LLM Failure Cause Estimation From Logs and Technical Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional failure cause analysis technologies, such as rule-based and failure-case-based methods, struggle with limited analyzable ranges and accuracy in identifying specific failure causes, especially in complex IT systems with multi-vendor environments, leading to inefficient failure recovery.
Innovation Solution
An estimation system utilizing a machine learning model, specifically a large language model (LLM), to analyze logs, generate prompts for failure classification and cause inference, and output specific failure causes by extracting relevant sentence chunks from technical documents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule-based analysis technology is used for failure cause analysis, then the analysis can be performed using predefined rules, but the analyzable range is limited to predefined failure causes
Solution Approach 1:
The system enables the failure analysis apparatus to automatically extract failure causes from technical documents without requiring manual rule creation or configuration. The machine learning model autonomously processes logs and documents to identify failure causes, eliminating the need for expert-defined rules while expanding the analyzable range to include undefined and novel failure causes.
2Reliability
If failure-case-based analysis technology is used, then the system can learn from past failure cases, but the analyzable range is limited to similar past cases
Solution Approach 1:
The system performs preliminary extraction of failure cause information from technical documents and stores it in advance. When a failure occurs, the machine learning model can quickly query and compare this pre-extracted information with current failure logs, enabling both rapid analysis of similar cases and the ability to handle novel failure scenarios that differ from past cases.
3Loss of information
If conventional failure cause analysis is performed, then the overall outline of failures can be analyzed, but specific failure cause analysis leading to countermeasures is limited
Solution Approach 1:
The system segments the failure analysis process into distinct stages: log analysis for failure classification, document extraction for specific failure cause identification, and inference for countermeasure recommendation. This segmentation allows the system to first determine the overall failure classification and then drill down to identify specific failure causes with high precision, ultimately providing actionable countermeasures.
4Extent of automation
If AI technology is used for failure cause analysis, then automation is improved, but the precision of specific failure cause estimation is insufficient
Solution Approach 1:
The system introduces technical documents as an intermediary between the automated log analysis and the final failure cause determination. The machine learning model extracts failure cause information from these documents and uses it to refine and validate its inferences, thereby maintaining high automation while significantly improving the precision of specific failure cause identification.
Data Source
AI summary
An estimation system includes, a memory having a machine learning model and, a processor coupled to the memory and configured to, analyze a log related to a failure to determine a failure classification of a failure content, cause the machine learning model to extract, from a predetermined technical document, a corresponding sentence corresponding to the determined failure classification, cause the machine learning model to infer a first failure cause based on the corresponding sentence extracted by the machine learning model, and output the first failure cause.