An example of Digital Twins Architecture – Azure Digital Twins

Introduction

In this post, I discussed the architecture of digital twins. This is a relatively new and emerging topic, and here I will look at Azure Digital Twins as an example to examine this architecture in depth. The architecture described below is from a cloud perspective

It is created by integrating other existing cloud products 

I expect that all providers will adopt this strategy of integration, as illustrated below:

Architecture

The steps are

• Model any environment, and bring digital twins to life in a scalable and secure manner
• Connect assets such as IoT devices and existing business systems, using a robust event system to build dynamic business logic and data processing
• Query the live execution environment to extract real-time insights from your twin graph
• Build connected 3D visualizations of your environment that display business logic and twin data in context
• Query historized environment data and integrate with other Azure data, analytics, and AI services to better track the past and predict the future

Source: Azure Digital Twins overview

Then, model the twins and determine the relationships between twins through twin graphs.:

You can create and query relationships through the twin graph. Integration with other services is through event routes. Event routes are an example of an event-driven architecture.  To send digital twin data to other Azure services or ultimately outside of Azure, you can create event routes

For example

• Connect Azure Digital Twins to Time Series Insights to track time series history of each twin
• Store Azure Digital Twins data in Azure Data Lake
• Analyze Azure Digital Twins data with Azure Synapse Analytics or other Microsoft data analytics tools
• Integrate larger workflows with Logic Apps​
• Send data to custom applications for flexible and customized actions

Analysis

While the above is comprehensive, in my view, it is one pipeline for creating the digital twin ie a design pipeline (the other two being the simulation pipeline and the deployment pipeline)

The actual deployment would include two additional critical parts – the feedback loop and the integration

In that sense, the overall architecture of a digital twin can be seen as three parts

1. Design pipeline 
2. simulation pipeline and 
3. deployment pipeline 

In terms of the design patterns, there are three possibilities

1. Time series
2. Event-driven
3. Graph

We have described the event-driven architecture above

But in reality, data may be acquired as a time series or as a graph

Note that by graphs, I mean programming constructs that operate directly on a graph structure comprising edges and nodes ex a graph neural network.

An example of a Graph driven digital twin is A Graph Neural Network-Based Digital Twin for Network Slicing Management.

Image source – Microsoft