Building a Manufacturing Digital Thread using Graph and Generative AI on AWS

With the majority of manufacturing organizations embracing digital transformation efforts, the role of data as a strategic asset is becoming increasingly clear. When harnessed effectively throughout the product lifecycle, manufacturers can unlock key insights to reduce overall costs, improve product quality, streamline supply chain and deliver differentiated solutions to customers. However, operationalizing a data-driven transformation is non-trivial. According to an Accenture study, 73% of the executives reported inability to derive data-driven insights from legacy data, assets and operations.

In most manufacturing organizations, the product lifecycle data is typically stored in enterprise systems such as Product Lifecycle Management (PLM), Enterprise Resource Planning (ERP), Manufacturing Execution Systems (MES), Customer Relationship Management (CRM), and other enterprise applications. Due to the disconnected nature of enterprise infrastructure, the generated data is often fragmented and goes unused. According to a study commissioned by Seagate and conducted by the research firm IDC, 68% of the data generated by the enterprises remains unleveraged. The report also identifies the missing link of data connectivity between the data producers and consumers.

Enabling manufacturing stakeholders to harness data across various stages of the product lifecycle to make better decisions has become a key imperative for business transformation. However, managing and contextualizing data across the product lifecycle is challenging due to variations in data representation and usage patterns that are dependent on the specific persona interacting with it. One approach manufacturers use to integrate and contextualize data is by creating a Digital Thread, which ensures a seamless and interconnected flow of authoritative data (e.g., requirements, product information, product cost, documents, supply chain, quality, defects, maintenance etc.,) generated by the product throughout its lifecycle. As the product traverses through various lifecycle stages – from ideation to design, manufacturing and in-field operations – it establishes a thread that connects diverse data sources to analyze the product performance at any instance of time and drives data-driven business outcomes.

Knowledge graph and generative AI for digital thread

While there are several approaches to establish integration between heterogeneous enterprise systems, such as point-to-point integrations between data sources, a practical method to establish a digital thread is by using knowledge graphs to represent manufacturing data and its relationships.

Knowledge graphs allows us to create a structured representation of connected data from various data sources to represent highly complex relationships between objects. It is designed to organize data by enabling efficient storage, retrieval, and visualization of complex relationships. Additionally, it enables data to be more accessible to various stakeholders, helping them gain better knowledge to make better decisions.

The digital thread, which connects the data across the lifecycle of a product from requirements to end-of-life, requires the ability to track dependencies and ensure end-to-end traceability. Knowledge graphs enable these semantic layers by connecting disparate data and enables seamless data accessibility across the enterprise.

Figure 1: Manufacturing Digital Thread – Knowledge Graph

In the graph shown in Figure 1, entities like requirements, product data, production order represent nodes, and edges represent the semantic connections between them. This digital thread graph allows seamless navigation through the interconnected manufacturing enterprise data, making it easy to understand the complex relationships. For example, it helps cross-functional teams and decision makers understand the relationship between a part and its associated defects and requirements, leading to improved quality outcomes and design processes. The knowledge graph queries facilitate extracting context-specific viewpoints and relevant information essential for decision-making processes among various stakeholders in the organization.

While knowledge graphs are ideal for developing a digital thread, automating graph queries and generating natural language text can pose challenges. To enhance the user experience, large language models (LLM) can be used to interpret complex graph data, analyze relationships to generate queries, and provide results in natural language based on the digital thread graph.

In this blog post, we explain how to create a manufacturing digital thread application by combining the capabilities of Amazon Neptune and Amazon Bedrock. Using generative AI technology with the graph enables business stakeholders to get insights faster in a conversational way.

Digital Thread Solution Framework

Data from various product lifecycle processes forms the foundation of this digital thread solution framework. The subsequent layer encompasses core enterprise systems, including PLM, ERP, and MES, which manage specific aspects such as people, processes, engineering, and manufacturing data within the enterprise. The next layer involves the knowledge graph, connecting data from these systems to establish relationships, derive insights, and facilitate a comprehensive understanding of the interconnected manufacturing enterprise data. Finally, large language models are integrated with the knowledge graph, enabling the creation of advanced graph queries and natural language capabilities.

 Figure 2: Manufacturing digital thread solution framework

As shown in Figure 2, the solution aims to accelerate innovation by seamlessly connecting data from various systems and generate insights through the manufacturing digital thread framework. This framework establishes an intelligent structure where data is interconnected, empowering manufacturing stakeholders in their decision-making processes.

Solution Architecture

The manufacturing digital thread solution architecture is implemented through the strategic integration of Amazon Neptune graph database and Amazon Bedrock, a fully managed generative AI service. The components are further enhanced by various AWS services, creating a comprehensive solution for the digital thread.

 Figure 3: Manufacturing digital thread solution architecture

As shown in Figure 2, the solution leverages over 15 AWS services and handles 10 essential functions to support the development of digital thread capabilities.

