Prescriptive Alert Messaging Using LLM Feedback for Asset Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing condition monitoring systems face challenges in providing prescriptive maintenance without requiring explicit configuration by system designers or users, due to the complexity of integrating with maintenance management systems and the variability in how failures manifest in sensor data across different machines.
Innovation Solution
A machine-agnostic method using a trained large language model that processes user feedback and historical operational data to provide prescriptive messaging by filtering and analyzing similarities and relatedness of asset conditions, without the need for explicit rule-setting or integration with maintenance systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If predefined rules or thresholds are used for prescriptive messaging, then the system can provide structured failure analysis, but the system complexity increases and requires extensive configuration by system designers
Solution Approach 1:
The system enables users to independently configure and refine prescriptive messaging by providing feedback on alert accuracy. Users can mark alerts as correct or incorrect, and the system automatically learns from this feedback without requiring system designer intervention or complex manual configuration.
Solution Approach 2:
The system implements a feedback loop where user responses to alerts (correct/incorrect markings) are continuously collected and used to refine the alerting model. This feedback mechanism allows the system to self-improve prescriptive messaging accuracy over time without increasing configuration complexity.
2Quantity of substance
If maintenance management system integration is implemented to collect failure information, then historical failure data becomes available for analysis, but the integration complexity increases significantly
Solution Approach 1:
The system collects and processes failure data directly from asset operational data without requiring integration with external maintenance management systems. The asset data itself contains sufficient information for prescriptive messaging, eliminating the need for complex system integrations.
3Reliability
If machine-specific rules are created for each asset type, then accurate prescriptive messaging can be provided, but the system becomes difficult to scale to new asset types
Solution Approach 1:
The system uses a universal alert refinement model that works across all asset types without requiring machine-specific configuration. The same feedback-based learning mechanism applies to diverse assets including turbines, compressors, and pumps, enabling the system to scale to new asset types while maintaining accuracy.
Solution Approach 2:
The system automatically adapts to new asset types through self-service learning from user feedback, eliminating the need for manual rule creation for each machine type. The model generalizes across asset types by learning from patterns in the feedback data.
4Measurement precision
If users are required to identify failure patterns manually, then training data can be obtained for learning algorithms, but the workload and expertise requirements increase for users
Solution Approach 1:
The system replaces complex manual pattern identification with simple feedback provision. Users only need to mark alerts as correct or incorrect, and the system automatically learns the underlying failure patterns from this simplified input, maintaining detection accuracy while dramatically reducing user effort.
Solution Approach 2:
The system performs the complex task of failure pattern identification automatically through self-service learning from user feedback, relieving users of the burden of manual pattern analysis while still achieving high detection accuracy.
Data Source
AI summary
A data processing system and methods for evaluating an operational condition of an asset are described. The asset is one of a plurality of assets that are monitored by a condition monitoring system. The condition monitoring system automatically issues alert messages based on the operational condition of the monitored assets. The methods use a trained large language model to generate output messages that provide a summary of contextually relevant content from feedback from previous alert messages in order to assist a user in evaluating an operational condition of an asset.