Building your digital twin solution using the Digital Twin Framework on AWS

This post was contributed by Jeremiah Habets, Ross Pivovar, Pramod Daya, Pallavi Chari, and Adam Rasheed from AWS

Customers tell us that they’re increasingly seeking holistic digital twin solutions spanning IoT, spatial computing, and predictive modeling domains. Integrating these diverse technical stacks presents challenges for builders. In a prior post, we described a four-level Digital Twin leveling index to help customers understand their use cases and the technologies required for delivering real outcomes.

Customers are also asking for guidance about where to start and how to expand into additional use cases by building upon their initial deployed solution. This evolving discussion reflects the growing complexity and potential of digital twin implementations across industries.

In today’s post, we’ll describe the Digital Twin Framework (DTF) reference architecture from AWSand some common use cases against which the framework can be applied. The DTF provides the blueprint for building and scaling digital twin solutions on AWS. This framework is extensible and allows integration of your preferred components as building blocks within the overall solution architecture. More importantly, the DTF enables our customers to begin building their digital twin solution focusing on one use case, and then expand to include additional cases over time.

Digital Twin Framework

Customers starting their digital twin journey worry about significant rework when expanding beyond initial use cases. They seek a holistic blueprint for building scalable solutions. In response, we created and published the Digital Twin Framework, shown in Figure 1, to provide a comprehensive guide for scaling digital twins over time.

Figure 1 Diagram showing overall digital twin framework reference architecture.

The DTF provides for a hybrid architecture with key components on-premises at the customer site, and the relevant data, models, and processing on AWS. The DTF breaks down the complexity of digital twins into three main domains: the Spatial Data Plane, the Industrial Data Fabric, and the Simulation Compute environment.

The Spatial Data Plane ingests, transforms, and stores spatial data files as dynamic 3D assets used for visualizing the digital twin. The Visual Asset Management System (VAMS) is one example of a Spatial Data Plane. VAMS is an open-source toolkit that ingests raw data sources like lidar scans, point clouds, CAD files, video, and photos to a spatial data lake.

With either the provided React website or custom scripts, users can upload assets to Amazon Simple Storage Service (Amazon S3) via Amazon API Gateway, ensuring that assets are uploaded securely, and with appropriate metadata for version control and governance. We use spatial data pipelines to define asset transformations, converting the source assets to real-time 3D files like glTF or Universal Scene Description (USD).

The Industrial Data Fabric outlines AWS Prescriptive Guidance using a collection of solutions for collecting data from various industrial data sources into AWS where it can be securely and economically accessed, analyzed, unified and harnessed for business and operational decision making. One common implementation of the Industrial Data Fabric uses AWS IoT Greengrass and AWS IoT Sitewise Edge to connect on-premises devices to AWS IoT SiteWise. AWS IoT SiteWise provides semantics and structure to the raw device data, organizing the data into a useful form for analysis and presentation. An AWS IoT TwinMaker knowledge graph enhances this conceptual understanding by creating rich interconnected representations of industrial assets, processes, and their relationships to provide for more sophisticated analytics, and better-informed decision-making.

Simulation and AI/ML models enable digital twins to predict current unmeasured variables or potential future events. Engineers build models of physical components, systems, or processes by combining industry-standard simulation software with high-performance computing (HPC) infrastructure and real-time IoT data, creating comprehensive and dynamic digital representations.

HPC services in AWS like AWS Batch, AWS Parallel Computing Service, or AWS ParallelCluster can achieve probabilistic outputs for scenario analysis at scale without large capital costs. They can also enable optimization of digital twin outputs from simple small CPU-bound calculations to large distributed simulations on millions of core or thousands of GPU.

Furthermore, we can connect digital twins with IoT data for self-calibration using tools like TwinFlow, Amazon Managed Workflows for Apache Airflow (MWAA), or open-source Ray. These enable large-scale graph orchestration, dynamically assessing decision points and deploying workflows based on LLM judgment or rules-based criteria. Calibration can be achieved offline with historical data using an HPC service or online with TwinFlow and IoT data to ensure a digital twin adapts to a changing environment and provides the most accurate predictions possible.

Finally, we need to present users with the results of all this data ingestion, storage, transformation, analysis, and simulation. The real-time 3D visualization can be rendered inside a Grafana dashboard using the AWS IoT TwinMaker plugin, or rendered in the cloud using Nvidia Omniverse or Epic Unreal Engine – and streamed into the dashboard as webRTC video. Some advanced customers create custom interfaces for their digital twins, rendering views of the digital twin on mobile devices or in AR/VR headsets. Others are experimenting with adding generative AI chatbots into their dashboards to provide additional context and ease of use.

Digital twin use cases

In industrial settings, we have identified some broad digital twin use case categories that commonly require the DTF – Engineering and Design, Process Optimization, Integrated Asset Performance Management, and Augmented Worker.

These combine IoT, spatial computing, and predictive modeling capabilities. Figure 2 illustrates these common deployments, while Figure 3 maps them onto a Venn diagram, showing their relationship to technical domains and the L1-L4 leveling guide. We’ll explore each category in detail below.

Figure 2 Diagram showing 4 digital twin use case categories of Augmented Worker, Asset Performance Management, Production & Process Optimization, and Engineering Design.

Figure 3 Venn diagram showing 4 digital twin use case categories overlaid on the 3 core technical domains required for digital twin solutions.

