Knowledge-Retrieval LLM Pipeline for Jet Engine Maintenance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automating maintenance activities for complex apparatuses like jet turbine engines using large language models can lead to incorrect instructions and results, especially when augmented with knowledge retrieval, posing challenges in aviation applications.
Innovation Solution
A knowledge-retrieval augmented Large Language Model pipeline is developed to automate maintenance recommendations, involving a control circuit that receives a textual description, retrieves relevant knowledge documents, generates a language generation prompt, and outputs candidate recommendations for review by a subject matter expert, with the option to validate and retrain the decoder using human feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large language model is augmented with knowledge retrieval to provide maintenance recommendations, then the relevance and information completeness of recommendations is improved, but the accuracy and reliability deteriorates due to generation of incorrect instructions
Solution Approach 1:
The patent introduces an intermediary verification step where a subject matter expert reviews and validates the candidate maintenance recommendation before it is finalized. This intermediary human reviewer acts as a mediator between the AI-generated recommendation and the final output, ensuring accuracy while maintaining the benefits of knowledge-retrieval augmentation.
Solution Approach 2:
The patent implements a feedback mechanism where the subject matter expert's validation or correction of the candidate recommendation feeds back into the system. This feedback loop allows the system to learn from human expertise and improve future recommendations, balancing automated efficiency with human oversight.
2Productivity
If a large language model is used to automate maintenance activities, then the productivity and efficiency are improved, but the reliability deteriorates due to generation of incorrect instructions and results
Solution Approach 1:
The system employs a subject matter expert as an intermediary to review and validate automated recommendations. This hybrid approach maintains the productivity benefits of automation while introducing human oversight to ensure reliability and accuracy of maintenance instructions.
Solution Approach 2:
The patent applies partial automation where the large language model generates candidate recommendations but does not fully automate the final decision-making process. The subject matter expert performs partial validation, creating a balanced system that leverages both automated efficiency and human expertise.
3Loss of information
If knowledge documents are retrieved and processed by the large language model, then the information completeness is improved, but the device complexity increases due to the pipeline architecture
Solution Approach 1:
The patent segments the maintenance recommendation system into distinct functional modules: a knowledge retrieval component that retrieves relevant documents, a large language model component that generates candidate recommendations, and a validation component reviewed by a subject matter expert. This segmentation allows each component to specialize in its function while working together in a coordinated pipeline.
Solution Approach 2:
The large language model serves multiple functions within the pipeline: it processes retrieved knowledge documents, generates candidate maintenance recommendations, and can incorporate subject matter expert feedback. This multi-functionality reduces the need for separate specialized components, managing complexity while maintaining information completeness.
Data Source
AI summary
A textual description of an issue pertaining to at least part of an apparatus is received as input. A data store is accessed and a plurality of knowledge documents that corresponds to the input is retrieved. A language generation prompt is then generated, as a function of both the input and the plurality of knowledge documents and output to a task-specific decoder that generates a candidate recommendation to address the aforementioned issue. That candidate recommendation is output to at least one human reviewer who reviews the candidate recommendation as a function of the plurality of knowledge documents and who then provides a corresponding human-validated recommendation to address the issue. The task-specific decoder can be retrained using the human-validation recommendation coupled with the corresponding textual description input.