Two-Stage Abnormality Detection for Noisy Plant Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large-scale plants like nuclear and thermal power plants, detecting subtle abnormalities in process signals is challenging due to complex systems, abrupt changes, and the interference of minute electrical noise signals, which can lead to erroneous learning and incorrect abnormality detection by single machine learning models.
Innovation Solution
A two-stage machine learning model comprising a model MA and model MB is used to accurately predict plant data features, including stepwise changes and electrical noise signals, by reducing data dimensions and restoring data to its original form, enabling precise abnormality detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single machine learning model is used to detect abnormalities in plant data, then the detection process is simple, but the accuracy is reduced due to erroneous learning from complex systems, abrupt changes, and electrical noise signals
Solution Approach 1:
The patent divides the single machine learning model into two separate models: Model MA for detecting abrupt changes and Model MB for detecting electrical noise signals. This segmentation allows each model to specialize in specific types of anomalies, improving overall detection accuracy while managing complexity through modular design.
Solution Approach 2:
The patent introduces a difference calculation unit that computes the difference between input data and output data as an intermediary step. This intermediary mechanism enables the system to separately analyze abrupt changes and electrical noise signals, resolving the contradiction by adding a processing layer that enhances accuracy without excessive complexity.
2Productivity
If data dimension reduction is applied to simplify processing, then computational efficiency improves, but information loss may occur affecting detection accuracy
Solution Approach 1:
The patent applies dimension reduction to the difference data (after subtracting Model MA's output from input data) rather than the original input data. This preliminary action of calculating differences first preserves critical information about abrupt changes and electrical noise signals while enabling efficient processing of the reduced-dimensional data in Model MB.
Data Source
AI summary
An abnormality determination device includes one or more processors. The processors input first input data to a first model to obtain first output data. The first output data is formed by restoring data with the reduced dimension to data with the same dimension as that of the first input data. The processors input second input data, which is a difference between the first input data and the first output data, to a second model, and obtain second output data. The second output data is formed by restoring data with the reduced dimension to data with the same dimension as that of the second input data. The processors obtain restored data that is a sum of the first output data and the second output data. The processors compare the first input data with the restored data and determine an abnormality in the first input data based on the comparison result.