1. Identify key stakeholders in the manufacturing organization
To start a digital thread journey, it’s essential to identify key stakeholders within the manufacturing organization. This includes system engineering, design engineering, manufacturing engineering, supply chain, operation and quality teams. Understanding their unique business interests and use cases provides the foundation for a connected digital thread.

2. Identify data sources for building the Digital Thread
Identify the data sources required to build a comprehensive digital thread. These may include PLM, ERP, CRM, MES/MOM, and other in-house enterprise applications. By identifying the key business problems, source systems and data, enterprises can ensure the inclusion of critical data for a holistic view of their operations.

3. Data Ingestion
Ingest data from the identified sources into AWS using services like AWS Data Migration Service (DMS) with Amazon S3 as target and AWS DataSync for large dataset movement into Amazon S3.

4. Upload Data to S3
Once data is successfully ingested, load the data into Amazon Simple Storage Service (Amazon S3).

5. Use Bulk Loader to load data into Amazon Neptune graph database
Leverage Amazon Neptune bulk loader capability to ingest the data stored in Amazon S3 into Amazon Neptune. The graph edges and vertices created in Amazon Neptune provides the basis for graph-based queries.

6. Use generative AI Model
Select a foundation model (FM) from Amazon Bedrock, a fully managed service that offers a choice of high-performing FMs from leading AI companies and Amazon through a single API, along with a broad set of capabilities for building generative AI applications.

7. Establish knowledge graph, LLM connection and orchestrate using langchain.
Establish the link between Amazon Bedrock (Claude 3.5 Sonnet), Amazon Neptune and integrate them seamlessly with langchain. It coordinates the process of generating the query from the foundation model, executing the query against the knowledge graph and returning the results in natural language to the user.

8. Create application layer
Create the application layer by combining the following: Streamlit App, AWS Fargate for running containerized applications, Amazon Elastic Container Registry for managing container images, Elastic Load Balancing (ELB) for traffic distribution, and Amazon Cognito for secure user authentication. This setup, deployed via AWS Copilot CLI, ensures a scalable, secure user interface for seamless stakeholder interaction with the digital thread data.

9. Security
Leverage Amazon VPC to operate the digital thread application in a secure and isolated network. AWS Identity and Access Management (IAM) enhances access control, while AWS Certificate Manager (ACM) manages certificates and AWS WAF ensures web application security.

10. Management and Governance
Leverage AWS CloudTrail to enhance transparency by tracking activities, Amazon CloudWatch to monitor resources and AWS CloudFormation for automated resource deployment of digital thread application.

For more information, please refer to the guidance, workshop and the sample code provided in the GitHub repository.

Figure 4: Manufacturing Digital Thread Advisor

Figure 4 showcases a conversational digital thread advisor combining enterprise data from PLM, MES, ERP, CRM systems, delivering intelligent insights through natural language queries.

Digital Thread use cases

Using a knowledge graph and generative AI based manufacturing digital thread enables organizations to unlock efficiency and innovation through seamless data integration. Key use cases include:

Requirements to Quality Traceability: Maintaining traceability between product requirements and quality can be a challenging task. By leveraging manufacturing digital thread, organizations can gain a well-defined view of how initial requirements influence the final product quality. This approach enables stakeholders to proactively identify issues and propose remedies before it leads to significant problems.

Streamlining Supplier Collaboration in Product Development: Collaborating with multiple suppliers and managing design specifications can make the product development a complicated process. By utilizing knowledge graphs and generative AI, organizations can seamlessly connect product and supplier information. This integration provides real-time insights into how supplier choices influence the final product, enabling more informed decision-making and streamlining the overall development process.

Optimizing Supply Chains for Sustainability: Achieving sustainability in supply chain is a critical goal for modern manufacturing businesses. By leveraging the digital thread, organizations can connect every aspect of their supply chain, from products, suppliers to logistics. This comprehensive view provides insights into carbon footprints and other sustainability metrics, enabling organizations to develop strategies that reduce carbon emissions.

Enhanced Customer Support with Contextual Insights: Providing effective customer support requires a deep understanding of the entire product journey. By utilizing a connected digital thread that includes design, manufacturing, and customer feedback, support teams can access comprehensive insights from the thread. This enables them to resolve issues more quickly and accurately, eventually enhancing customer satisfaction and loyalty.

The manufacturing digital thread capabilities using graph and generative AI extend beyond the use cases mentioned above, providing opportunities to transform operations across the entire product lifecycle and across industries.

Conclusion

In this post, we explored how to build the digital thread using knowledge graphs and generative AI on AWS. Knowledge graphs and large language models are essential for creating a connected manufacturing digital thread. Graphs provide the data connectivity and traceability needed to navigate complex manufacturing processes, while LLMs extract valuable insights from the graph and presents them in a natural language. By harnessing the power of knowledge graphs and generative AI, manufacturing organizations can enhance data accessibility, collaboration, traceability, efficiency, and enable faster data driven decision making process.