Production Graph Monitoring for Reconfigurable Factory Analytics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current production monitoring technologies in manufacturing require significant programming and configuration changes, leading to production interruptions when updating monitoring processes to handle new data or configurations, as they are based on time series analysis and require reprogramming of human machine interface (HMI) coding.
Innovation Solution
A graph-driven approach for production process monitoring using a processor and memory with modules for executing a production ontology and graph engine to instantiate a production graph, allowing real-time data population and machine learning for predictive analytics without the need for reprogramming interruptions, decoupling the graph model from control and HMI programs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional time series analysis-based production monitoring is used, then production monitoring functionality is achieved, but reprogramming interruptions are required when updating monitoring processes to handle new data or configurations
Solution Approach 1:
The system segments the monitoring functionality into independent modules: a graph engine that handles data visualization and a machine learning engine that handles analysis. These modules can be updated independently without requiring system-wide reprogramming, allowing new monitoring capabilities to be added without production interruptions.
Solution Approach 2:
The patent implements dynamic configuration capabilities where the graph engine can adapt to new data types and configurations at runtime without requiring reprogramming. The system dynamically loads and executes monitoring algorithms, enabling flexible updates while maintaining continuous operation.
2Reliability
If complex reprogramming of HMI coding is performed to handle new events, then monitoring coverage is improved, but production operation must be halted
Solution Approach 1:
The machine learning engine automatically processes and analyzes production data without requiring manual HMI coding for each event type. The system self-configures monitoring rules and algorithms, eliminating the need to halt production for reprogramming while maintaining comprehensive monitoring coverage.
Solution Approach 2:
The graph engine and machine learning engine serve multiple monitoring functions through a single unified platform. Rather than requiring separate HMI code for different event types, the universal engine handles various monitoring tasks including real-time visualization, anomaly detection, and predictive analytics, all without requiring production stoppage.
3Reliability
If traditional production monitoring systems are used, then basic monitoring is achieved, but significant engineering effort is required for configuration and programming
Solution Approach 1:
The patent replaces traditional manual HMI coding and configuration mechanisms with automated graph-based visualization and machine learning algorithms. The graph engine automatically generates visual representations of production data, and the ML engine automatically configures analysis parameters, significantly reducing engineering effort while maintaining robust monitoring functionality.
Solution Approach 2:
The system uses configurable parameters and metadata to define monitoring behavior rather than hard-coded programming. By changing parameters and configuration data rather than reprogramming, the system adapts to new monitoring requirements with minimal engineering effort, maintaining high reliability through data-driven configuration.
Data Source
AI summary
A system and method are disclosed for production process monitoring using graph learning. A production ontology is generated based on data received from an engineering design process, the ontology including selection of production machines, definition of a product, and workflow design. A production graph is instantiated based on the production ontology. Production process data is read from control systems of the production environment and the production graph is populated with the production process data to generate a time series of production graphs. Prediction information is received from historical production graphs of related production processes. Offline runtime analytics are performed on the production graph to yield analytics results including a plurality of predictors. The predictors include knowledge from the received prediction information leveraged with a weight sharing initialization.