Empowering predictive maintenance with Amazon Bedrock

In the competitive field of industrial operations, minimizing downtime is crucial for financial success. The challenge lies in managing complex data spread across various systems within industrial environments. Quickly identifying and solving issues without advanced analytics can be difficult.

Operators often spend time consulting manuals to diagnose problems and follow protocols. Automation can streamline this process. The key is detecting anomalies early and providing troubleshooting instructions promptly.

Amazon Web Services (AWS) can help make things easier.

AWS IoT SiteWise, Amazon Lookout for Equipment, and Amazon Bedrock form a powerful combination. AWS IoT SiteWise collects and visualizes time-series data from industrial assets at scale. Amazon Lookout for Equipment uses machine learning to detect subtle anomalies, preventing costly downtime. Amazon Bedrock steps in by giving operators step-by-step instructions to address identified anomalies. Together, this trio processes asset data, identifies anomalies, and prescribes efficient repairs. Empowered operators can quickly resolve problems – minimizing downtime and enhancing operational efficiency.

In this technical blog post, we’ll explore the integration of AWS IoT SiteWise, Amazon Lookout for Equipment, and Amazon Bedrock to create an advanced industrial analytics system. The focus will be on detecting emerging motor failures, retrieving relevant manuals, and generating comprehensive repair plans. The goal is to maximize uptime, boost productivity, reduce costs, and revolutionize industrial operations.

Overview of solution

The industrial analytics solution that we’re describing employs AWS IoT Greengrass and AWS IoT SiteWise to collect and visualize time-series data from industrial assets at scale. AWS IoT Greengrass lets Internet of Things (IoT) and industrial devices collect and analyze data closer to where that data is generated, react autonomously to local events, and communicate securely with other devices on the local network. In this solution, an AWS IoT SiteWise Edge gateway serves as the intermediary between industrial equipment and AWS IoT SiteWise. It is possible to deploy AWS IoT SiteWise Edge gateway software on any device that can run AWS IoT Greengrass.

AWS IoT SiteWise lets you model industrial equipment and processes using AWS IoT SiteWise asset models (“templates”) and assets. Further, the industrial analytics solution defines relationships among these assets and constructs an asset hierarchy under ISA-95from the International Society of Automation (ISA).

Figure 1. Architecture

This solution uses the new AWS IoT SiteWise integration with Amazon Lookout for Equipment, which provides multivariate anomaly detection without the user needing deep technical experience in machine learning. Amazon Lookout for Equipment can use data from up to 300 sensors per equipment and train an optimal model using automated machine learning. Examples of equipment you can use include compressors, pumps, motors, boilers, robots, and Computer Numerical Control (CNC) machines, and many more.

Amazon Bedrock steps in by giving operators instructions to address identified anomalies. It is a fully managed service offering diverse foundation models from leading artificial intelligence (AI) companies and provides capabilities for constructing generative AI applications with a focus on security, privacy, and responsible AI practices.

The industrial analytics solution uses Amazon Titan, an exclusive family of models within Amazon Bedrock that are created by AWS and pretrained on large datasets. Specifically, it incorporates the Amazon Titan Text Embeddings model to generate embeddings (semantic vectors) from a knowledge base containing runbooks with incident classes, preconditions, root causes, resolution steps, and operational information related to applications.

The AI models generally available through Amazon Bedrock, like Anthropic’s Claude family, examine anomaly descriptions and context in the vector database to create suitable responses. These responses might include troubleshooting steps, seeking additional details, adding work notes, or alerting users about system issues. The solution includes an integrated Anomaly Detection Dashboard to keep track of the system’s operational status, improving solution reliability.

Walkthrough: Anomaly Detection and Knowledge-Driven Maintenance

Anomaly detection

Figure 2. Anomaly detection

The AWS IoT SiteWise gateway connects to data sources, such as the Ignition OPC Unified Architecture (UA) server, to fetch real-time data from the plant floor. Powered by AWS IoT Greengrass components, the gateway software simplifies this process. AWS IoT SiteWise then receives this data from the AWS IoT Greengrass gateway. Real-time monitoring of diagnostic data from the industrial plant is made possible with AWS IoT SiteWise Monitor. The collected data is normalized to maintain consistency for analysis. Amazon Lookout for Equipment identifies anomalies by analyzing real-time data using a model built from historical data provided during configuration. When an anomaly is spotted, Amazon Lookout for Equipment initiates an AWS Lambda automation, generating a detailed prompt for Amazon Bedrock. This seamless flow results in efficient data processing, analysis, and response in an industrial setting.

Static knowledge analysis

Figure 3. Static knowledge analysis

Documentation, made up of manuals and runbooks, resides in a bucket in Amazon Simple Storage Service (Amazon S3), an object storage service. The processing workflow involves using Amazon Bedrock (Amazon Titan model) for digital documents and Amazon Textract for handling printed text, handwriting, and layout elements within the documentation. In the event of an anomaly detected by Amazon Lookout for Equipment, an AWS Lambda function is initiated to create a prompt. This prompt is then forwarded to Amazon Bedrock to access static knowledge. The query output, combined with the alert generated by Amazon Lookout for Equipment, is transmitted to Amazon Bedrock. Subsequently, a well-defined and formatted custom repair plan is generated. The final step involves the AWS Lambda function sending the custom repair plan through Amazon Simple Notification Service (Amazon SNS)—a fully managed pub/sub service—and archiving it in an Amazon S3 bucket for future reference.

Conclusion

Industries operating on slim profit margins have a critical need to prevent unplanned downtime and operational inefficiencies. In this blog post, you’ve learned how the integration of AWS IoT SiteWise, Amazon Lookout for Equipment, and Amazon Bedrock can help meet this need.

Amazon IoT SiteWise and Amazon IoT Greengrass provide you comprehensive insights by collecting and transforming data from IIoT edge gateways. Machine learning in Amazon Lookout for Equipment can help you detect anomalies early, averting costly disruptions. Amazon Bedrock steps in to provide you with personalized troubleshooting guides, leveraging institutional knowledge and automating the resolution process. This intelligent loop—from data ingestion to anomaly identification and repair prescription—empowers you to swiftly address emerging issues, significantly reducing downtime and associated costs.

If you’re enthusiastic to see a demonstration or dive deeper into the integration of these AWS services for your specific industrial setup, please contact us. We’re here to help you use these technologies and drive positive change in your industrial operations.