AWS re:Invent 2025 -Agentic Code Generation for Industrial Analytics & Predictive Maintenance-PEX320

Summary of AWS re:Invent 2025 Presentation: Agentic Code Generation for Industrial Analytics & Predictive Maintenance

Common Challenges in Industrial IoT and Manufacturing

• Merging operational technology (OT) and information technology (IT) systems with different processes, software, and environments
• Slow digital transformations due to lack of skills and expertise in IoT, AI, and digital twins
• Data analysis gap and skills shortage to create predictive models
• Costly downtime and the need for more frequent scheduled maintenance
• Manufacturing skills gap due to an aging workforce and knowledge loss

AWS Stack for Agentic Code Generation

• Infrastructure layer: AWS Trainium, Inferentia, and other AI compute options
• Platform layer: Amazon SageMaker for data science and model training
• Agentic layer:
  • Amazon Bedrock for large language models and knowledge base access
  • Agent Core for managing and executing agentic applications
  • Strands SDK for building multi-agent workflows

Tutorial 1: Autonomous Maintenance Response and Repair Plan Generation

• Prerequisite: Prepare AWS IoT SiteWise environment with asset data and anomaly detection
• Workflow triggered by anomaly detection in SiteWise
  • Agent 1 (Ops Data Collector): Retrieves sensor data from SiteWise
  • Agent 2 (Knowledge Retriever): Looks up relevant equipment manuals and SOPs from Bedrock knowledge base
  • Agent 3 (Report Generator): Summarizes findings, generates HTML repair plan report, and stores it in S3
• Benefits:
  • Automated response to equipment anomalies
  • Leverages equipment knowledge to provide maintenance recommendations
  • Generates a shareable report for the operations team

Tutorial 2: Advanced Analytics for Predictive Maintenance

• Objective: Analyze historical data to identify common patterns and root causes of recurring anomalies
• Process:
  • Loads equipment sensor data and anomaly history into a Pandas DataFrame
  • Uses Agent Core's code interpreter to dynamically generate data analysis code (Python, NumPy, Matplotlib)
  • Performs correlation analysis, identifies normal operating ranges, and detects data quality issues
  • Provides recommendations for maintenance and equipment settings adjustments
• Benefits:
  • Empowers process engineers to quickly analyze equipment health without relying on data scientists
  • Uncovers hidden insights and relationships in sensor data
  • Enables proactive maintenance and optimization of equipment performance

Key Takeaways

• Agentic code generation leverages large language models and multi-agent workflows to automate complex industrial analytics and maintenance tasks
• Integrates AWS services like SiteWise, Bedrock, and Agent Core to create a comprehensive industrial IoT and predictive maintenance solution
• Addresses manufacturing skills gaps by democratizing data analysis and empowering domain experts to drive operational improvements
• Provides both reactive (anomaly response) and proactive (advanced analytics) capabilities to optimize equipment uptime and performance

Technical Details

• AWS services used: IoT SiteWise, Bedrock, Agent Core, Strands SDK, S3
• Anomaly detection and repair plan generation workflow using three specialized agents
• Advanced analytics leveraging code generation and dynamic execution within Agent Core

Business Impact

• Reduces costly equipment downtime and unplanned maintenance through automated anomaly response
• Improves operational efficiency and agility by empowering domain experts to quickly analyze and optimize equipment performance
• Addresses manufacturing skills gaps by democratizing data analysis and enabling a new generation of domain-driven industrial analytics
• Enhances predictive maintenance capabilities and enables proactive optimization of equipment settings and operations

Examples

• Anomaly detection and repair plan generation for a welding robot in an e-bike assembly line
• Advanced analytics to identify common patterns and root causes of recurring anomalies across a fleet of similar equipment