Sensor-Tagged Support Workflow for Context-Specific LLM Advice
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for processing systems in facilities like chemical plants and factory manufacturing lines lack flexibility in generating advice based on sensor data, as they are limited to predefined templates, restricting the output of tailored recommendations.
Innovation Solution
A support system utilizing a large language model that assigns unique tags to measurement data items from sensors, allowing external information to be associated with these tags for generating flexible advice through a prompt-based approach, including evaluation target data and information to generate reports.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If preset template data is used to generate advice, then the system structure is simple and easy to implement, but the output advice lacks flexibility and cannot provide tailored recommendations
Solution Approach 1:
The system changes the parameter of advice generation from fixed template-based output to dynamic LLM-generated output. By adjusting the prompt parameters (inputting different measurement data items and their relationships), the system can generate flexible, context-specific advice without requiring complex rule-based systems, thus resolving the contradiction between flexibility and complexity
Solution Approach 2:
The patent introduces an intermediary component (the LLM with prompt engineering) between the measurement data and the advice output. This intermediary processes the input data and generates appropriate advice dynamically, avoiding the need for complex predefined rules while maintaining system simplicity. The intermediary enables flexible output without proportionally increasing system complexity
2Loss of information
If multiple measurement data items are processed individually, then each data item can be evaluated separately, but the advice lacks context and cannot provide comprehensive operational support
Solution Approach 1:
The system merges multiple measurement data items into a single comprehensive prompt for the LLM. By combining related measurement data (such as sensor readings with their relationships to other system components) into one unified input, the system retains contextual information and enables the LLM to generate holistic advice that considers interdependencies between different data points
Solution Approach 2:
The LLM serves as a universal processing component that can handle multiple types of measurement data items simultaneously. Rather than creating separate processing paths for each data type, the system uses a single multi-functional LLM that can process various sensor data, maintenance records, and operational parameters together, reducing overall system complexity while preserving context
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Multiple measurement data items corresponding to measurement values collected by sensors is each assigned a first tag unique to the sensor collecting the measurement data item. External information is assigned a second tag corresponding to the first tag assigned to a related measurement data item. The measurement data items include an evaluation target measurement data item assigned the first tag matching an evaluation target tag. The external information includes evaluation target external information assigned the second tag corresponding to the first tag matching the evaluation target tag. The support system includes a processor configured to generate a prompt including the evaluation target measurement data item, the evaluation target external information, and an instruction to generate the advice based on that data item and that information, acquire an answer including the advice from a large language model by inputting the prompt into the model, and generate the report using the answer.