Time Series Model in Azure Time Series Insights Gen2

This article describes Time Series Model, the capabilities, and how to start building and updating your own models in the Azure Time Series Insights Gen2 environment.

Summary

Traditionally, the data that's collected from IoT devices lacks contextual information, which makes it difficult to find and analyze sensors quickly. The main motivation for Time Series Model is to simplify finding and analyzing IoT or Time Series data. It achieves this objective by enabling the curation, maintenance, and enrichment of time series data to help prepare consumer-ready datasets for analytics.

Scenario: Contoso's new smart oven

Consider the fictitious scenario of a Contoso smart oven. In this scenario, suppose that each Contoso smart oven has five temperature sensors, one for each of four top burners and one for the oven itself. Until recently, each Contoso temperature sensor sent, stored, and visualized its data individually. For its kitchen appliance monitoring, Contoso relied on basic charts, one for each individual sensor.

While Contoso was satisfied with its initial data and visualization solution, several limitations became apparent:

• Customers wanted to know how hot the overall oven would get when most of the top burners were on. Contoso had more difficulty analyzing and presenting a unified answer about the conditions of the overall oven.
• Contoso engineers wanted to verify that the top burners being run simultaneously wouldn't result in inefficient power draw. There was difficulty cross-referencing which temperature and voltage sensors were associated with each other and how to locate them in the store.
• The Contoso quality assurance team wanted to audit and compare the history between two sensor versions. There was difficulty determining what data belonged to which sensor version.

Without the ability to structure, organize, and define the overarching smart oven time series model, each temperature sensor maintained dislocated, isolated, and less informative data points. Turning these data points into actionable insights was more difficult since each dataset lived independently of the others.

These limitations revealed the importance of smart data aggregation and visualization tools to accompany Contoso's new oven:

• Data visualization proves useful when you're able to associate and combine data into a convenient view. An example is showing voltage sensors along with temperature sensors.
• Managing multidimensional data for several entities along with comparison, zooming, and time range functionalities can be difficult to accomplish.

Time Series Model provides a convenient solution for many of the scenarios encountered in this fictitious example:

• Time Series Model plays a vital role in queries and navigation because it contextualizes data by allowing comparisons to be drawn across time ranges and between sensor and device kinds. (A)
• Data is further contextualized because data persisted in Time Series Model preserves time series query computations as variables and reuses them at query time.
• Time Series Model organizes and aggregates data for improved visualization and management capabilities. (B)

Key capabilities

With the goal to make it simple and effortless to manage time series contextualization, Time Series Model enables the following capabilities in Azure Time Series Insights Gen2. It helps you:

• Author and manage computations or formulas leveraging scalar functions, aggregate operations, and so on.
• Define parent-child relationships to enable navigation, search, and reference.
• Define properties that are associated with the instances, defined as instance fields, and use them to create hierarchies.

Components

Time Series Model has three core components:

• Time Series Model instances
• Time Series Model hierarchies
• Time Series Model types

These components are combined to specify a time series model and to organize your data.

A time series model can be created and managed through the Azure Time Series Insights Explorer. Time Series Model settings can be managed through the Model Settings API.

Time Series Model instances

Time Series Model instances are virtual representations of the time series themselves.

In most cases, instances are uniquely identified by deviceId or assetId, which are saved as time series IDs.

Instances have descriptive information associated with them called instance properties, such as a time series ID, type, name, description, hierarchies, and instance fields. At a minimum, instance properties include hierarchy information.

Instance fields are a collection of descriptive information that can include values for hierarchy levels, as well as manufacturer, operator, and so on.

After an event source is configured for the Azure Time Series Insights Gen2 environment, instances are automatically discovered and created in a time series model. The instances can be created or updated via the Azure Time Series Insights Explorer by using Time Series Model queries.

The Contoso Wind Farm demo provides several live instance examples.

Instance properties

Instances are defined by timeSeriesId, typeId, name, description, hierarchyIds, and instanceFields. Each instance maps to only one type, and one or more hierarchies.

