Plant Diagnosis Pattern Analysis for Early Abnormality Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current plant monitoring techniques, such as the Mahalanobis-Taguchi method, face limitations in accuracy and timeliness for detecting abnormalities, often failing to detect device issues until they have progressed significantly due to empirically set thresholds.
Innovation Solution
A diagnosis device that generates a diagnosis target pattern by plotting monitoring data against plant output data over time, allowing for early detection of abnormalities through pattern recognition, even if data does not exceed predetermined thresholds, and includes units for monitoring data acquisition, pattern generation, and pattern diagnosis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the Mahalanobis-Taguchi method is used to monitor plant operating state, then it is possible to comprehensively diagnose the plant with a single index (Mahalanobis distance), but the accuracy of detecting abnormality is limited and abnormality may be detected only after device damage progresses
Solution Approach 1:
The patent segments the single Mahalanobis distance index into multiple component indices by decomposing it into partial correlations with individual monitoring data items. This allows analysis of specific parameter relationships rather than treating all parameters as a single aggregated value, thereby improving abnormality detection accuracy while maintaining diagnostic simplicity.
Solution Approach 2:
The patent introduces a new dimension of analysis by examining the relationship between Mahalanobis distance and individual monitoring data items. Instead of relying solely on the magnitude of the Mahalanobis distance, the system analyzes how the distance correlates with specific parameters, adding a relational dimension that enhances detection precision.
2Ease of manufacture
If the control value (threshold) of the Mahalanobis distance is set empirically, then the diagnosis method is simple to implement, but abnormality may be detected after device damage progresses to some extent
Solution Approach 1:
The patent performs preliminary analysis by calculating partial correlations between Mahalanobis distance and individual monitoring data items before setting thresholds. This preliminary characterization of parameter relationships enables the establishment of more accurate, relationship-based thresholds that detect abnormalities earlier, reducing the time loss while maintaining implementation simplicity.
Solution Approach 2:
The patent changes the parameter basis for threshold setting from empirical fixed values to dynamically calculated partial correlation values. By using correlation-based parameters that reflect actual parameter relationships, the system achieves earlier abnormality detection without complicating the implementation process.
3Reliability
If multiple state quantities are monitored to determine plant operating state, then comprehensive monitoring is achieved, but it takes a lot of skill to monitor the trend of state quantities and determine abnormality
Solution Approach 1:
The patent introduces partial correlation coefficients as intermediary metrics that mediate between multiple monitoring data items and the overall Mahalanobis distance. These intermediaries translate complex multivariate relationships into interpretable correlation values, maintaining comprehensive monitoring while reducing the skill required to interpret trends and determine abnormalities.
Data Source
AI summary
A diagnosis device for diagnosing a plant based on an operating state of the plant includes a monitoring data acquisition unit configured to acquire a plurality of monitoring data which are measurement values of a parameter related to the operating state of the plant measured at different times, a diagnosis target pattern generation unit configured to generate a diagnosis target pattern that is a plot pattern where each of the plurality of monitoring data is plotted against plant output data of the plant, and a pattern diagnosis unit configured to diagnose the plant based on the plot pattern of the diagnosis target pattern.