customer_Name and Customer Name. The export is interrupted and user is notified with the specific error message.

Lakehouse delta table columns therefore use the sanitized column names, whereas semantic model columns use the original column names.

Relationships

The published data model includes predefined relationships that handle filtering and interconnectivity of visuals. You don't need to manually create more relationships unless you connect other data sources. For this scenario, use the Power BI composite semantic model and build relationships on top of that model.

Some relationships are hidden in the model. This limitation exists because only one relationship between the same two tables can be active in semantic models. You can use inactive relationships by using the corresponding DAX functions.

Data model summary

From a logical perspective, the data model consists of:

• Object and Event Data: Data related to objects and events.
• Visuals data: Data related to summarizing items derived from objects and events, usually used for visualization purposes.
• Process execution data: Data related to identified process executions (process instances) based on a leading object.

The following sections describe the subsets and included entities.

Object and Event Data

The content of process data entities changes when process model data is refreshed.

When you work with these entities, you can:

• Access the raw process data.
• Access the measures calculated based on the process data.

Entity	Description
Object Types	List of all types of objects in the process.
Objects	List of all objects in the process with reference to object type, lifecycle of the object, and relevant object level attributes as additional columns (when ingested).
Events	Entity holds list of all events in the process with reference to particular node in the Nodes table and ingested event level attributes as additional columns.
EventObjectRelations	Entity holds all relationships between Events and Objects “touched” by the execution of an event.

Visuals data

Visuals data entities are recalculated only when there's a data refresh for the process model. They include items derived from the raw data by process mining algorithms and usually serve visualization purposes.

Entity	Description
Mining Attributes	Holds values of available mining attributes - perspectives to look at the objects. By default the entity holds the values of Activity attribute.
Edges	Entity holds all edges identified between nodes with reference to particular object, nodes and including the relevant calculated measures.
Nodes	Entity holds all nodes with reference to specific object and including the relevant calculated measures.

Process execution data

Process execution data entities are calculated as process instances of interconnected objects based on a leading object type.

Entity	Description
ProcessExecutions	List of process executions, given a leading object type including throughput time of the whole execution and start and end time of the process execution instance.
ProcessExecutionObjects	Object referenced by a particular process execution.

Power BI composite model

Use the Power BI composite model on top of the semantic model published by Power Automate Process Mining and create the necessary modifications for these scenarios:

• You need to create more data sources.
• You need to create more entities.
• You need to create more relationships.
• You need to create more custom DAX (Data Analysis Expressions) queries.

Important

The semantic model uses DirectLake access mode, but its option is set to Automatic. This setting means that using nonoptimal DAX queries or incorrectly setting a composite model might result in fallback to DirectQuery mode. This setting means that your report doesn't break, but you might experience lower performance.

Learn more about creating Power BI composite data models on top of DirectLake semantic models in Building a composite models on a semantic model or model.

Semantic model refresh

By default, Power Automate Process Mining automatically keeps the semantic model up to date.

For large datasets, data refresh of underlying tables in OneLake might take longer. This delay can cause potential inconsistencies in the report. Although there's eventual consistency at the end of data refresh (semantic model is explicitly refreshed), you might want to remove the potential intermediate inconsistencies by turning off the Keep your Direct Lake data up to date flag in the Settings screen of the semantic model.

Before you update this screen, you need to take ownership of the semantic model by selecting Take over at the top of the Settings screen.