Instances have the following JSON representation:

{
  "timeSeriesId": ["PU2"],
  "typeId": "545314a5-7166-4b90-abb9-fd93966fa39b",
  "hierarchyIds": ["95f0a8d1-a3ef-4549-b4b3-f138856b3a12"],
  "description": "Pump #2",
  "instanceFields": {
    "Location": "Redmond",
    "Fleet": "Fleet 5",
    "Unit": "Pump Unit 3",
    "Manufacturer": "Contoso",
    "ScalePres": "0.54",
    "scaleTemp": "0.54"
  }
}

Time Series Model hierarchies

Time Series Model hierarchies organize instances by specifying property names and their relationships.

You can configure multiple hierarchies in a given Azure Time Series Insights Gen2 environment. A Time Series Model instance can map to a single hierarchy or multiple hierarchies (many-to-many relationship).

The Contoso Wind Farm demo displays a standard instance and type hierarchy.

Hierarchy definition

Hierarchies are defined by hierarchy id, name, and source.

Hierarchies are represented in JSON as:

{
  "hierarchies": [
    {
      "id": "6e292e54-9a26-4be1-9034-607d71492707",
      "name": "Location",
      "source": {
        "instanceFieldNames": [
          "state",
          "city"
        ]
      }
    },
    {
      "id": "a28fd14c-6b98-4ab5-9301-3840f142d30e",
      "name": "ManufactureDate",
      "source": {
        "instanceFieldNames": [
          "year",
          "month"
        ]
      }
    }
  ]
}

In the previous JSON example:

• Location defines a hierarchy with parent states and child cities. Each location can have multiple states, which in turn can have multiple cities.
• ManufactureDate defines a hierarchy with parent year and child month. Each ManufactureDate can have multiple years, which in turn can have multiple months.

Hierarchy example

Consider an example where hierarchy H1 has building, floor, and room as part of its instanceFieldNames definition:

{
  "id": "aaaaaa-bbbbb-ccccc-ddddd-111111",
  "name": "H1",
  "source": {
    "instanceFieldNames": [
      "building",
      "floor",
      "room"
    ]
  }
}

Given the instance fields used in the previous definition and several time series, the hierarchy attributes and values appear as shown in the following table:

Time Series ID1 and ID4 are displayed as part of hierarchy H1 in the Azure Time Series Insights Explorer because they have fully defined and correctly ordered building, floor, and room parameters.

The others are classified under Unparented Instances because they don't conform to the specified data hierarchy.

Time Series Model types

Time Series Model types help you define variables or formulas for doing computations. Types are associated with a specific instance.

A type can have one or more variables. For example, a Time Series Model instance might be of type Temperature Sensor, which consists of the variables avg temperature, min temperature, and max temperature.

The Contoso Wind Farm demo visualizes several Time Series Model types associated with their respective instances.

Type properties

Time Series Model types are defined by id, name, description, and variables.

Types conform to the following JSON example:

{
  "types": [
    {
      "id": "1be09af9-f089-4d6b-9f0b-48018b5f7393",
      "name": "DefaultType",
      "description": "Default type",
      "variables": {
        "EventCount": {
          "kind": "aggregate",
          "value": null,
          "filter": null,
          "aggregation": {
            "tsx": "count()"
          }
        },
        "Interpolated Speed": {
          "kind": "numeric",
          "value": {
              "tsx": "$event['Speed-Sensor'].Double"
          },
          "filter": null,
          "interpolation": {
              "kind": "step",
              "boundary": {
                  "span": "P1D"
              }
          },
          "aggregation": {
              "tsx": "right($value)"
          }
        }
      }
    }
  ]
}

Time Series Model types can have many variables that specify formula and computation rules on events. Read more about how to define Time Series Model variables

Next steps

• Read the Time Series Model reference documentation for more information on how to edit the model via APIs.

• Explore the formulas and computations you can create with Time Series Model variables

• Learn about querying data in Azure Time Series Insights Gen2