Anomaly Sign Detection Model with Process Value Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly sign detection techniques for plant apparatuses using AI face challenges in accurately distinguishing normal states from abnormal states due to insufficient data, leading to false positives and reduced reliability, especially in dynamic environments like power plants where seasonal variations and operational changes complicate data analysis.
Innovation Solution
An anomaly sign detection system that uses machine learning to calculate correction values for actual process values based on reference values, generating corrected process values to improve data correlation and reduce false positives by preprocessing data with an autoencoder network, thereby enhancing detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If unsupervised learning is applied to detect anomalies in time-series data from power plants, then the system can handle the infinite variety of abnormal states, but normal states are incorrectly identified as abnormal due to insufficient learning data
Solution Approach 1:
The patent applies data augmentation techniques to preliminarily expand the normal state data before training the unsupervised learning model. By generating synthetic normal data through transformations (rotation, translation, scaling, noise addition), the system prepares sufficient training data in advance, enabling the model to reliably distinguish normal variations from actual anomalies while maintaining the ability to handle infinite abnormal states
2Quantity of substance
If data augmentation is applied to time-series process data, then the learning data quantity increases, but the data cannot be augmented by conventional techniques like adding noise or changing frequency components
Solution Approach 1:
The patent changes the parameters of time-series process data through transformations such as time-shifting, scaling, rotating (in phase space), and adding realistic noise patterns that respect the physical constraints of the process data. These parameter changes generate diverse synthetic data while maintaining the inherent relationships and constraints of the original process variables
3Quantity of substance
If conventional data augmentation techniques are used on time-series data, then data quantity increases, but false positives increase and detection reliability decreases
Solution Approach 1:
The patent converts the challenge of data scarcity into a benefit by using unsupervised learning on augmented normal data to establish a robust baseline of normal behavior. The model learns the inherent patterns and variations of normal operation, enabling it to identify true anomalies more accurately while reducing false positives from insufficient training data
Data Source
AI summary
According to one embodiment, an anomaly sign detection system comprising one or more computers configured to: calculate a correction value for correcting at least one actual process value from the at least one actual process value and at least one reference process value; determine whether each of plurality of actual process values is correlated with the at least one reference process value or not, based on correction-necessity coefficient of determination; use the correction value for correcting at least one actual process value determined to be correlated with the at least one reference process value among the plurality of actual process values; generate learning input data including at least one corrected process value as the at least one actual process value corrected by the correction value; and perform machine learning by inputting the learning input data to anomaly sign detection model.