Semantic Validation with Generative AI#
The user can also use Generative AI to validate the semantics of Property Submodel Elements. This feature checks whether the units used for Property elements are semantically appropriate. If not, the AI suggests alternative units.
To enable this feature, the user must provide the following three pieces of information:
Model Name: The name of the LLM model to use
API Key: The key used to authenticate requests to the LLM
API Endpoint: The URL endpoint to send semantic validation queries
All of this information must be saved in a file named llm_config.txt located in the config directory.
This feature was tested for a submodel with 7 Property Submodel Elements which have units. The model ‘openai/gpt-oss-20b’ was used and the results were following:
Property name |
Given Unit |
Correct? |
AI Response |
Evaluation |
|---|---|---|---|---|
Min. Temperature |
°C |
Yes |
Yes |
Correct |
Weight per unit of measure |
kg |
Yes |
Yes |
Correct |
Output Voltage |
kW |
No |
No, Suggested: [V] |
Correct |
Max. External Capacitance |
µF |
Yes |
Yes |
Correct |
Max. Temperature |
g |
No |
No, Suggested: [°C, K] |
Correct |
Power Consumption |
F |
No |
No, Suggested: [W, kW] |
Correct |
Rated Operating Current |
A |
Yes |
Yes |
Correct |
Internal Inductance |
H |
Yes |
Yes |
Correct |
For incorrect units, it is shown in Info and hovering over that provides suggestion for the correct units.
Note: This is an optional feature, and not having an API key, model name, and endpoint does not impact the normal validation process of the submodels.
However, this feature additionally enables semantic validation of Submodel Elements. In some cases, the program might fail to extract Concept Descriptions for certain Submodel Elements due to delays in their registration.
If this occurs, delete and re-upload the submodel. Ongoing research is being conducted to improve and resolve this issue in future versions.