Causal Abnormality Estimation Using Normal OT Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality detection and diagnostic systems fail to accurately identify causal abnormality factors in OT systems, as they rely on correlation between data points rather than causal relationships, and require collection of abnormal data, which is time-consuming and inefficient.
Innovation Solution
An estimation apparatus comprising a correlation index estimation unit, an abnormality degree calculation unit, and a causal effect estimation unit that estimates correlation index information, abnormality degree, and causal effect to identify causal abnormality factors without needing abnormal data, using machine learning techniques like neural networks and probabilistic models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If correlation-based abnormality detection is used, then abnormality detection capability is provided, but causal abnormality factor cannot be detected
Solution Approach 1:
The patent replaces traditional correlation-based statistical methods with a neural network-based causal inference system. The neural network learns causal relationships from normal operation data alone, substituting the mechanical correlation calculation approach with an intelligent system that can identify causal abnormality factors without requiring abnormal data collection.
2Measurement precision
If abnormal data collection is performed to specify abnormality factors, then known abnormality factors can be specified, but time and effort are consumed
Solution Approach 1:
The patent performs preliminary learning during the normal operation phase, where the neural network is trained to recognize causal relationships between variables. This preliminary action enables the system to immediately identify causal abnormality factors when abnormalities occur, without requiring time-consuming abnormal data collection and processing.
Solution Approach 2:
The system uses normal operation data to train itself and build the causal relationship model. The neural network serves itself by learning from readily available normal data, eliminating the need for external abnormal data collection efforts and making the system self-sufficient in identifying causal factors.
3Ease of operation
If correlation-based abnormality determination is used, then abnormality detection is simplified, but causal abnormality factor that does not directly contribute cannot be specified
Solution Approach 1:
The patent substitutes the simple correlation calculation mechanism with a neural network-based causal inference system. While the neural network adds computational complexity, it automatically learns causal relationships from data, providing both simplicity in operation (automatic identification) and precision in causal factor detection without requiring manual analysis.
Data Source
AI summary
An estimation apparatus 1 includes: a correlation index estimation unit 2 configured to receive a variable output by a component 21 as an input, and estimate correlation index information indicating a range that a value of the variable and a correlation can take after a predetermined time, at a normal time; an abnormality degree calculation unit 3 configured to calculate an abnormality degree using the correlation index information; and a causal effect estimation unit 4 configured to estimate a causal effect expressing an index indicating that an abnormality propagates to a variable output by the component 21.