Augmented Worker

These use cases aim to enhance workforce productivity and safety by providing easy access to information, guidance, and remote experts. The technical domains underpinning this use case are data connectivity along with relevant visualization using spatial computing, and AI assistants as shown in Figure 1. User experiences range from 2D dashboards to immersive 3D environments for training on Standard Operating Procedures (SOPs), and remote expert assistance.

We increasingly see the industry moving towards a “3D way of working”. For example, remote operators use VR glasses to immerse themselves in a context-aware 3D virtual environment, accessing sensor data, engineering information, and live video feeds. Maintenance technicians using AR glasses or tablets can view overlaid engineering data and service history, with the ability to share real-time views with remote experts. This visual immersion is especially helpful in complex or hazardous environments where workers can follow precise steps, troubleshoot issues and practice in a virtual environment before working in the physical site.

Furthermore, generative AI based chatbots enhance workforce experience by generating tailored responses to unexpected or unfamiliar issues, and can provide adaptive training content for skill development. Customers deploying such use cases through digital twins seek to boost safety, reduce errors, accelerate training, for a more skilled, confident, efficient, and protected workforce. We’ve made reference solutions for Augmented Worker with the DTF available in the AWS Solutions Library.

Asset Performance Management (APM)

Maintaining high value assets is a critical challenge for industrial operations, where unexpected failures lead to costly downtime. Typical reactive or schedule-based maintenance approaches are inefficient and lack predictive insights. APM through digital twins is a systematic approach to improving the reliability and availability of physical assets such as machinery, industrial plants, or fleets of assets throughout their life cycle.

Goals include improving operational efficiency, enhancing asset lifespan, and reducing unplanned downtime. APM use cases go beyond simply monitoring assets and predicting failures to prescriptive maintenance.

Primary sub-use cases include:

1. Contextual operational awareness via effective visualization of merged IoT, ERP, and maintenance data
2. Anomaly detection and Remaining Useful Life (RUL) prediction
3. Prescriptive analytics and scenario planning.

IoT sensors collect asset operating data, enabling real-time behavior modeling, anomaly detection, and failure prediction. Simulation models improve asset utilization by optimizing asset performance and assessing different maintenance strategies. Spatial technologies allow maintenance teams to visualize asset health and performance metrics overlaid on physical equipment, providing contextualized information access.

You can read about a reference deployment of the DTF with Cadent Gas, AWS, and our partner, Bosch. Learn how to use Amazon Bedrock and services like AWS IoT TwinMaker for digital twins for predictive maintenance through the DTF in this blog post.

Process Optimization

Digital twins address inefficiencies in complex manufacturing processes, traditionally plagued by resource waste and production bottlenecks. These twins focus on streamlining performance, managing waste, and reducing emissions at the plant level.

Key capabilities include:

1. Improving operational awareness through real-time monitoring across facilities
2. Identifying anomalous and sub-optimal plant performance
3. Conducting what-if analysis via simulation before implementing changes
4. Developing optimized operations plans considering inventory and deliveries
5. Optimizing site layout using a dynamic virtual model of the facility incorporating equipment usage, traffic flow, and environmental condition data.

Virtual simulations test site layout changes and spatial technologies optimize asset placement based on workflow efficiency, safety, and to account for varying conditions. You can read about how Amazon Reliability and Maintenance Engineering (RME) is building a digital twin to improve safety through Prevention through Design.

Open-source AWS frameworks like Twinflow orchestrate models, provide workflow traceability, and optimize compute for scalable digital twins. An integrated approach enhances production efficiency, reduces waste, and enables dynamic responses to changes. It helps create leaner, more adaptive manufacturing environments, as compared to using traditional data analysis and costly trial-and-error adjustments.

Engineering Design

Digital twins enhance product design, reducing reliance on costly, time-consuming physical prototypes. They accelerate development cycles, cut costs, and provide powerful tools for predicting outcomes, testing variations, and optimizing performance. They provide engineers and designers with powerful tools to predict outcomes, test variations, optimize performance, and enhance decision-making.

Customers often want to accelerate their engineering design cycles by employing AI-augmented simulation to iterate faster on their designs, and deploying remote engineering collaboration tools for remote teams to work closely together.

In the area of simulation, we’re exploring techniques to augment HPC simulations with ML methods to increase the number of designs you can evaluate using predictive models that run-in seconds versus HPC simulations that can take hours. We’re also exploring generative AI-based agentic workflows to assist users in building and running their simulations using natural language queries.

In the area of remote collaboration, we’re seeing the use of virtual reality and augmented reality during engineering design to allow remote users interact within the spatial visualization to improve collaboration. We’ve put a reference implementation of the DTF for Product Design Optimization in the AWS Solutions Library.

For more information on Smart Industrial Machines overall, and how the framework supports this, you can refer to a dedicated post discussing these concepts in detail.

Summary

In this post, we discussed four categories of digital twin use cases being explored by our customers. Each of these use cases leverages technologies at the intersection of IoT, spatial computing, and predictive modeling.

Building digital twin solutions across these domains can be challenging and we described how the AWS Digital Twin Framework (DTF) reference architecture enables customers to build extensible, scalable solutions on AWS.

In future posts, we will detail these use cases, and others to continue to highlight how the DTF helps our customers not just initially deploy, but scale digital twins in their organization.

If you want to request a proof of concept or if you have feedback on the AWS tools, please reach out to us at ask-hpc@amazon.com.