Causal Factor Inference for Fast Abnormality Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies struggle to provide non-expert users with accurate and actionable insights from machine learning results, particularly in understanding the mechanisms behind abnormality detection in complex data sets.
Innovation Solution
A factor inference device and method that utilizes a knowledge model expressed as a network of nodes representing events in a process, along with an information creation unit, node extraction unit, data combining unit, and factor inference unit to infer and present the factors contributing to a phenomenon, such as abnormality in sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are used for abnormality detection with large number of variables, then prediction accuracy is improved, but inference time and computational load increase
Solution Approach 1:
The patent segments the inference process into two stages: (1) pre-calculation of importance scores for each variable using machine learning during normal operation, and (2) rapid factor inference during abnormality detection by combining pre-calculated importance scores with current sensor data. This segmentation allows accurate predictions while reducing real-time inference time.
Solution Approach 2:
The system performs preliminary actions by pre-calculating and storing importance scores for each variable in the knowledge model before abnormality occurs. When abnormality is detected, the system quickly retrieves these pre-computed scores and combines them with current data to infer factors, avoiding time-consuming calculations during critical moments.
2Device complexity
If machine learning models output only variable influence without causal relationships, then computational complexity is reduced, but user understanding of mechanisms deteriorates
Solution Approach 1:
The patent introduces a knowledge model as an intermediary that connects machine learning output with human understanding. The knowledge model stores causal relationships between variables and phenomena in a structured format (e.g., directed acyclic graphs). When abnormality is detected, the system queries the knowledge model to trace causal pathways from abnormal variables to potential factors, providing both computational efficiency and mechanistic understanding.
Solution Approach 2:
The system replaces complex causal reasoning calculations with a pre-structured knowledge model that encodes domain expertise. Instead of performing heavy computational analysis of causal relationships in real-time, the system uses the knowledge model to quickly retrieve and present causal pathways, substituting mechanical computation with structured knowledge representation.
3Measurement precision
If expert knowledge is required to interpret machine learning results, then inference accuracy is improved, but ease of operation deteriorates
Solution Approach 1:
The system enables self-service by automatically performing factor inference without requiring expert user intervention. The factor inference unit automatically combines current sensor data with pre-stored importance scores and queries the knowledge model to infer factors and present results. Non-expert users simply need to input abnormality data and receive interpreted results, eliminating the need for users to possess deep domain knowledge while maintaining high inference accuracy.
Data Source
AI summary
A factor inference device for inferring a factor of a phenomenon in a process includes: a knowledge model acquisition unit configured to acquire a knowledge model including events occurring in the process as nodes, the knowledge model being expressed in a network form connecting the nodes, with respect to a causal relationship of the phenomenon; an information creation unit configured to create information including at least an abnormality index related to the event based on data collected from the process; a node extraction unit configured to search for information on the nodes based on a name of the data collected from the process to extract a corresponding node; a data combining unit configured to associate the extracted node with the corresponding information; and a factor inference unit configured to infer and present the factor based on a structure of the knowledge model and the information associated with the node.