Here's the gist of how I'm doing things and what kinds of inference/modeling are enabled
Almost everything we are using falls under OWL Lite with the exception of the owl:hasValue
property which we use to define a 1-1 mapping between sets of tags and names of classes.
Because of this, the current solution falls under OWL DL.
This also means that we need a reasoner, but thankfully not one that supports OWL Full :). OWL-RL is a nice choice that works with RDFlib.
# G is our RDFlib Graph
# apply reasoning to expand all of the inferred triples
import owlrl
owlrl.DeductiveClosure(owlrl.OWLRL_Semantics).expand(G)
# get namespaces
print(list(G.namespaces()))
G.query("SELECT ?x WHERE { ?x brick:hasTag tag:Equip }")
We can serialize the expanded form of the graph to disk if we need to use a SPARQL query processor that does not support reasoning.
- The Brick class namespace is
https://brickschema.org/schema/1.1.0/Brick#
- Classes belong to
owl:Class
and are arranged into a hierarchy withrdfs:subClassOf
- Equivalent classes (the members of the classes are the same) are related with the
owl:equivalentClass
property - Definitions given with
skos:definition
- We are eliminating equipment-flavored classes where it makes sense
- e.g.
brick:AHU_Average_Exhaust_Air_Static_Pressure_Sensor
is just aAverage_Exhaust_Air_Static_Pressure_Sensor
that is a point of an AHU.
- e.g.
- Classes are equivalent to a set of tags (see below)
The root classes we have defined are:
Equipment
Location
Point
Tag
Substance
Quantity
(Relationships are the Brick term for owl ObjectProperties between instances of classes)
At the surface level, relationship work the same as they did in the original Brick.
All the same relationships still exist (where I remembered to define them), and they have their
inverses defined using owl:inverseOf
.
Domains and ranges are defined in terms of classes. Stating that the rdfs:range
of a relationship
is of class brick:Equipment
means that the object of the relationship should be an instance of the
brick:Equipment
class.
This prototype includes sub-relationships in addition to relationships.
Sub-relationships can be used in place of the super-relationship to add more detail to the nature of that relationship.
The only example so far is feedsAir
being a subproperty of feeds
.
Something to figure out is how we could infer the feedsAir
relationship; maybe if the two endpoint equipment have the air
tag and a feeds
relationship?
This may be something that needs to be explicitly specified rather than inferred.
- Tag ontology namespace is
https://brickschema.org/schema/1.1.0/BrickTag#
- We use Haystack tags and define our own set including them
- Tags should have definitions, but this is not included yet
- Sets of tags have a 1-1 mapping with a class name
- definitions given using the
skos:definition
property
This is accomplished by declaring a Brick class (e.g. Air_Temperature_Sensor
) as equivalent to an anonymous class, which is an owl:Restriction
that is the intersection of entities that have certain tags.
# in turtle format
brick:Temperature_Sensor a owl:Class ;
rdfs:subClassOf brick:Sensor ;
owl:equivalentClass [ owl:intersectionOf (
[ a owl:Restriction ;
owl:hasValue tag:Sensor ;
owl:onProperty brick:hasTag
]
[ a owl:Restriction ;
owl:hasValue tag:Temperature ;
owl:onProperty brick:hasTag
]
) ] .
The first owl:Restriction
is the set of all classes that have tag:Sensor
as the value for one of their brick:hasTag
properties.
This means that a temperature sensor :ts1
could be defined in two different ways and the reasoner would infer the other triples:
# using classes
:ts1 a brick:Temperature_Sensor
# using tags
:ts1 brick:hasTag tag:Temp
:ts1 brick:hasTag tag:Sensor
Brick now defines a hierarchy of substances (substances.py
) and a hierarchy of quantities (quantities.py
).
Substances and quantities can be related to equipment and points.
Not all of this is implemented. In the current prototype, sensors are related to substances and quantities
through the brick:measures
relationship.
:ts1 a brick:Temperature_Sensor
:ts1 brick:measures :Air
# this implies the following
:ts1 a brick:Air_Temperature_Sensor
We can further subclass substances to provide system- or process-level context to their definitions:
:ts1 a brick:Sensor
:ts1 brick:measures brick:Return_Air
:ts1 brick:measures brick:Temperature
# implies...
:ts1 a brick:Return_Air_Temperature_Sensor
Brick uses OWL restrictions to refine classes based on such relationships.
For this example, because :ts1
measures Air (specifically the brick:Air
class), OWL infers our sensor as a brick:Air_Temperature_Sensor
.
Here's what that the definition looks like in turtle:
brick:Air_Temperature_Sensor a owl:Class ;
rdfs:subClassOf brick:Temperature_Sensor ;
owl:equivalentClass [ owl:intersectionOf ( [ a owl:Restriction ;
owl:hasValue brick:Temperature ;
owl:onProperty brick:measures ] [ a owl:Restriction ;
owl:hasValue brick:Air ;
owl:onProperty brick:measures ] ) ] .
Note: we are using classes as values here, which is different than the rest of Brick. This is called "punning". This is to avoid having to create instances of substances for our sensors to measure and so on, but reserves the possibility to implement this in the future. Instances of substances can model regions/chunks of "stuff" in a stage of a process, e.g. the water entering a chiller or the mixed air region of an air handling unit.
Rather than getting lost in the Sisyphean bikeshedding of how to format everything as YAML, we're just using Python dictionaries so we don't have to worry about any (well, not that much) parsing logic.
definitions = {
"Lighting_System": {
"tagvalues": [ # Lighting_System class is equivalent to the Lighting tag
(BRICK.hasTag, TAG.Lighting),
# if you have more required tags add them as their own tuple in the list
],
# defining subclasses. This can be nested ad-infinitum
"subclasses": {
"Lighting": {
"subclasses": {
"Luminaire": {},
"Luminaire_Driver": {},
},
},
"Interface": {
"subclasses": {
"Switch": {
"subclasses": {
"Dimmer": {},
},
},
"Touchpanel": {},
},
},
},
}
}
define_subclasses(definitions, BRICK.Equipment)
For now, the code is the documentation. Look at equipment.py
, point.py
, etc for examples and how to add to each of the class hierarchies.
- pull all tags from haystack
- add Tag definitions
- add alarms, other setpoints and the rest of the Brick points
Authors: Gabe Fierro