Digital Buildings Abstract Model

In addition to providing a general syntax and primitives for building a smart-building ontology, Digital Buildings provides an abstract model with the types generated when modeling Google’s own buildings. This document describes the conventions used in the concrete model and provides additional explanations of individual types.1

Model Origins and Design Philosophy

The DB model was created with the intent of providing a consistent API to Google’s building portfolio for high-level analysis and control. The model is intentionally reductionist, covering only the points necessary for operations like gross fault detection, setpoint control, environmental and energy monitoring. Building a full-featured BMS in the cloud was an explicit non-goal. Instead, the model focuses on defining minimal set of fields for atomic units of functionality and composing them into equipment in consistent and easy-to-follow ways (we hope we succeeded).

Naming and Inheritance Conventions

The concrete model uses some conventions that are not [yet] encoded into the ontology structure. Due to the utility of these conventions, they will likely be added to the model in a future revision.

General Types

The first of these is the concept of a General Type. A general type is simply a broad category of types for which there may be many variations. An example might be an air handling unit (AHU) or a heat exchanger (HX). Each of these general types is instantiated as an entity type (sometimes with no fields) and inherited by any type that represents a specific variation on the class.

General types are given 1-4 character abbreviations as names (e.g., AHU, HX, VAV, etc.), and each child (entity) type begins its name with <general type>_ (e.g., VAV_SD_...).

Almost all general types are tagged as is_abstract: true and inherit from EQUIPMENT (with the exception of systems, which are themselves general types).[^2] General types can also inherit individual fields.

General types, by convention, are kept in a file called GENERALTYPES.yaml that is scoped to a particular namespace (e.g., HVAC, METERS, etc.).

Abstract Functional Groups

Abstract functional groups (aka abstract types) are sets of fields required for understanding a particular discrete device function. For example, airflow control would have a field for the flow setpoint, flow sensor, and damper command. These field sets are given short names, like SD (short for “supply damper”) and marked as is_abstract: true so they cannot be assigned directly to entities.

Functional groups, by convention, are contained in a file called ABSTRACT.yaml that is scoped to a particular namespace.

Use of optional fields

Optional fields are used for fields that provide nice-to-have information in a type. For instance, a device may have an additional sensor that is used for monitoring only. Adding these fields as optional to the general type or functional groups they’re likely to be seen with decreases the diversity of types without affecting analysis.

Type Composition and Naming

For the most part, types have one level of inheritance hierarchy below abstract types. A typical type will implement its general type and the functional group(s) that represent its functionality. Ideally, the constructed type will have no fields directly assigned to it.

Typically the type is named by the general type and its inherited functional groups with _. Functional groups are roughly ordered by significance, which is somewhat subjective; the order is not inherently meaningful, but for readability some type of ordering should be used (an alternative might be alphabetically).

Specific types are, by convention, organized into files by general type, named by that class (ex: VAV.yaml).

Specific Model Details

For further detail on types in specific namespaces see the following:

Notes

  1. Explanations are primarily for HVAC at this time, as it is one of the most complex systems to model. \ [^2]: Sometimes devices are so bespoke as to not be worth defining as canonical; however, it should be be the goal of the modeler to reduce these devices to sufficiently meaningful abstract concepts and to build the appropriate canonical representation for such combinations. ↩